Modelling Water Quality for

Water Distribution Systems

A thesis submitted for the degree of

Doctor of Philosophy

by
Stefan Heinrich Maier

Department of Manufacturing and Engineering S~'St(,IllS

BruneI University, Uxbridge

February 1999
Dedicated to my parents,
Hanna and Heinz Maier

1
Acknow ledgernents

Special thanks go to my supervisor, Dr. Roger Powell, for guiding and supporting
me throughout this project. He always made time for helpful discussions and his
enthusiasm was an inspiration to me. I am also very thankful to my industrial
supervisor, Tony Woodward, who was equally supportive. helpful and challenging.
Much support for my work came from the team of people at Thames Water
Utilities Ltd. that was working on the pipe rig during the time of my experiments.
I am indebted to all of them for their help, frequently starting at six o'clock in
the morning, but in particular the contributions of Anaick Delanoue, Dr. David
Holt and Dr. Sarah McMath were invaluable.
I am also much obliged to Thames Water Utilities Ltd. who provided, among
other things, the experimental set-up, the large-scale "TORUS" pipe rig, and
granted me the use of the obtained and previous data.
Last but not least, many discussions with my father, Prof. Heinz Maier, and
my friends and colleagues, especially Dr. Johannes Andersen and Silvio Hrabar,
did help me to clarify (or discard) my ideas. All my family and my friends both
in England and back home in Germany were an invaluable support throughout
the time of my research work at Brunel and I am very privileged to have them.

11
Abstract

Maintaining water quality in distribution systems has become a prominent issue

in the study of water networks. This thesis concentrates on disinfectant and
particle counts as two important indicators of water quality.
The models discussed in this work are based on data collected by the author.
The experimental set-up and procedure are described and observations of par-
ticle counts, particle counter size distributions, mono chloramine as disinfectant.
temperature, heterotrophic plate counts and epifluorescence microscopy counts
are reported.
A model of the response of particle counts to an increase in flow is developed.
This model is obtained from specification derived from the data and assumptions,
and is validated by its interpretability and its fit to data. A local shear-off density
and an initial biofilm shedding profile were introduced and thus a linear model
for this part of the water quality dynamics could be obtained. A procedure for
the identification of the parameters of the local shear-off function and for the
determination of the biofilm shedding profile is presented. This profile can be
used to provide information about the status of the distribution system in terms
of shear-off from the biofilm on the pipe walls.
Monochloramine decay dynamics are investigated. The chlorine meter data
is preprocessed with the help of titration data to correct meter drift. The data
is then used in calibrating two different possible chlorine models: a model with a
single decay coefficient and a model with bulk decay coefficient and wall demand
(as used in Epanet). Important difficulties in identifying these parameters that
come about because of the structure of the models are highlighted. Identified

111
decay coefficients are compared and tested for flow, inlet chlorine and temperature
dependence.
The merits and limits of the approach to modelling taken in this work and
a possible generalisation are discussed. The water industry perspective and an
outlook are provided.

IV
Contribution to Knowledge

The accompanying thesis is based on work carried out by the author at Brunel
University between October 1995 and December 1998.
All work and ideas in this thesis are original, unless otherwise acknowledged
in the text by references. The work has not been submitted for another degree in
this university, nor for the award of a degree or diploma at any other institution.
In the following, a list of the contributions of this thesis to knowledge in
the fields of water quality modelling for water distribution systems and system
identification is provided.

1. A qualitative characterisation of data collected by the author shows the

structure of the particle counts data and highlights relationships between

particle counts and bacteriological data.

2. The preprocessing of pipe rig data, in particular the chlorine meter data, is
addressed in some detail. It gives novel insight into the degree of uncertainty
associated with chlorine data and presents an original procedure for the
correction of these errors.

3. The approach to the development of the system structure from data that is
taken in this thesis is novel. The modelling work of this thesis is based on
data and is obtained with the help of specifications and assumptions and
validated by fitting and physical or biological interpretation.

4. The particle counts model, resulting from the approach taken in this thesis.
is another important new contribution. The concept of a local shear-off
 function allows the development of a linear model that describes shear-off
from biofilms that comes about due to a flow increase.

5. The particle counts model describes dynamic transient biofilm detachment

in terms of particle counts, a process that previously found little attention
and no mathematical description in the literature.

6. The determination of the biofilm shedding profile opens a new way for
testing distribution systems for critical areas in terms of particle shear-off.
Thus, the state of the system can be investigated.

7. Finally, through the comparison of the parameter estimation of two possible

monochloramine decay models, novel insights in the limitations and the
usefulness of these models are presented.

Parts of chapter 1, chapter 3, work related to section 8.2 and results from
previous experiments were published at the Doctoral Research Conference, De-
partment of Manufacturing and Engineering Systems, Brunel University [96].
Parts of chapter 4 were included in the Proceedings of the International Confer-
ence on Biofilms in Aquatic Systems [98]. Earlier results of the work presented
in section 5.3 and chapter 7 were published at the International Conference on
Computing and Control for the Water Industry [97].
Contents

I Background 1
1 Introduction 2
1.1 Overview..................... 2
1.2 Water Quality in Water Distribution Systems 3
1.3 Processes in Water Pipes . . o
1.3.1 Suspended Biomass . 5
1.3.2 Biofilm... 6
1.3.3 Nutrients . 6
1.3.4 Disinfectant 6
1.3.5 Particles.. 7
1.4 A Systems View .. 7
1.5 Thesis Organisation. 8

2 Literature Review 9
2.1 Introduction................ 9
2.2 Water Distribution Systems . . . . . . . 10
2.3 Water Quality for Distribution Systems. 12
2.4 Biofilms in Water Distribution Systems 14
2.4.1 Definition and Structure . 14
2.4.2 Biofilm Processes . . . . . 15
2.4.3 Detachment from Biofilms 17
2.4.4 Disinfection of Biofilms .. 19
2.4.5 Measuring Biofilms . . . . 23
2.4.6 Biofilm Kinetics and Models 24
2.5 Suspended Material. . . . . . . . . 27
2.5.1 Suspended Biomass . . . . . 27
2.5.2 Kinetics of Suspended Bacteria 28
2.5.3 Inanimate Particles and Particle Counters 29
2.5.4 Propagation of Conservative Substances . 30
2.6 Chlorine as Disinfectant . 32
2.6.1 Chlorine Chemistry and Simple Chlorine Kinetics 32
2.6.2 Modelling of Reactive Substances 34
2.6.2.1 Models . 3-1
2.6.2.2 Case Studies . 37
2.6.2.3 Variability of Decay Coefficients. 40
2.7 Temperature. . . -12

\"11
 2.8 Multi-Quality Xlodels . -12
2.9 Algorithms and Simulation. . . . . . . . . . . -1-1
2.10 System Identification and Physical Xlodelling . -1.)
2.11 The Approach of This Thesis -17
2.12 Conclusions . . . . . . . . . . . . . . . . . . . -19

II Data 50
3 Experimental Work 51
3.1 Experimental Set-Up . . . . 51
3.1.1 The Pipe Rig . . . . . ... 51
3.1.2 The Pipe Rig as Input-Output System 53
3.2 Measurement Method. . ... 5-1
3.2.1 Heterotrophic Plate Counts 00
3.2.2 Bottled Samples for Heterotrophic Plate Counts 00
3.2.3 Epifuorescence Microscopy 56
3.2.4 Particle Counters 56
3.2.5 Chlorine Meters . . . . 01
3.2.6 Chlorine Titrations 58
3.2.7 Bottled Samples for Chlorine Titrations 58
3.2.8 Volumetric Flow Meter .. . . . 59
3.2.9 Temperature, pH and RedOx Potential 59
3.3 Experiments........ . 60
3.3.1 Pipe Rig Experiments During 1996 60
3.3.2 Bottle Experiments During 1996 . 62
3.4 Problems Encountered 63
3.5 Summary . . . . . . .. .. 64

4 Qualitative Characterisation of the Data 66

4.1 Total Particle Counts. 67
4.1.1 Observations 68
4.1.2 Discussion and Hypotheses. 70
4.1.2.1 Peaks .. 70
4.1.2.2 Occasional Particle Step 72
4.1.2.3 Slow Signal .. . . . . 72
4.1.2.4 Inlet Particle Counter Signal 73
4.1.2.5 Particle Counts after Decrease of Flow 74
4.2 Particle Counter Size Distributions 74
4.2.1 Observations . . 75
4.2.2 Discussion and Hypotheses. 76
-1.3 Disinfection . . I ,

4.3.1 Observations in the Pipe Rig. 11

-1.3.2 Observations in Bottled Samples 79

-1.-! Temperature. .. 81
-1.5 Bacteriological Data .... 81

YIn
 4.5.1 Heterotrophic Plate Counts . . . . . . . . . . . . . . 81
4.5.1.1 Observations in the Pipe Rig During SFTs . 82
4.5.1.2 Long-Term Observations in the Pipe Rig. 8-1
4.5.1.3 Observations in Bottled Samples 85
4.5.2 Epifluorescence Microscopy Counts . 88
4.5.3 Discussion.................. 90
4.5.3.1 Heterotrophic Plate Counts . . . 90
4.5.3.2 Epifluorescence Microscopy Counts 91
4.5.3.3 Summary 92
4.6 Long-Term Effects 93
4.7 Conclusions . . . . . . . . 93

III Model 95
5 Preprocessing of Data 96
5.1 Introduction............. 96
5.2 Switching the Particle Counters . . 97
5.2.1 Ratios of Particle Counters. 97
5.2.2 Original Order of Particle Counters 100
5.3 Chlorine Meter Data . . . . . . . . . . . . 101
5.3.1 Correction of Chlorine Meter Errors. 101
5.3.2 Alternative Correction Procedures. 106
5.3.3 Summary .. . . . . . . . . . . 109
5.4 Travel Times (Delay Times) . . . . . . . . 109
5.4.1 Calculation of the Travel Time .. 110
5.4.2 Possible Errors in Calculating the Travel Time. III
5.4.3 Error Correction 113
5.4.3.1 Length Correction . . . . . . . . . . . 114
5.4.3.2 Adjustment of the Meter Extraction Flows. 115
5.4.3.3 Final Correction 116
5.5 Implementation 120
5.6 Conclusions . . . . . . . . . . . . 121

6 Modelling of Particle Counter Data 123

6.1 Introduction.................. 123
6.1.1 Purpose . . . . . . . . . . . . . . . . 123
6.1.2 Particle Counters in the Test System 123
6.1.3 Results......... 124
6.2 Model Specifications . . . . . 124
6.3 Numerical Statistical Models. 127
6.3.1 Total Particle Counts. 127
6.3.1.1 All attached biomass in a cell is sheared-off 129
6.3.1.2 Clumps of random size are sheared-off . . . 130
6.3.1.3 A constant fraction of the remaining biofilm is
sheared-off . . . . . . . . . . . . . . 132

IX
 6.3.1.4 A decreasing fraction of the remaining biofilm is
sheared-off 133
6.3.2 Particle Counter Size Distributions . . . . 134
6.3.3 Conclusions................. 137
6.4 Parametric Model of the Total Particle Response 137
6.4.1 Assumptions............ 138
6.4.1.1 Peaks . . . . . . . . . . . . . . 138
6.4.1.2 Occasional Particle Step . . . . 139
6.4.1.3 Basic Model of the Slow Signal 139
6.4.1.4 Extended Model of the Slow Signal 142
6.4.1.5 Inlet Particle Counter Data . . . . 144
6.4.1.6 Model for Particle Counts After Flow Decrease 144
6.4.1.7 Discussion of Assumptions . . . . . . . 145
6.4.2 Development of the Positive Input Slope Model 146
6.4.3 Development of a Negative Input Slope Model 149
6.4.4 Examples . . . . . . . . . . . . 151
6.5 Parameter Identification . . . . . . . . 155
6.5.1 Data Preparation (Peak Erase) 156
6.5.1.1 Peak Erase by Hand . 156
6.5.1.2 Automatic Peak Detect and Erase 157
6.5.1.3 Inclusion of Peak Models. 158
6.5.2 Time Constant Ta = 1a . . . . . . . 159
6.5.3 Parameters k 1 (f) and k 2 (f ) . . . . 160
6.5.4 Issues in Parameter Determination 164
6.5.4.1 Impact of Peak Erase .. 164
6.5.4.2 Normalisation of Data for Simulation . 165
6.6 The Initial Biofilm Shedding Profile. 166
6.7 Results.............. 168
6.7.1 Parameters 169
6.7.2 Biofilm Shedding Profile 172
6.8 Discussion.......... 177
6.9 Validation and Limitations. 182
6.10 Implementation 183
6.11 Conclusions . . . . . . . . . 184

7 Comparison of Models of Monochloramine Decay 185

7.1 Introduction......... 185
7.2 Review of Chlorine Kinetics . . . . . 186
7.3 Total Chlorine Kinetics. . . . . . . . 188
7.4 Determination of Decay Coefficients. 190
7.4.1 Parameter Estimation for the Single Decay Coefficient :\Iode1190
7.4.2 Parameter Estimation for the Model With Two Decay Co-
efficients . . . . . . . 193
7.4.3 Issues in Estimation 19-1
7.5 Bottled Samples. . . . . . 195
7.6 Analysis of Pipe Rig Data . 197

x
 7.6.1 Single Decay Coefficient . . . 197
7.6.2 Combined Decay Coefficients 20-1
i . I Discussion... 206
7.8 Implementation ')0-
_ i

7.9 Conclusions . . 20S

IV Conclusion 210
8 Discussion and Conclusions 211
8.1 Xlerits of this Approach to :\Iodelling 211
8.2 Generalisation........ 212
8.3 Water Industry Perspective 213
8.4 Future Perspective 21-1
8.·S Conclusions . . . . 216

Bibliography 218

Appendix 231
A Model-Relevant Data 232
A.1 Total Particle Counts . . . . . . . . . 232
A.2 Size Distributions of Particle Counts 2-10

B Calculations for the Determination of k 1 (f), k 2 (f ) 258

B.1 Introduction. . . . . . . . . . . . . . . . . 258
B.2 Calculation of the Laplace Back-Transform 259
B.3 Calculation of the Full Integral 261
BA A Test of Validity . . . . . . . . 263

C Implementation in Matlab 265

C.1 Overview. . . . . . . . . . . . . . . . . . 26.S
C.2 Using the \Yater Quality Toolbox . . . . 266
C.2.1 Reading Data and Time Formats 266
C.2.2 Processing of Raw Data, Visualisation and Graphical Input 266
C.2.3 Data Analysis . . . 267
C.2.4 Starting a Session. 268
C.3 Summary . . . . . . . . . 268

D Reynolds Numbers 269

Xl
List of Figures

1.1 Criteria for water quality in water distribution systems -1

1.2 Processes in water pipes 0

3.1 Schematic drawing of the "TORUS" pipe test rig at Kempton Park. 52
3.2 The pipe rig as an input-output system. . . 54
3.3 Experimental Schedule showing flow and total chlorine at the pipe
rig inlet, three baseline regimes of Experiment 1 are indicated 60

4.1 First example of total particle counts: SFT 5 . 68

4.2 Second example of total particle counts: SFT 8 69
4.3 Particle Counter Size Distribution of sample point 1. ....-
10
4.4 Total chlorine vs. travel time and time at the inlet, 15.10.1996
(step flow trial 7). . . . 77
4.5 Total chlorine vs. travel time and time at the inlet, 26.11-29.11.1996
(step chlorine trial 1). . . . 78
4.6 Total chlorine in free chlorine bottles . . . . 80
4.7 Total chlorine in mono chloramine bottles . . . 80
4.8 Temperature on the left y-axis over the full experimental time pe-
riod with inlet flow on the right y-axis for comparison with the
schedule of the experiments .. 82
4.9 Heterotrophic plate counts of inlet tank (INLET) and three sample
points along the rig (SP 1, SP 2, SP 3) and inlet flow.. 83
4.10 Heterotrophic plate counts of sample point 2, inlet tank and flow. 84
4.11 HPC of bottles without chlorine 85
4.12 HPC of free chlorine bottles 86
4.13 HPC of monochloramine bottles. 87
4.14 SFT 7: Epifluorescence microscopy counts of 'clumps' of 10 or
more bacteria . . . .. 89
4.15 SFT 8: Epifluorescence microscopy counts of 'clumps' of 10 or
more bacteria . . 90
4.16 Simplified pattern in heterotrophic plate counts; inlet flow values
are given on the right y-axis. . . 91
4.17 Comparison of patterns in epifluorescence counts of 'clumps' of 10
or more bacteria and the particle counter size distribution; inlet
flow values are given on the right y-axis. 92

5.1 Physical quantities and measured data 96

xu
5.2 Discrete Fourier transform of the multiplicative error function c\7(t).102
5.3 Corrected and uncorrected chlorine data of sample point 1. 104
5.4 Corrected and uncorrected chlorine data of sample point 3. . 105
5.5 Alignment of particle counts with original parameters.. . . . 113
5.6 Alignment of particle counts of SFT 1 with final correction. . 115
5.7 Alignment of particle counts of SFT 4 with final correction. . 116
5.8 Alignment of particle counts of 14.12.96 (baseline flow) with final
correction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.9 Alignment of particle counts of SFT 6 with final correction. . 118
5.10 Alignment of particle counts of SFT 8 with final correction. . 119
5.11 Alignment of total chlorine with final correction. . . . . . . . 120

6.1 Qualitative Simulink model of particle counter response to a step

increase in flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
6.2 Schematic showing the discretisation of the pipe rig into n cells 128
6.3 Particle counts response to an increase in flow if all attached biomass
in a cell is sheared-off. Probability of shedding is 0.01. 129
6.4 Particle counts response to an increase in flow if all attached biomass
in a cell is sheared-off. Probability of shedding is 0.5. . . . . . . . 130
6.5 Particle counts response to an increase in flow if clumps of random
sizes are sheared-off. Probability of shedding is 0.5. . . . . . . . . 131
6.6 Particle counts response to an increase in flow if 1% of the remain-
ing biofilm is sheared-off. Probability of shedding is 0.5. . . . . . . 132
6.7 Illustration for the subsequent remaining amounts of biofilm with
a decreasing fraction (double-exponential) shear-off. Always 50 %
of the remaining fraction is sheared-off. . . . . . . . . . . . . . . . 133
6.8 Particle counts response to an increase in flow if a decreasing frac-
tion of the remaining biofilm is sheared-off. Probability of shedding
is 0.5. 134
6.9 Particle counter size distributions. . . . . . . . . . . . . . . . . . . 135
6.10 Particle counter size distributions. . . . . . . . . . . . . . . . . . . 136
6.11 Linear system with input flow f(t) and output total particle counts
p(t) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
6.12 Linear system with apparent initial shedding profile b*('I9) as input
and output total particle counts p(t), positive input slope model.. 149
6.13 Linear shear-off function and corresponding p~rticle counter re-
sponse for a flat Biofilm Shedding Profile. 151
6.14 Exponential shear-off function and corresponding particle counter
response for a flat Biofilm Shedding Profile. 152
6.15 Sum of two exponential shear-off functions (second order system)
and corresponding particle counter response for a flat Biofilm Shed-
ding Profile. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
6.16 Decreasing Biofilm Shedding Profile and corresponding particle
counter response for an exponential shear-off function.. . . . . . . 153
6.17 Increasing Biofilm Shedding Profile and corresponding particle counter
response for an exponential shear-off function. . . . . . . . . . . . 153

XU1
6.18 Quadratic Biofilm Shedding Profile and corresponding particle counter
response for an exponential shear-off function. . . . . . . . . . . . 15-1
6.19 Effect of reduction of negative edge threshold in peak detect for
apparent initial biofilm shedding profile of SFT 6 16-1
6.20 Unfiltered and filtered apparent initial biofilm shedding profile of
SFT 2 . . . . . . . . . . . . . . . . . . . . . . . . 167
6.21 Apparent initial biofilm shedding profile of SFT 1 173
6.22 Apparent initial biofilm shedding profile of SFT 2 17-1
6.23 Apparent initial biofilm shedding profile of SFT 3 175
6.24 Apparent initial biofilm shedding profile of SFT 4 176
6.25 Apparent initial biofilm shedding profile of SFT 5 177
6.26 Apparent initial biofilm shedding profile of SFT 6 178
6.27 Apparent initial biofilm shedding profile of SFT 7 179
6.28 Apparent initial biofilm shedding profile of SFT 8 180
6.29 Apparent initial biofilm shedding profile of SFT 9 181

7.1 Simple first order decay of total chlorine. . . . . . 189

7.2 Monochloramine in glass bottles and an exponential function is
fitted to all data. . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
7.3 Free chlorine in glass bottles and an exponential function is fitted
to all data.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
7.4 Decay Coefficients in period 12, as example of estimated single
monochloramine decay coefficients K of data with low standard
deviation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
7.5 Decay Coefficients in period 30, as example of estimated single
monochloramine decay coefficients K of data with high standard
deviation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
7.6 The combined decay coefficient K as used in Epanet versus the
wall demand k w ' 204

A.l SFT 1: Total particle counts during high flow 233

A.2 SFT 2: Total particle counts during high flow 234
A.3 SFT 3: Total particle counts during high flow 235
A.4 SFT 4: Total particle counts during high flow 236
A.5 SFT 5: Total particle counts during high flow 237
A.6 SFT 6: Total particle counts during high flow 238
A.7 SFT 7: Total particle counts during high flow 239
A.8 SFT 8: Total particle counts during high flow 240
A.9 SFT 9: Total particle counts during high flow 241
A.I0 SFT 10: Total particle counts during high flow. 242
A.ll SFT 11: Total particle counts during high flow . 243
A.12 SFT 12: Total particle counts during high flow. 244
A.13 SFT 1: Particle counter size distributions. 246
A.14 SFT 2: Particle counter size distributions. 2-17
A.15 SFT 3: Particle counter size distributions. 248
A.16 SFT 4: Particle counter size distributions. 2-19

Xl\"
A.17 8FT .5: Particle counter size distributions. 2.50
A.18 8FT 6: Particle counter size distributions. :251
A.19 8FT 7: Particle counter size distributions. :2.j :2
A.20 8FT 8: Particle counter size distributions. .)-
_.J.'3

A.21 8FT 9: Particle counter size distributions. 2.j-!

A.22 8FT 10: Particle counter size distributions.. :25:)
A.23 8FT 11: Particle counter size distributions.. :256
A.24 8FT 12: Particle counter size distributions.. 2.5,

D.1 Reynolds numbers versus volumetric flow ..

I
II
,
j

I
I,

I
I,

Xy

~
List of Tables

2.1 ext values 0 • • • • • • • • • • • •

21
2.2 Monochloramine Decay Constants and Flow Rate . . . . . . . . . .J1
2.3 Comparison between physical modelling, system identification and
"hy bn'd" approaches. . . . . . . . . . . . .J6
3.1 Bottles taken for bacteriological analysis 63
3.2 Bottles taken for chlorine titrations . . . 6.J
5.1 The ratios of the particle counters before and after switching 98
5.2 Percentage PFTO.3 of the maximal value of the Fourier transform at
f = 0.3d~Y· . . . 0 • • • • • • • • • • • • • • • • • • • • 107
5.3 Additive error residuals in rng/L . 108
5.4 Errors on abscissa after alignment of particle counts. 11.J
6.1 Thresholds for peak erase and filter window sizes. . . 168
6.2 Investigated time periods of SFTs. . 169
6.3 Data of model parameters a and k 1 (f) for step flow trials 1 to 9. 169
6.4 Statistical analysis of data concerning time constants Tao ... 0 170
6.5 Approximation constants e' and e" during step flow trials 1 to 9. 171
6.6 Final values of total particle counts at high flow during step flow
_')
trials 1 to 9. . . . . . . . . . . . . . . . . . . . . 0 • • • • • • • • • 1 1-

7.1 Decay coefficients and other fitted parameters of mono chloramine

and free chlorine bottled samples. . . . . . . . . . . . . . . . . . . 196
7.2 Monochloramine decay coefficient data. . . . . . . . . . . . . . . . 198
7.3 Pairs of time periods with average decay coefficients that are not
significantly different. . . . . . . . . . . . . . . . . . . . 202
7.4 Results of the estimation of combined decay coefficients . 206

XYl
Principal Symbols and
Abbreviations

1(t) uni t step function

A cross-sectional flow area of pipe
AOC Assimilable Organic Carbon
ATP adenosine triphosphate
a decay coefficient in particle counts model
BSP Initial Biofilm Shedding Profile
B(s) Laplace transform of b( x )
B*(s) Laplace transform of b* ('l9)
BDOC Biodegradable Dissolved Organic Carbon
b(x) initial biofilm shedding profile, part of particle counts model
b" ('l9) initial biofilm shedding profile as function of modified variable
b1 ( x), b2 ( X ) initial biofilm shedding profiles for extended specifications
i, (i), b, (i) attached and suspended biomass in cell i, respectively
b5 ( · ) fifth order Butterworth filter with zero-phase distortion
c(x, t) total chlorine concentration
non-reactive fraction of total chlorine (coo = 0 for free
chlorine)
Cm ( x, t), Cm ( t ) chlorine meter reading
Cb ( x, t), Cb ( t ) result of bench titrations of chlorine
Cc ( x, t), Cc ( t ) corrected chlorine reading (after preprocessing)
C , C' , c" validation constants in particle counts model
Cl (T), C2 (T) parameters of the exponential tail of the particle counts model
D molecular diffusivity of substance in fluid, D = ;c
DOC Dissolved Organic Carbon
d pipe diameter
6 small positive pertubation
6(t) Dirac's delta impulse
6' (t) derivative of 6(t)
e.g. for example
EPS Extracellular Polymeric Substances
fA! (ti) multiplicative error function, for chlorine preprocessing
fA(t i ) additive error function, for chlorine preprocessing
(f[(t) fA! (t i ) reverse-calibrated

(~~ (t) f.\f(t i ) reverse-calibrated and interpolated

f~\R(t) (A (ti) reverse-calibrated and interpolated

..
XVll
 E~f'o (t) E~(t) with the zero-frequency value set to zero
E~F(t) filtered error function E~7 (t)
EW (t) EM(t i ) interpolated within recalibration windows
E~W (t) EA (ti) interpolated within recalibration windows
E~,O(t) E~W (t) with the zero-frequency val ue set to zero
EWF(t) filtered error function E~r (t)
f(t) flow rate
H total static head
HPC Heterotrophic Plate Counts
i.e. that is
K chlorine decay coefficient
kb bulk demand, part of the chlorine decay coefficient
kw wall demand, part of the chlorine decay coefficient
kf flow dependent mass transfer coefficient
k l , k2 parameters in particle counts model
k= sx. k2
ratio of parameters in particle counts model
kd biofilm detachment coeficient in 1/second
L pipe length
Lf biofilm thickness for biofilm detachment kinetics
fi(t) polynomial for linear interpolation in the interval
[ti, t i+l ]
MDPE medium density polyethylene
M.V. Maximum Value (in particle counts signal)
1/ kinematic viscosity of fluid
p(t) particle counter signal at a fixed point x = xo
Pd(X, t) local particle shear-off density
Q pump discharge
RH hydraulic radius
SFT Step Flow Trial
SP Sample Point
Sc Schmidt number
5h Sherwood number or :-;usselt number for mass
transfer
5(s) Laplace transform of s(t)
s(t) local shear-off function, part of particle counts model
Sinv (t) inverse Laplace transform of 5- 1 (s) = sts)
SI(t), S2(t) local shear-off functions for extended specifications
a standard deviation
T temperature
T travel time to a fixed point Xo, T = T(XO)
t; time constant in particle counts model, Ta = ~
TK time constant in chlorine decay models. T K = ~
TOC Total Organic Carbon
t, to. t l · t• time
t~ jth recalibration time
J

xnn
T(X), T(X, t). T(i,j) travel time
{) modified variable of biofilm shedding profile.
{)=T-T
() threshold in peak erase algorithm
!1t. !1T hydraulic time step: sampling time
Re Reynolds number
r(t i ) reverse-calibration factor at time t,
rd biofilm detachment rate in kg/(m 2 s)
p biofilm density for biofilm detachment kinetics
F volume
VOl, 112, ~!23 pipe rig volume between two sample points
V, v(t) velocity of water
W set of recalibration windows
Wj jth element (recalibration window) of W
w number of windows in W
distance along the pipe rig

XIX
Glossary

Biofilm: A collection of microorganisms and their extracellular prod-

ucts bound to a solid surface. Biofilms are generally organic
although there might be entrained inorganic debris.

Epifluorescence Microscopy: Measurement method for counting individual or clumps

of live, injured and dead bacteria using fluorescent staines.

Heterotrophic Plate Counts: Measurement method for counting bacteria colonies

(injured or dead bacteria are not counted, 'clumps' cannot
be distinguished from single bacteria).

Initial Biofilm Shedding Profile: Snapshot of the initial amount of particles com-
ing off the pipe walls as a function of location (i.e. distance
or length); only particles that exhibit some attachment to
the walls are included thus implying that they are part of
the biofilm.

Local Particle Shear-off Density Pd(X, t): The density per unit length of particles
that are sheared-off the pipe walls given as a function of
time and distance.

Local Shear-off Function s(t): Particles that are sheared-off the pipe walls as a
function of time only.

Shear-off: Continuous removal of biofilm due to shear forces (also

known as erosion).

Sloughing: Detachment of large amounts of biofilm, aided by shear

forces.

Step Flow Trial: Step increase of flow rate and subsequent step decrease
(after between 14 hours and several days); frequently used
as input signal for the pipe rig.

Titration: Chemical bench analysis, used for the determination of

chlorine content (in terms of free chlorine. monochloramine
and dichloramine) in water samples.

xx
 Part I

Background

1
Chapter 1

Introduction

1.1 Overview

This work is concerned with water as distributed in potable water supply systems.
Elements of these large-scale systems are pipes, valves. pumps, reservoirs etc. The
hydraulics of these systems are governed by non-linear equations involving the
flow and head loss in each element.
In recent years there has been an increasing awareness of the importance of
maintaining water quality in water distribution systems [35, 33]. This has given
rise to the need to utilise computer models to simulate changes in water quality
in order to predict and control water quality throughout distribution networks
[59].
It is the aim of this work to quantify some important aspects of water quality
in water distribution systems. In the long run, this should lead to the possible
use of water quality in a similar way as the other states of the system (like flow
or head), e.g. for estimation, control or as part of a cost function for the purpose
of optimisation.
To achieve this, state variables that describe water quality (i.e. describe it
to an extend considered sufficient for the purposes of this work) are defined and
the underlying equations or mathematical system structure which describes there
relationships is found. In this work, the methodologies of system identification

2
are used for that end. The water quality variables investigated in this thesis are
foremost particle counts and mono chloramine, but bacteriological counts (like
heterotrophic plate counts or epifluorescence microscopy) are discussed as well.
A model of the response of the particle counts to an increase of flow is de-
veloped and this model is used to determine parameters and a biofilm shedding
profile that give information about the condition of the pipe system in terms of
biofilm (or other) shear-off. In addition, two published models of chlorine de-
cay are adopted for mono chloramine, the decay coefficients are estimated and
compared, and they are investigated for dependence with other water quality
variables.

1.2 Water Quality in Water Distribution Sys-

tems

The definition of water quality as such is by no means straightforward. Gener-

ally, it is perceived to be a collection of upper or lower limits on selected possible
contaminants of water [143, 31]. Thus, a wide range of substances and physical
parameters could be considered, however, the models presented in the literature
frequently concentrate on water age, the propagation of conservative substances
(e.g. particulate matter) or reactive chemical substances, such as disinfectants,
as in [59, 123, 154] among others.
Other relevant substances include heavy metals, various toxins, dissolved oxy-
gen, chlorides, total chlorine, and bacterial counts (e.g. heterotrophic plate count,
HPC), among others. Turbidity or pH are generally also considered. Since this
work is concerned with treated water, bacteria are considered to be crucial in
determining the quality of water. These possible water quality criteria are illus-
trated in fig. 1.1.
Obviously, the treatment process is an important determinant of further water
quality, however. even under the most advanced treatment small numbers of
bacteria enter the distribution system. These bacteria have a potential of growth

3
 ~~\ ~r~ /-
Chlorides • ~ty Cri~ 4 pH

To~J 1 . .
Bacterial Counts
(e.g. HPC)

Figure 1.1: Criteria for water quality in water distribution systems

and multiplication, especially since the water is otherwise quite "clean:'. Thus
important limiting factors for bacterial growth and multiplication, and therefore
for water quality, are the amount of nutrients available and the possible presence
of a disinfectant.
This thesis is concerned with the quality of the water between the outlet
of the treatment plant and the consumer's tap. In this context, the disinfec-
tant concentration is of paramount importance (rather than the nutrient content.
which normally cannot be influenced after treatment). The most frequently used
disinfectants in the U.K. are free chlorine (Ch) and monochloramine (NH2 CI).
Since the disinfectant reacts with a number of substances in water, its residual is
decaying.
Another important factor is the bacteria concentration in the system. Because
of the difficulties in taking pipe cutouts, water utilities generally use plate counts
or epifluorescence microscopy counts as indication of biofilm growth and detach-
ment. This seems to be justified if the system is exposed to high flow rates since
it has been reported in the literature that increased biofilm detachment may be
due to increased shear stress (among other things) [20, 122, 26, 144]. In this work
different methods of measuring biofilm detachment under high shear stress con-
ditions are investigated (heterotrophic plate counts, epifluorescence counts and
particle counts). Shear stress is introduced by an experiment consisting of a series
of flow increases.
 Biofilm

\ death

Nutrients

Figure 1.2: Processes in water pipes

1.3 Processes in Water Pipes

There are numerous chemical and biological processes taking place in any section
of pipe in a water distribution system, which have an effect on water quality
parameters. In Fig. 1.2, a schematic drawing of these processes is given.

1.3.1 Suspended Biomass

Bacteria or, more generally, biomass entering a distribution system may undergo
any of the following processes:

• Growth and multiplication of living, non-injured bacteria. For this purpose.

they utilise nutrients and respire.

• Respiration. On this level of organisms this can by viewed as a direct loss

of mass (since O 2 is inhaled and CO 2 is exhaled).

• Death, which makes the dead bacteria part of the pool of nutrients.

• Adhesion. Some of the bacteria entering a pipe section will settle and adhere
to the pipe walls and thus form a layer of biofilm.

• The number of suspended bacteria will be increased by sloughing parts of

biofilm due to shear stress.

1.3.2 Biofilm

The bacteria settling on pipe walls form a layer of biofilm. A possible definition of
"biofilm" is given in [20] (after [100]): "a biofilm is a collection of micro-organisms
and their extracellular products bound to a solid (living or inanimate) surface
(termed a substratum)" (p.57). Biofilm bacteria are involved in the same pro-
cesses as suspended bacteria, i.e. growth and multiplication, respiration, death,
adhesion (increase of amount of biofilm bacteria) and sloughing (or shear-off).
In particular the decrease of biofilm bacteria due to sloughing (i.e. due to shear
stress) is flow dependent.

1.3.3 Nutrients

Nutrients are depleted by all living biomass present. Dead biomass, however,
adds nutrients to those already present.

1.3.4 Disinfectant

Disinfectants are generally highly reactive substances. Thus, their pure form
declines in number over time to form new chemical species. This chemical reac-
tion can kill bacteria, however the disinfectant will also react with many other
substances.
The more highly reactive a disinfectant, the faster it decays, and generally the
better it disinfects (while there is something left). Free Chlorine (CI2 ) is highly
reactive and therefore decays quickly, whilst monochloramine (NH 2CI) decays
more slowly.
In the system under consideration, monochloramine is chosen as disinfectant
since a residual of this substance is present longer than a free chlorine residual.
However, this implies a slower rate of reaction with all water constituents includ-
ing bacteria. The combination of all chlorine species (free chlorine. monochlo-
ramine, dichloramine, etc.) is generally termed total chlorine. There is normally
no free chlorine in the considered test system.
It is suggested by some authors that monochloramine reacts better with

6
biofilm than free chlorine since it penetrates deeper layers of the film due to
its slower reaction rate [85].

1.3.5 Particles

Suspended particles in the bulk liquid of more than 2 uitv in diameter can be
counted through on-line particle counters or may be measured through micro-
scopic techniques. They may be inanimate or animate, conservative or reactive.
Animate (i.e. reactive) particles are chiefly the suspended bacteria mentioned
above, but inanimate particles are mainly debris like pieces of pipe material or
entrained dirt. In the case discussed in this work it appears that animate particles
form the predominate portion of counted particles.

1.4 A Systems View

The approach taken in this paper is inspired by systems science, viewing the test
rig as a system of independent variables (flow rate, initial disinfectant concentra-
tion), dependent variables (disinfectant residual, biofilm, particles, bacteriological
counts), parameters (e.g. temperature, pipe material, source water quality, initial
biofilm present) and their relationships.
For the purpose of this project, a system is considered to be a theoretical
object rather than e.g. the existing pipe rig. A system can be viewed as a set of
variables and parameters together with their relationships. A suitable definition
is e.g. given by Forrester [53]: "As used here a 'system' means a grouping of parts
that operate together for a common purpose" (p.l-l). In the case considered in
this work, this common purpose exists in a quite abstract way as being water
quality or effecting water quality.
The associated model is the transformation of the system into the abstract
domain of mathematics.
To make this transformation easier and to enhance the understanding of the
svstrm. only a simplified version of the svstern is transformed. This simplification

-
I
must preserve, however, all relevant variables, parameters. and relationships of
the system for the specified purpose, i.e. explanation or prediction.

1.5 Thesis Organisation

Chapter 1 gives an introduction to the problem of water quality modelling, chap-

ter 2 reviews the available literature in the field. The experimental work, as
described in chapter 3, was undertaken at Thames \Vater Utilities Ltd. During
the period of June 1996 to December 1996 the author was almost full-time at
the experimental facility to take care of the required work. Chapter --! describes
the results obtained from these experiments. This characterisation of the data
is used as a basis for the analysis of the experimental results in part III of this
work. In this part, the preprocessing of chlorine meter and particle counts data is
covered in chapter 5, the particle counts model in chapter 6 and the comparison
of two monochloramine models in chapter 7. The thesis finishes with a discussion
and conclusions in chapter 8. Appendices provide a comprehensive collection
of data plots (appendix A), mathematical derivations for some of the particle
counts model (appendix B) and an overview of the implementation of this work
in Matlab (appendix C).

8
Chapter 2

Literature Review

2.1 Introduction

The literature in the fields of water distribution systems, water quality and math-
ematical modelling is very extensive and only a representative sample can be
presented here. Naturally, the focus lies on those publications which are most
relevant to this work, but also some work of more marginal significance for this
thesis has been include, especially in the sections concerning water quality.
Grayman and Clark [59] give an overview of water quality modelling for dis-
tribution systems. They distinguish three different types of models of water
distribution systems:

• hydraulic models "which simulate the flow quantity, flow direction, and
pressure in the system" (p.81),

• steady state water quality models "which determine the movement of con-
taminants, their flow paths and travel times through the network under
steady state operational and demand conditions" (p.81), and

• dynamic water quality models "which simulate the movement and transfor-
mation of substances in the water under time varying conditions" (p.81), i.e.
demand, pressure or flow vary e.g. diurnally (or in other words: transient
phenomena are also modelled).

9
 Besides a brief explanation of these types of models, an example of a model of
the propagation of a conservative (i.e. non-reactive) substance is given. In this
case example the usefulness of a model for prediction purposes and analysis of a
distribution system is demonstrated. The source of pollution (two leaks) could
be found and validated in this investigation.
The publication by Shamir and Howard [131] gives also an overview of wa-
ter quality modelling, some general thoughts and a "system's approach." The
overview covers water quality as used in simulation models, methods of analysis.
global network analysis and an approach to the optimisation of a hydraulic or a
combined hydraulic/water quality model.
In addition to the types of models discussed in the above publications. other
models which focus more on biological processes were developed recently [62, -19],
see sections 2.4.6 and 2.5.2.

2.2 Water Distribution Systems

All the work of this thesis is aimed at potable water distribution systems. These
systems are governed by hydraulic equations (flow and head) and are normally
large-scale and non-linear.
A comprehensive introduction into water distribution systems is given in the
book "Operational Control of Water Systems: Structures, Algorithms and Ap-
plications" by Brdys and Ulanicki [17]. It uses a systems view in investigating
water systems and covers in depth, among other things, modelling, simulation,
scheduling and operational control of water distribution systems, together with
several case studies.
From the rich literature on hydraulic modelling only few titles are selected for
this review.
Shamir and Howard's [130] publication contains an introduction to the pro-
cesses involved in obtaining hydraulic models. It describes the Hardy-Cross
method, which is, however, in the meantime hardly used due to poor numeri-
cal efficiency on computers.

10
 The article by Alexander et at. [2] contains some basic considerations concern-
ing the hydraulic modelling of water distribution systems and presents some of
the features of their simulation program.
Several possible uses of hydraulic modelling are given: (1) analyse problems,
(2) develop emergency operating procedures, (3) find water system development
plans, (4) determine priorities of improvement in a capital improvement pro-
gramme, (5) find out the capabilities of the existing system for different de-
mands, (6) compare different scenarios, (7) evaluate effects of major customers.
(8) consider benefits or operational problems of several water systems, (9) eval-
uate effects of proposed major changes, and (10) develop a real-time operational
control system.
For the development of such a model it is necessary to have a schematic dia-
gram of the system which defines sources of supply, pumping, storage, a distribu-
tion pipeline grid, etc. This could be obtained from physical features (topography,
supply, reservoirs, pipelines) and operational records (source pumpage, reservoir
levels, booster pumps, demand).
The underlying theory are basically "Kirchhoff's Laws" for water systems:

(1) Continuity: the rate of flow into each junction (node) must be equal to the
rate of flow out of that node (including demands),

(2) Loop: total head loss around a loop is zero.

The head loss-flow relationship is for most compounds nonlinear. Thus, either
Newton's method or the Hardy-Cross method are used for linearisation and nu-
merical solution.
The hydraulic simulation program utilises Hardy-Cross and Hazen-Williams
formulae for the head loss vs. flow calculations. It can simulate source pumps
(H = a - bQc, with a, b, c=constants, H =total static head. and Q=pump dis-
charge), booster pumps, altitude valves, check valves. and pressure reducing
valves among other things.
Cohen [38] develops a dynamic hydraulic model for Amsterdam's distribution
system. Advantages of dynamic models over static ones include (1) the possibility

11
to test the exploitation of the system, i.e. pumps, reservoirs. and controlling com-
ponents, (2) the possibility to compare different solutions in the design stage. (3)
the availability of travel times, (4) diagnosis capabilities, and (5) the possibility
to simulate calamities.
Controlled valves, pumps or boosters are necessary to be able to control the
flows in the network. In this case the dynamic simulation can be used to optimise
the system performance, even in terms of quality.
Appropriate laws of hydraulics, diagrams of the network, and statistical esti-
mates of the drinking water consumption are used to build the model. Quite an
extensive derivation of the equations is given. A model of linear partial differential
equations is constructed.
A example of a simple network is given and solution methods for large-scale
systems are sketched. Finally a brief description of the DYNA software package
is included.
The publication by Males [99] is concerned with a graph-theoretic approach
to hydraulic modelling. Definitions of graphs are given briefly.
Water distribution systems are viewed as directed graphs (node-link networks)
or digraphs. Indices describing the complexity of these graphs are possible (e.g.
cyclomatic number, accessibility index, dispersion index). Specific indices for
water systems are possible.
A public domain program WADSTAT can calculate these measures.
The paper by Andersen and Powell [3] highlights the problems associated
with the controlling elements that are generally present in water systems (like
non-return valves, pressure reduction valves or pressure sustaining valves).

2.3 Water Quality for Distribution Systems

As indicated in the introduction, water quality is generally defined in terms of
criteria which concern certain contaminants of water.
Train's book "Quality Criteria for Water" [143] contains a definition of water

quality criteria:

12
 "A specific concentration of water constituents which. if not exceeded,
are expected to result in an aquatic ecosystem suitable for the higher
uses of water."

Train gives a list of a number of water quality parameters, e.g. chemical elements:
or physical parameters like pH or temperature, and states the significance of each
parameter on one to ten pages each.
The book "Water Quality Surveys" [73] published by the C~ESCO and the
World Health Organisation (WHO) wants to be "a guide to the collection and in-
terpretation of water quality data" (subtitle). It includes the treatment of manv
practical matters and technical aspects concerning the collection and interpreta-
tion of water quality data. This book is, however, not restricted to distribution
systems.
Dojlido and Best [47] in their book "Chemistry of Water and \Yater Pollution"
give an extensive account of the chemistry involved in systems concerned with
water quality. In the last chapter, a water quality index is defined. This index
incorporates unit indices for each water quality parameter (range of 0 to 100) and
weightings of each parameter found through questionaires.
Sinha et at. [133] present a "Combined Water Quality Index" to evaluate the
water quality in distribution systems. This water quality index takes into account
dissolved oxygen, pH, chloride concentration, turbidity, residual chlorine, conduc-
tivity, and MPN counts", and weights the normalised values of these parameters
at each node with the flow through that node.
Adam and Kott [1] focus on the maximum growth rate of suspended bacteria
and conclude "it seems that the maximum growth rate technique could serve as an
additional parameter needed in the evaluation of water quality in the distribution
system" (p.1412).

ll\IPN counts are not defined in this paper.

13
2.4 Biofilms in Water Distribution Systems

2.4.1 Definition and Structure

Wimpenny [157] states that "[there] is no simple definition of biofilms since these
vary so dramatically in geometry and composition" (p.I}, however, he still at-
tempts to define a biofilm as "a predominantly two-dimensional microbial cornmu-
nity which forms at a solid/liquid interface and which may become spatially het-
erogeneous by virtue of physico-chemical gradients that develop within it" (p.1).
Bryers [20] gives a slightly different definition of biofilm (according to the 1984
Dahlem Konferenz [100)): "a biofilm is a collection of microorganisms and their
extracellular products bound to a solid (living or inanimate) surface (termed a
substratum)" (p.57).
Van der Wende and Characklis [151] define biofilm as "a layer of microorgan-
isms in an aquatic environment held together in a polymeric matrix attached to
a substratum [... ]. The matrix consists of organic polymers that are produced
and excreted by the biofilm microorganisms and are referred to as extracellular
polymeric substances (EPS)" (p.250).
According to this paper, biofilms are sometimes continuous, evenly distributed
layers, but "often quite patchy in appearance" (p.250). A maximum thickness of
a few hundred micrometres is said to be possibly reached by distribution system
biofilms. Biofilms are further described as generally heterogeneous (i.e. contain-
ing different distinct micro environments of bacterias). In addition, the film "often
contains organic or inorganic debris" (p.250).
The paper of Murga et al. [105] gives microscopic evidence for the thickness
variability of biofilm. Thickness profiles of several frozen single species biofilms
are presented. Depending on the species of bacteria in the biofilm, different
profiles are observed, ranging from a fairly homogeneous layer (for Pseudomonas
aeruginosa) to a very patchy biofilm (Klebsiella pneumoniae).
Stoodley et al. [138] found that biofilm after long periods (17 days) of either
laminar or turbulent flow exhibited different structures, either it contained cell
clusters intersected by channels (low flow) or it had developed streamers from

14
the trailing edges of the clusters (high flow). Keevil [80] distinguishes between a
basal layer of concentrated biofilm (thickness of ca. 5 /Lm) and a raised layer of
up to 100 uu: consisting mainly of micro colonies (a group of identical cells) that
form streamers (a string of cells, attached at the base and otherwise floating in
suspension) .

The book "Biofilms" by Characklis and Marshall [27] gives a comprehensive

overview of many aspects of biofilms, like their structure, rates of transfer pro-
cesses, models and applications in technology.

2.4.2 Biofilm Processes

In his paper "Biologically Active Surfaces: Processes Governing the Formation

and Persistence of Biofilms" [20] Bryers gives a well structured account of the
processes involved in the formation of biofilms:

• Deposition-related processes, i.e. the substratum preconditioning, the cel-

lular particle transport, and adhesion.

• Metabolic processes, i.e. growth and replication, substrate conversion (e.g.

in wastewater treatment), and extracellular polymer production.

• Removal processes, i.e. predator harvesting, shear-related removal, abra-

sion, and sloughing.

Van der Wende and Characklis [151] focus on examples of biofilm occurrences
in water distribution systems (i.e. case examples, however almost no quantita-
tive data are included), its relevance for these systems, and possible disinfectant
methods.
They define breakthrough as "the increase in bacterial numbers in the dis-
tribution system resulting from viable bacteria passing through the disinfection
process" (p.249) and growth as "the increase in viable bacterial numbers in the dis-
tribution system resulting from bacterial growth downstream of the disinfection
process" (p.249/250). Mentioned as main risks to water quality due to biofilms
are bad taste, odour, and a high number of bacteria.

15
 Quantitative results concerning possible disinfection of biofilms due to free
chlorine in a test reactor system are provided. Monochloramine is said to be
more effective in biofilm disinfection without data being given in this paper.
LeChevallier et al. [84] investigate which particular bacteria are part of biofilms.
Biofilm samples were taken through scraping and pipe coupons. The samples
were analysed with heterotrophic plate count (HPC) on R 2A agar and "five-tube
most-probable-number procedure" (p.2716) to enumerate coliforms. Total organic
carbon (TOC) and assimilable organic carbon (AOC) were also analysed.
Results include that chlorine had little impact on the occurrence of coliforrns:
a free chlorine residual of 1.0 mg/Iitre was insufficient for full disinfection; and
flushing did not necessarily reduce the amount of bacteria present.
In the introduction, Nagy and Olson [106] are cited as stating that "HPC
bacteria levels increase 1 log2 for every 10 years of service" (p.2714).
Donlan and Pipes [48] use a specifically developed "corporation sampling de-
vice (CSD)" to monitor biofilm in water supply systems. The CSD inserts sam-
pling cylinders in water mains which can be removed for lab assays. The inves-
tigated parameters were temperature, total chlorine, HPC, solution composition,
pH, alkalinity, TOC, and other chemical parameters. Some strong relationships
between the following were found;

• water temperature and HPC of bulk water,

• water temperature and chlorine concentration,

• flow velocity and HPC of sampling cylinders,

• cylinder HPC and season,

• HPC of attached bacteria (cylinders) and HPC of bulk water.

No strong relationships between individual nutrient parameters and attached mi-

crobial density could be found.

:2by a factor of 10

16
 :-Iartin et at. [101] use two glass plates, 33 x 2--1 ern, exposed to a flow of tap
water, to measure biofilm. Steady state was reached after only two days.
Servais et at. [129] compare bacterial data of several French distribution sys-
tems. According to them, bacterial growth can be controlled by both, chlorine
and the absence of nutrients. They establish the biodegradable dissolved organic
carbon (BDOC) as a good measure for available nutrients in water and there-
fore a good source for the estimation of the amount of bacteria present in the
absence of chlorine. The biofilm was evaluated from cast-iron coupons by an
original method based on exoproteolytic activity [136], "which has been shown
to be proportional to bacteria biomass" (p.12) [8].
The data from the distribution systems is evaluated In several tables and
graphs. Only very little bacterial activity was found in water with chlorine resid-
uals of 0.1-0.3 mg/Iitre.

2.4.3 Detachment from Biofilms

Detachment is a biofilm process like the ones discussed in the previous section.
However, since the detachment from biofilms is the main consideration in our
particle counts model of chapter 6, a separate section is devoted to it.
Bryers, in the above mentioned paper [20], distinguishes between four different
types of biofilm removal processes. Predator harvesting can be quite common, it
is the grazing of microbial biofilm by protozoa. Shear-related removal (or erosion)
is a continuous process. Bryers sees a connection between shear, biofilm thickness
and flow velocity. Abrasion is not very important in potable water distribution,
but features in some biological reactors. It is the removal of biofilm due to the
collision with particles. Sloughing, finally, is an event where large amounts of
biofilm are detached. It is an apparently random, discrete process.
Pevton and Characklis [113, 114, 115] utilise the same equipment as used in
[151]. Thev focus on the relationships between glucose concentration or rotational
speed and biofilm thickness or detachment. In the investigated test reactor system
glucose concentration is proportional to both biofilm thickness and detachment

17
whereas rotational speed has no influence on both. In addition they correlate
various published models of biofilm detachment with their experimental results.
Stewart [137] provides a short summery of the previously published models of
biofilm detachment that are also used in [115]. All models describe the detach-
ment rate Td and contain the following:

T di = detachment rate of component i, in kgl (m 2 sec),

kd = detachment rate coefficient, in varying units,
Pi = density of component i in the biofilm, in kg/rn",
Lf = biofilm thickness, in m,
7 = fluid shear stress, in N 1m2 .
Various equations for the detachment rate have been put forward, Stewart
gives the following examples:

r« kdPiLf (2.1)

r« kd(PiLf )2 (2.2)

Tdi kdPiLJ (2.3)

0 58
r« kdPi L f7 . (2.4)

r« k d7Pi (2.5)

Rittman proposed eq. (2.1) and the shear stress dependence of eq. (2.4) [122].
He reviews this and other models in [122] which stands for a number of contri-
butions to the field by him.
Characklis et al. [28] proposed eq. (2.2), Gujer and Wanner [61] proposed
equation (2.3) as part of their model of biofilm processes (cf. section 2.4.6), and
Bakke et al. [6] suggested the first-order (linear) dependence on shear stress in
eq. (2.5).
Both, Peyton and Characklis [115] and Stewart [137], propose more compli-
cated detachment models. Peyton and Characklis suggest a dependence on the
product of the nutrients utilisation rate, yield coefficient of biomass on substrate.
flow velocity and biofilm thickness. Stewart includes the (local) growth rate of

18
bacteria (which also depends on nutrients and yield) and a variable depth of
shear-off.

Dukan et al. [49] investigate three different detachment models (cf. sec-
tion 2.5.2): (1) detachment is proportional to the number of fixed bacteria (2) de-
tachment takes place only over a certain thickness proportionally to the number
of fixed bacteria and their growth rate, or (3) detachment takes place through-
out the thickness of the biofilm, otherwise like (2). They choose model (1) for
the detachment from the chlorinated layer of the biofilm and model (3) for the
non-chlorinated layer.
All of the publications discussed in this section are chiefly concerned with
medium- or long-term variability of biofilm thickness due to shedding (or shear-
off). Thus, the time dependence is generally assumed to be linear (i.e. a rate
per time unit is found) or slowly varying. Only Peyton and Characklis [115] and
Bakke and Olsson [7] mention that it was observed that shear-off briefly increases
during a change of shear force, however, with an return to original values after
five retention times in the experimental apparatus (a chemostat reactor). They
do not investigate the transient dynamics any further.
In the work of the present thesis it is found that transient (i.e. short-term)
phenomena can have a decisive impact on the amount of biomass or inanimate
particles in the distribution system over periods of several days. In addition, tran-
sient detachment may account for the removal of significant amounts of biofilm.
Therefore, the transient phenomena are investigated in detail to obtain informa-
tion about the amount of biofilm present in the system (cf. chapter 6).

2.4.4 Disinfection of Biofilms

LeChevallier et al. [85] examine the potential of several biocides (in particular free
chlorine and monochloramine) for the inactivation of biofilm bacteria. They show
experimentally that monochloramine is a more efficient disinfectant of biofilms.
While for free chlorine the resistance of biofilm bacteria is 150 to 3000 times
higher than the resistance of suspended bacteria, resistance to mono chloramine

19
was found to be only 2 to 100 times higher for attached as opposed to unattached
bacteria.
In LeChevallier et al. [86] it is investigated what happens with biofilm in
distribution systems if a disinfectant residual is applied.
A small (21.95 m or 72 ft) pipe system with controlled pH, temperature and
flow rate consisting of different pipe materials is investigated. Chlorine and
monochloramine are the independent variables and ammonia, phosphate, tur-
bidity, water conductivity, TOC, glucose, and biofilm are measured. Biofilm is
measured by

• 25 mm x 25 mm glass coverslips placed into pipes at quick-disconnect

couplings; thickness is measured microscopically,

• permanently mounted electrodes monitoring the increase in electrical resis-

tance (however reported results "did not appear to be related to biofilm
thickness or plate count density", p. 15),

• scraping the interior of pipes and measuring the amount of viable bacteria
through heterotrophic plate count (HPC).

Results include the following:

• The effectivity of disinfectants depends heavily on the pipe material.

• Monochloramine is (if higher than a certain threshold of roughly 2 mg/I]

more effective against biofilm (especially on cast iron).

• Corrosion has a major impact.

• ext values (concentration of disinfectant x time for the inactivation of 99%

of the biofilm) do not differ much between attached and dissolved bacteria,
see table 2.1.

The "attached" bacteria were scraped from the pipe wall and than put in

suspension.

• Biofilm thickness is between 60 and l IOum.

20
 Table 2.1: ext values
attached unattached
Free Chlorine 0.1 mg;nm O.OSmg;nm
Monochloramine 95 mgmm
I
14.Smglmm

• There is no linear relationship between disinfectant concentration and dis-

infectant efficiency (i.e. doubling disinfectant concentration does not result
in twice the inactivation).

Clark et al. [36] present a study of the effects of various disinfectants (0 3 , free
C1 2 , NH 2CI) on biofilm.
A pilot facility at Nancy is used to investigate biofilm. It consists of two sets
of three loops with an average retention time of 24 hours per loop. The biofilm
is evaluated from coupons consisting of pipe materials with a wetted surface of
2cm 2 placed at the end of a sampling device probe which was inserted flush with
the pipe wall. The bacterias on the surface of the sampling device are removed
through sonication.
It is assumed that the biofilm reaches steady state after three to five weeks
(depending on disinfectant residual). In lab assays the following was analysed:

• free and total chlorine,

• dissolved organic carbon (DOC),

• biodegradable dissolved organic carbon (BDOC),

• bacterial density

from water samples (bulk water),

from coupons (biofilm),

HPC were used to enumerate viable bacteria and microscopic counting
to find total bacteria.

• disinfection by-products.

21
It is observed that "the biofilm build-up increased with residence time in the loop,
and increased dramatically after disinfectant residuals disappeared" (p.27-1).
In Block et at. [13] the same experimental set-up as in [36] is used.
Biofilm accumulation seems to be not prevented by chlorine. An equation
for the biofilm growth rate (from [152]) is given and the relationship between
the amount of total attached cells and DOC is provided (found through linear
regression) .
Block et at. [12] use again a very similar test apparatus to the one used in the
previous papers [36, 13]. Results include the following

• chlorine slows down the formation of biofilm but does not prevent it,

• there exist log linear relationships of attached cells vs. chlorine concentra-
tion, planktonic cells vs. chlorine concentration, and planktonic HPC vs.
chlorine concentration,

• chlorine was more effective than mono chloramine (on suspended bacteria
and biofilm),

• a model was made but only graphs are presented,

• BDOC increases the amount of cells even if chlorine is present.

In the report by Packer [112] of the Centre for Applied Microbiology and Re-
search (CAMR) a test apparatus consisting of a seed vessel and a test chemostat
is used to investigate the water quality of the water treated by the Advanced
Water Treatment Centre (AWTC) of Thames Water Utilities at Kempton. The
AWTC was also used as a source for the experiments described in this thesis.
This report investigates which particular bacteria form part of the biofilm
in the test chemostat for both with and without mono chloramine disinfection.
Some results are that coliforms disappear after three days of high monochloramine
residual; coliforms live in biofilm longer than suspended; and the A\YTC water
does not have a significantly changed nutrient content compared to water from
previous treatment.

22
2.4.5 Measuring Biofilms

Frequently, biofilm samples were taken through scraping and pipe coupons [-15.
84, 86, 36, 12, 129]. Others use devices which are inserted into water pipes for a
period of time and then evaluated [48, 101, 86].

Jass et at. [75] use a modified Robbins device coupled to a chcrnost at. The
Robbins device is a turbular flow system filled with a microbiologically controlled
feed water, containing circular test pieces (coupons) for the evaluation of biofilm
on different surfaces in a reproducible manner. The modified Robbins device
includes retractable pistons from the inside of a rectangular section pipe [19].
Van der Kooij and Veenendaal [149, 150] use a special biofilm monitor devel-
oped by the Dutch company KIWA NV Research and Consultancy to measure
biofilm. This biofilm monitor consists of a glass column (diameter 2.5 em; length
60 cm) containing glass cylinders on top of each other. The column is supplied
with the water to be investigated with a constant velocity (0.2 ~). Periodically
one or two cylinders are taken from the column. Subsequently, biomass den-
sity according to adenosine triphosphate (ATP) analysis and heterotrophic plate
counts is obtained.
In [149] a large amount of data (in graphs) on biofilm formation with no
chlorine present is given. Generally, the growth of a biofilm starts after 10 days
and is normally monitored for 75 to 150 days. This paper claims that the analyses
of AOC and BDOC is not enough to determine biofilm growth, instead the biofilm
is evaluated with the ATP method and the biofilm formation rate (BFR) and the
biofilm formation potential (BFP) are evaluated. (BFR and BFP are defined by

the authors in [149].)

In [150] again the biofilm formation characteristics BFR and BFP are eval-
uated (using the biofilm monitor). Other methods used to determine "the rate
and extent of bacterial multiplication in samples of the water to be investigated"

(p, SSl-2) are:

• colonv counts .

• turbidity measurements.

23
 • ATP analysis,

• microscopic techniques,

• assessing the degree of removal of dissolved organic carbon (DOC). i.e. the
biodegradable dissolved organic carbon (BDOC) ~ or

• assessing the concentration of easily assimilable organic carbon (AOC).

The experimental set-up used for the experiments discussed in this thesis,
the Kempton Park "TORUS" pipe rig of Thames \rater [69] contains coupons
flush with the pipe wall. Due to the difference between pipe wall and coupons
in surface structure if investigated under the microscope, the small size of the
coupons (1 cm 2 ) and the limited number of coupons (for continuous sampling),
it was decided to consider only indirect measures of biofilm in this study (i.e.
particle counts, suspended bacteria).

2.4.6 Biofilm Kinetics and Models

Kinetics are the first step towards a full model. Investigations in biofilm kinetics
imply that the data is collected with the intention to produce a rigorous descrip-
tion of the underlying processes, generally in mathematical terms, possibly at a
later stage or by somebody else. A model is normally a set of many one-equation
kinetic descriptions, each of a process that interacts heavily with the others.
The paper by LeChevallier [83] is concerned with the major factors of microbial
water quality i.e. breakthrough, growth and disinfection. According to the
author "[there] is no unified model to describe microbial processes in distribution
systems, although almost all of the parameters have been studied and could be
modeled" (p.365). However only an outline of a possible model is presented. A
number of useful references are given.
Breakthrough consists of bacteria passing through the treatment process into
the distribution system. Tvpical amounts of bacteria getting into the system
"mav range between 10 and 100 cfu/rnl" (p.365) (plate counts on R 2 A agar).

Although injured bacteria may comprise a significant portion of the total cell
population they are not normally found with standard heterotrophic plate counts
(HPC). These bacteria may survive due to attachment to organic or inorganic
particles.
The growth rate of biofilms depends on

• the physiochemical (i.e. chemical, thermodynamical) properties of the in-

terface,

• the physical roughness of the surface,

• the physiological factors of the attached microorganisms,

• shear forces, and

• the possible reactions of disinfectants with extracellular polymer substances

(EPS).

Factors influencing bacterial growth are

• temperature and rainfall,

• nutrients availability,

• ineffectiveness of disinfectant residual,

• corrosion and sediment accumulation, and

• hydraulic effects.

The most important nutrients required for bacterial growth are carbon, m-
trogen and phosphorous in the ratio 100:10:1 (C:N:P), thus assimilable organic
carbon (AGC) is growth limiting. Therefore, the author suggests combining a
hvdraulic model with AGe levels, bacterial growth rates and disinfectant kinet-
ICS.

Finally, disinfection does not totally prevent bacteria from proliferating. Some
factors t hat have an effect on bacterial survival in chlorinated waters are

• at tachment to surfaces,

25
 • bacterial aggregation,

• age of biofilm,

• encapsulation,

• previous growth conditions,

• alterations in bacterial cell wall and

• the choice of disinfectant.

Again a number of useful references are given.

In Trulear and Characklis [145] an annular reactor is used as the test equip-
ment (at 30°C ± O.5°C). The thickness of the biofilm on a removable slide is
determined microscopically. Again glucose is the source of nutrients for the bac-
teria.
A typical experiment progression normally includes three phases: induction,
accumulation and plateau.
This paper includes a number of equations of e.g. the mass balance across
the reactor of glucose accumulation or the accumulation of attached biomass
among others. Data is given for the net biofilm accumulation and several other
dependent variables changed through the independent variables rotational speed
and glucose concentration. In their discussion the authors try to explain some of
the observed effects. An interesting concept in this context is the active biofilm
thickness which represents a maximum depth of penetration of glucose into the
biofilm (dependent on glucose concentration).
Block et al. [12] present some results of a model differential equations which
also includes biofilm, however, without presenting the model.
The papers of Gujer and Wanner [62. 61] develop a model for a mixed pop-
ulation biofilm. They develop a model of biofilm processes from physical (or
biological) principles, in particular mass transfer relations. :\umerous equations
are provided.
Gujer and' Vannor discuss briefly the merits of deductive modelling (as is done
by t hem) over inductive modelling, which they see as using a "black box." One

26
of the chief advantages of their approach is that "the underlying mathematical
derivations are consistent" (p.398) and "that each assumption must be clearly
identified and its significance discussed" (p.398). The approach taken forward in
this thesis aims at a "grey box" (thus it is basically inductive. i.e. it starts from
data) rather than a system with no explanatory power, which implies that the
given advantages of deductive (consistency, clear assumptions) are still aimed at
(cf. section 2.11).

Dukan et al. [49] also include the biofilm in their model of water quality. see
section 2.5.2.

2.5 Suspended Material

The following papers are concerned with material that is suspended in water
distribution systems, which may include both biomass and inanimate material.

2.5.1 Suspended Biomass

Some papers mentioned in section 2.4 are partially concerned with bulk fluid
bacteria, e.g. Nagy and Olson [106], Servais et al. [129], Block et al. [12] or
LeChevallier et al. [85].
Huck [71] gives a review on measurement methods of suspended organic mat-
ter. There are a number of different methods discussed and also examples (mainly
in water treatment processes) given.
On a conceptual level, Huck distinguishes between measurements aimed at
bacterial regrowth and measurements aimed at the effects on chlorine demand.
In the former case, AOC should be investigated, in the latter BDOC.
Also the paper of Maier et al. [93] discusses many measurement principles
for different parameters. In particular the following parameters are dealt with:
turbidity, colour, Uv'-absorption, nitrate, fluor carbohydrates, mangan, fluor.
aluminium, ammonia, phosphate, temperature, conductivitv, pH. oxygen. redox-
potential, chlorine, and ozone.
 The main result of the paper by Haas et al. [63] is that "[the] results of
the statistical analyses did not support the hypothesis that the physiochemical
measurements would suffice to produce reliable predictions of microbial water
quality" (p. 475).

In the article by Goshko et al. [57] no strong relationships between plate counts
and other parameters like chlorine residual or turbidity could be established.
Samples were obtained from real distribution systems.

2.5.2 Kinetics of Suspended Bacteria

The model used by Block et al. [12] also includes suspended biomass, however,
the model equations are not presented.
Dukan et al. [49] present a model that incorporates biodegradable dissolved
organic carbon (BDGC), temperature, residual free chlorine, pH, hydraulic con-
ditions (flow, diameters), free (i.e. suspended) bacteria and biofilm bacteria. For
simplicity, the biofilm is modelled as a uniformly distributed over the pipe sur-
face, however, they distinguish a chlorinated and unchlorinated layer within the
biofilm. Equations of bacterial growth and its temperature dependence are given
with source, the action of chlorine on bacteria is also presented mathematically
(its source is not entirely clear from the paper). The model of bacteria mor-
tality takes into account protozoa grazing which implies that the temperature
dependence of bacteria mortality does not follow an Arrhenius law.
A number of equations are provided, e.g. chlorine disappearance kinetics,
the concentrations of BDGC, free bacteria and active free bacteria, mass balance
equations, both for chlorinated and non-chlorinated biofilm, for BDGC, fixed
bacteria and active fixed bacteria and the mortality of fixed and free bacteria.
Detachment of bacteria from biofilm is identified as an important factor for
the amount of free bacteria. Three different detachment models are presented
and two of them are chosen for t he chlorinated and unchlorinated biofilm laver.

respccti velv.
Finallv some initial results are given. Thev find steady state thresholds for

28
BDOC (0.2-0.25 ppm) and temperature (16°C ± 1°C) below of which no signifi-
cant multiplication of free bacteria takes place. They also simulated the response
to a step in BDOC from 0.5 ppm to 0.4 ppm. The BDOC changes rapidly in a
system with a maximum retention time of 60 hours (change within tens of hours).
The active fraction of fixed bacteria not attained by free chlorine and the quan-
tity of fixed biomass change within 10-25 days. The concentrations of active free
bacteria and total free bacteria settle down to a nearly stable value after about
5 days but still exhibit some slow trend until about 30 days.
The publication of Piriou et at. [117] gives some more information on the
model introduced in [49]. It gives a short description of the mathematical model
and of the software package Piccobio. Model validation experiments, both on a
experimental pipe rig and real distribution systems, are presented together with
some simulation results. Apart from the thresholds already discussed in [-!9] they
also find the water quality problems and the location of a booster chlorination
station through simulation. Piriou et at. [118] give another overview of the func-
tionality of Piccobio, however, without mentioning the name.

2.5.3 Inanimate Particles and Particle Counters

Apart from bacteria, inanimate particles may be present in the bulk liquid of
water pipes. They (together with the biomass) can be measured by particle
counters, provided their size falls into an appropriate size range.
The paper by O'Shaughnessy et at. [110] evaluates the performance of particle
counters in comparison with counts made by microscopic processes. In particular
a forward-angle light scattering (FALS) particle counter sensor was compared to
scanning electron and optical microscope counts, and the FALS sensor was com-
pared to a light obscuration sensor (as used in the experimental set-up used for
this work, the "TORUS" pipe rig). It was found that particle counters undercount
in comparison to both microscopic counts, with the degree of undercounting in-
creasing as size range increases. Furthermore, the comparison of particle counters
showed that light scattering sensors were more accurate than light obstruction

29
sensors when measuring latex spheres. However, when counting microorganisms,
light obstruction sensors were better than light scattering counters. Since coun-
ters are calibrated for latex spheres, sizing particles, in particular if the particles
are not spherical, is frequently inaccurate and therefore, "total counts made by a
particle counter within a particular size range should be viewed relative to pre-
vious counts and not as an absolute indication of the number of particles with
diameter associated with that size range" (p.69). A number of references are
included.
The American Water Works Association Research Foundation report "A Prac-
tical Guide to On-Line Counting" by Hargesheimer and Lewis [65] gives an ex-
tensive account of the practicalities involved in using particle counters. They
discuss the use of particle counters (they have been used most frequently for
the evaluation of filter efficiency in water treatment), state the different types
of counters (light obstruction or scattering), elaborate on how the sensors work,
show different ways of looking at counts (e.g. total counts or size distributions),
give a wealth of information on how to achieve best results when using particle
counters and talk about quality assurance during operation. Both, batch (i.e.
one-off) particle counters and continuously on-line counters are discussed.
Hatukai et al. [66] use particle counters to monitor water treatment. They
find (among other things) that particle counters are more sensitive to changes
of water quality as result of modification of operating conditions than turbidity
meters.
In the next section, the publications by Woodward et al. [159] and Ta [140]
will be discussed (among others) which are both investigating particle counts in
the "TORUS" pipe rig, which was also used for the experimental work presented
in chapter 3 of this thesis.

2.5.4 Propagation of Conservative Substances

The first work on water quality was concerned with the age of water. This.
however, gives naturally rise to the investigation of conservative (i.e. non-reactive)

30
substances since the methodology is the same. the difference is simply the use of
a 'substance instead of just water.
The C.S. Environmental Protection Agency (EPA) report written by Clark
et al. [35] gives the most extensive account of the example for the propagation
of conservative substances in a actual water distribution system also used in
[59. 32~ 30, 60]. The results of the steady state model and the development of
the dynamic model are presented.
It is remarked that already "hydraulic analysis and verification of Bows in a
system is an art in itself' (p.2).
Grayman et al. [60] introduce the concept of "dynamic water quality models"
as opposed to steady state models. They focus on conservative substances, how-
ever mention how non-conservative substances could be dealt with (first order
kinetics).
This paper gives a rather detailed account of the practicalities involved in
developing dynamic models, e.g. the trade-off between simulation time and model
accuracy involved in the choice of the discretisation time step. The concept of a
dynamic model is described in words.
In addition, the same example as in [59~ 35, 32, 30] is given.
Clark and Goodrich [32] includes a brief overview of modelling very similar
to that in [59] and a short discussion of trihalomethanes (TTH~ls) simulation
(which is more extensively given in [37]). In addition. it states some ideas on how
the actual human exposure to water contaminants could be modelled.
Clark [30] gives an extensive case study of the application of both a steady
state and a dynamical model for the prediction of various water quality parame-
ters (like amount of disinfectant by-products present or substance propagation).
A fluoride tracer experiment in an actual distribution system is described (same
example as in [59, 32, 60, 35]).
To model the system, EPA~ET [124~ 123]. a public domain hydraulic and
water quality modelling package provided by the CS Environmental Protection
Agency is used.
The paper of Biswas et al. [10] gives a number of interesting references. The

31
authors apply methods of fluid dynamics to develop a model of deposition, to
calculate velocity fields or to investigate cavities and bends. A lead accumulation
and transport model is presented consisting of partial differential equations. In
addition, results from modelling chlorine are presented.
Kennedy et al. [81] did fluoride and chlorine tracer studies to determine travel
times. Results of these experiments are given graphically.
Woodward et al. [159] investigate the behaviour of particles in the Kempton
test rig of Thames Water. Results from a step flow trial validate the generally
made assumption of plug flow (for turbulent flow conditions), since the particle
profile is propagated through the rig without dispersion.
The number of particles increases for a short period of time sharply with
flow step increase. The explanation for this phenomenon is the existence of "hot
spots", in which loose deposits accumulate preferably. These deposits are lifted
into suspension with a step increase of the flow rate.
The internal report of Thames Water Utilities Ltd. "Analysis of the Particle
Counts and the Chlorine Decay in PIPE RIG During the Step Flow Trials" by
Tuan Ta [140] deals with the same experiments as [159, 160]. As additional
information it contains the way in which the travel times were calculated. The
average of two appropriate subsequent flow data points was used to calculate the
volume covered in a given time interval.

2.6 Chlorine as Disinfectant

2.6.1 Chlorine Chemistry and Simple Chlorine Kinetics

An extensive treatment of chlorine chemistry and other issues involved in chlori-

nating potable water, wastewater, cooling water, industrial processes, and swim-
ming pools is given in White's "Handbook of Chlorination and Alternative Dis-
infectants" [156].
A good, broad and concise overview on chlorination is given in the Blue Pages
of the International Water Supply Association on "The Practice of Chlorination:

32
Application, Efficacy, Problems and Alternatives" [74]. This booklet gives a
biological and chemical view on chlorine in distribution systems without including
quantitative kinetics (e.g. decay coefficients). In addition, brief mention is made
of the use of organic chloramines and mono chloramine as disinfectant.
The report of Thomas [141] gives a summary of what was going on in chloram-
ination in Australian Water Authorities at that time. The information provided
in this report includes that

• the aim of most water authorities is to obtain 0.2 to 0.5 mg/l at the extrem-
ity of their system, however sometimes only the monochloramine residual
decreases to 0.05 mg/I;

• the ammonia injection point can be clogged with calcium carbonate;

• in laboratory tests mono chloramine is 25-100 times less effective than free
chlorine, however field results are better, possibly due to an initial reaction
of free chlorine as disinfectant before it forms mono chloramine;

• microbial results (in particular concerning coliforms) differ substantially for

the systems considered;

• the amount of disinfection by-products (THMs) is far smaller with chlo-

ramination compared to free chlorination; and

• false free chlorine residuals in the presence of ammonia and false monochlo-
ramine residuals due to indistinguishable organic chloramines are observed.

Norton and LeChevallier [109] investigate the effect of chloramination on water

quality in distribution systems. Monochloramine doses in excess of 2 mg/Iitre
were used with the result that in general, chloramination seemed to be more
effective against coliforms and live bacteria as measured with HPC than free
chlorine. For evaluation, they consider HPC, coliforms, corrosion, disinfection
by-products and nitrification among other things. Many references are included.
The article of Ram and Malley [120] states also the possibility of a false free
chlorine reading for various tests. In tests performed under laboratory conditions

33
the apparent chlorine residual was sufficient to determine that bacteria will not
grow (with actual chlorine doses of 0.5 to 1 mg/I).
Feng [52] discusses the behaviour of organic chloramines in some detail. It is
found that organic chloramines have no or little disinfecting power, in particular
the lethal activities of organic chloramines are lowest at pH near neutral (pH 7).
Water distribution systems generally hold a pH of about 8.
Gotoh [58] assumes an approximate first order decay of free chlorine. As
possible reasons for chlorine decay he gives" (1) degradation of free chlorine, (2)
reaction with contacting pipe material, and (3) reaction with oxidising substances
in water" (p. SS 21-18). The first order decay factor is determined for cast iron
pipes with no lining, epoxy resin-lined cast iron pipes and mortar-lined ductile
cast iron pipework. Other materials were investigated but results are given only
for these materials. In addition, a field survey procedure is provided.
In an experiment carried out by Nic Proux and reported by Rory Todd [142] it
was established that nitrite can not exist together with free chlorine in the same
sample, i.e. nitrite will react with free chlorine to form other chlorine species.

2.6.2 Modelling of Reactive Substances

Modelling reactive substances is the natural next step from modelling conservative
substances as discussed in section 2.5.4. In this section, model equations for
chlorine decay and case studies of their application will be discussed together
with a more detailed focus on the issue of decay coefficient variability.

2.6.2.1 Models

Rossman [123] describes EPANET, a public domain program package for hy-
draulic and water quality modelling developed by the U.S. Environmental Pro-
tection Agency. EPANET is described in detail in [124].
Rossman gives the equations of the hydraulic model (possible utilisation of
Hazen- Williams, Darcy-Weisbach or Chezy-Manning formulas) and the water
quality model equations together with a short introduction to the computer im-

34
plementation.
EPANET can model the propagation of conservative and reactive substances
(i.e. free chlorine). The chlorine decay coefficient K for first order chlorine decay
depends on both bulk water demand k b and wall demand k w . Thus,

(2.6)

(with kf=mass transfer coefficient between bulk flow and pipe wall and RH=hy-
draulic radius). The mass transfer coefficient kf depends among other things on
the Reynolds number and therefore on the flow.
The chlorine dynamics are

ac(x, t) _ _ L ac(x, t) _ K ( ) (2.7)

at - A ax c x, t

(with c=chlorine concentration, t=time, x=axial distance, f=flow, A=area) and

the solution is

(2.8)

with

(2.9)

where t., t2 represent time, Xl, X2 represent upstream and downstream distance
along the pipe, respectively, and v = ~ is the constant velocity of water in the
pipe,
Colin and Grapin [40] combine their own hydraulic model (AREMAN) with
a water quality model of the form

(2.10)

with Ci=chlorine concentration, Ez=axial dispersion coefficient. v=water velocity

35
in the pipe, Ci=tracer concentration, x=space variable, <Pi=production term and
index i=ith pipe.
They use highly parallel hardware and thus can solve systems with upto 10000
of these partial differential equations (PDE).
In Biswas et al. [11] a more complicated model of chlorine decay: incorporating
the biofilm demand, is given. The model equations are derived by methods of
fluid dynamics and are valid only for steady state, i.e.

ac
-=0 (2.11)
at

(with c=chlorine concentration).

As the model of chlorine decay the following equation is found:

ac =
g(r)- 1
--
c a2 + - a (rac)
A o- Alc
ax Pea ax2 r ar ar- - (2.12)

(with g(r) = 1 for plug flow, r=radius, Pea = l;, l=pipe length, v=flow velocity,
d=diffusion constant).
If axial diffusion is neglected and plug flow is assumed, the following equation
is obtained.

ac =
ax
Ao~
r
(r
ar Br
ac) _ Al c. (2.13)

The wall demand of the biofilm is one boundary condition of this partial differ-
ential equation.
Taking into account all boundary conditions the radial average of the chlorine
concentration is obtained to be:

(2.14)

"Thus, three dimensionless parameters, A o, AI, and .4. 2 , govern the chlorine
decay in the distribution system. A o accounts for the radial diffusion and depends
on the pipe length, on the effective diffusivity of chlorine and on the flow rate

36
throughout the system. Al depends on the reactivity of chlorine with species
such as viable cells or chemical compounds in the bulk liquid phase and on the
residence time in the system. A 2 is a wall consumption parameter depending on
the wall consumption rate, on the pipe radius and on the effective diffusivity of
chlorine" (p.1718). The An are the solution of AnJI(An) = A 2Jo(A n) where Jo and
J 1 are the Bessel functions of the first kind of order zero and one.
In addition, an algorithm to solve this equation numerically is provided.
The paper by Lu et al. [91] is following-up [11]. It gives an extensive model
of dynamics in water systems for conservative substances, chlorine and biofilm,
in particular biofilm thickness. It encompasses 13 partial differential equations
and is based on fluid dynamics and mass balance equations. The wall demand is
taken into account and within the bulk fluid first order kinetics of chlorine species
are assumed.
A number of simulation results are graphically depicted.
Chen and Brdys [29] develop a frame work for the estimation of parameters
and states of a system including both quantity and quality with known measure-
ment bounds. Quality is defined through the concentrations of reactive substances
as used in EPANET [123]. Brdys et al. [16] take this one step further by devel-
oping an operational control scheme for quantity and quality (again as a possibly
reactive concentration).

2.6.2.2 Case Studies

Wable et al. [154] use Piccolo to model chlorine concentrations in water distribu-
tion systems.
First order kinetics for chlorine decay are derived from the chemical equations.
The network model is based on a hydraulic model and plug flow, no radial or axial
diffusion, perfect mixing at node, and conservation of fluxes are assumed. After
the hydraulic regime is known they find the residence time, the concentration of
any compound, and the origin of the considered water volumes. The procedure
to achieve this requires the integration of the kinetics over the residence time for
each pipe, and for each node the mean of the concentration is averaged by the

37
flow rate.

They consider single pipes in an actual distribution system (Paris) to deter-

mine the free chlorine demand and show that a log-linear relationship between
the decay coefficient and temperature exists. Similarly, TOC is related to free
chlorine decay.

The paper of Sharp et al. [132] attempts to measure free chlorine decay
through measuring flow (i.e. travel time) and chlorine concentrations. An un-
certainty analysis was done and results from experiments in an actual system are
given.

Liou and Kroon [88] give another example of the application of a dynamic
model of the propagation of conservative substances to an actual distribution
system and outline the theory used for non-conservative substances.
LIQVARS by Stoner Associates is used to develop the steady state hydraulic
model. The additional dynamic model divides the pipes in volume elements and
propagates them with every time step.
Hunt and Kroon [72] tune first order free chlorine decay coefficients to actual
data.
Chambers and Joy [25], in a Water Research Centre (WRc) report prepared
for Thames Water, aim to validate first-order kinetics for free chlorine decay.
They did an extensive sampling programme, but less extensive statistical anal-
yses. Problems occurring during the project are discussed (e.g. some of the
measurements were discarded because the sampling locations were not represen-
tative for what they theoretically should stand for). The precision of the chlorine
measurements could be significantly improved through the adoption of an accu-
rate sampling procedure.
It is noted that the chlorine decay in bulk water samples (bottle tests) is very
different from the actual chlorine decay in the system (for which the author of
the current work assumes the biofilm to be responsible).
Clark et al. [34] present a case example of modelling chlorine decay III an
actual distribution system. In this system there is a high wall demand.
It is remarked that storage tanks. due to long retention times, may cause

38
serious water quality problems. In addition, it is mentioned that a "simulated
distribution system or a pipe loop" (p.885) was at the time of submisson under
construction.

Heraud et al. [67] present results from sampling in a real system in France
together with the application of chlorine models incorporated into Piccolo (cf.
[49]). Sampling in a real distribution system brings the draw-back of fluctuations
in inlet chlorine (0.28-0.97 mg/I) and temperature (l1.6-24.1 0C).
Sampling was performed at hydrants at a flow rate of about 1.4 lis. No
velocity is given (or sample tube diameter). A transient response of the sampling
process (due to hydrant deposits?) was observed and for this reason the sampler
waited 5 min for bacterial counts to level off before the sample was taken.
The analytical programme contained analyses of temperature, dissolved oxy-
gen, redox potential, turbidity, iron, particle counts (method or results not fur-
ther specified), TOC, BDOC, UV 254nm absorbance, free chlorine ('Chlorscan' mi-
crosensors are discussed in the paper, including measurement principle and fur-
ther literature; their limit of detection is 0.02 mg/I), trihalomethanes (THM),
HPC on PCA medium at 22°C and 37°C, HPC on R2A at 22°C, total coliforms,
fecal streptococceae, Aeromonas hydrophilia and Pseudomonas aeruginosa.
A microbial evolution over time and location was observed. There is a small
decrease in bacterial counts over a period of 50 min (fast decay of chlorine) and
significant changes happened over a period of several months (influenced by both
chlorine and temperature).
The free chlorine model is based on first order exponential decay with decay
coefficients dependent on the pipe material (old grey cast iron, cement-lined grey
cast iron, and other) and pipe diameter. From representative data for two diam-
eters the decay coefficients for other diameters were calculated using a equation
given in the paper (p.69). Neither derivation nor source are provided, but it is
mentioned that Piccolo is used for calculating the coefficients (i.e. this equation
might be built into Piccolo, or it could be otherwise empirical).
Chlorine model and measurements agree reasonably well. According to the
authors the differences are due to the shortcomings in hydraulic modelling (e.g.

39
inaccurate demand patterns). The model identified regions with no chlorine and
thus it found effective places for booster chlorinations.
Piriou et al. [118] includes some more information on the chlorine meters that
were also used in [67].
Clark et al. [37] present the propagation of a conservative substance (trihalo-
methane, TTHM) and a first order decay of free chlorine.
The chlorine decay constant incorporates in this paper the wall demand k w
as used in EPANET (found through trial and error). It is demonstrated that
TTHMs are closely related to UV absorption.

2.6.2.3 Variability of Decay Coefficients

One important issue in chlorine decay modelling is the question if the decay
coefficient is a constant parameter (i.e. first order decay) or varies with flow,
biomass, pipe material or other factors. EPANET [124] uses a decay coefficient
which includes a flow and biofilm dependence through the wall demand. Heraud et
al. [67] use decay coefficients dependent on the pipe material. Other relationships
are possible.
Some researchers validate the assumption of first order free and total chlo-
rine decay, like Chambers et al. [24] of the WRc. Only little statistical analysis
is included. The simulation packages WATNET and WATQUAL are used for
hydraulic and water quality modelling, respectively.
Burgess et al. [21] give an outline of an experiment procedure to develop
dynamic water quality models incorporating first order chlorine decay with wall
demand. The results of this project are presented in Burgess et al. [22]. It is
observed that first-order kinetics seem to fit the data, however the decay factor
is system dependent. A temporal variation of typically about 40 % of the mean
decay coefficient is generally observed. No explicit expression for the wall demand
is provided.
Lungwitz et al. [92] investigate chlorine decay in an actual distribution system.
The relevant equations, as given in Rossman's paper on EPANET [123, 124], are
provided. It is observed that the chlorine decay coefficient assumes different

40
values depending on whether chlorine is increased or decreased at a pipe inlet
(booster chlorination). This phenomenon is explained with axial diffusion due to
a concentration gradient.
Elton et al. [51] are concerned with both the propagation of conservative
substances (including source blending) and the propagation and reaction of non-
conservative substances, i.e. free chlorine. For chlorine decay first order kinetics
are assumed.
Four different real-world water distribution systems were investigated. how-
ever only few results of models are given. It was found "that chlorine decay rates
depend on initial concentration, pipe diameter, pipe material, pipe condition. wa-
ter temperature, and pH. These are in addition to the influence of source water
characteristics" (pA9). The authors work for Severn Trent Water and used the
Stoner Workstation Service to model these distribution systems.
Schneider et al. [128] find that the free chlorine decay coefficient depends
on flow and other parameters like pH and temperature. They use methods of
continuous time system identification, the Poisson Moment Functions [148], and a
least squares estimator for decay coefficient determination. First and second order
chlorine decay models are investigated. Woodward et al. [160] found evidence
that the monochloramine decay constant is flow dependent. Results as given
in table 2.2 are presented. The same set-up as for the work of this thesis (the
"TORUS" pipe rig) is used.

Table 2.2: M onochloramine Decay Constants and Flow Rate

Flow rate (I/s) Decay constant (h -1 )

0.07 0.051 ± 0.007
0.25 0.095 ± 0.008
0.9 0.173 ± 0.008

In addition, a relationship between monochloramine concentration found through

titration and chlorine concentration measured on-line is graphically depicted in

[160] .

41
2.7 Temperature

Fransolet et al. [54] establish a strong relationship between bacterial growth in

distribution systems and temperature. A number of data are presented. however.
without developing a mathematical model.
In Chambers [23], a WRc report following up [25], the long term variability
of the chlorine decay rate is investigated.
A temperature dependence of the form

is established (K =decay coefficient, 7 =temperature, .-L B=constants). However.

only few data could be provided to validate this equation and even within this
data rather large long-term changes, which do not fit the model, are observed
(thus only part of the data are used to validate the model).
A similar temperature dependence is used in Piccolo. The Piccolo Reference
Manual [127] includes a brief description of the Quality Module of Piccolo.
It is possible to use a chlorine decay coefficient K H 20 which depends on total
organic carbon (TOC) and temperature 7 according to

K H 20 = TOC x exp(14.4) x exp( -6050/7) (2.1.5)

(TOC in mg/l, 7 in K, !{H20 in min -1) and to use diameter dependent decay
constants within the distribution system.

2.8 Multi-Quality Models

1\1 ult i-qualitv models are concerned with distribution systems that mix water of
different quality.
The article by Reike et al. [121] is geared towards the use of poor quality
water (salty water) in irrigation systems in Israel. This is only possible if it is
sufficient ly diluted with high quali t~" fresh water. \Yater quality is considered in
terms of salinity.
Hierarchical two-level control is suggested to achieve sufficiently good water
for irrigation. The upper control level is concerned with finding an optimal steady
state distribution of flow rates, pressures and quality within the network. Parts
of this algorithm can be performed off-line before hand, as e.g. a path finding
algorithm based on graph theory. The flow rates between each source and each
consumer are chosen as the state variables of the system for the optimisation
problem.
The lower control level is concerned with maintaining the optimal values of
flow, pressure and quality. Special emphasis is given to a method of predictive
control which takes account of dead-times in the system.
In this paper no explicit quality model or values of any parameters are given.
However, an objective function (cost function) is provided. This work is based
on simulation, no experiments are recorded.
Ostfeld and Shamir [111] deal with the design of multi-quali tv distribution sys-
tem models in steady state. In addition, they give an outline of the optimisation
problem with a possible solution.
The optimal design is done concerning discharge (Which source and what is
its relative contribution of each source to total supply? Is there a source node
where supply reaches its bound?), quality (Is there a node at which a threshold
concentration is reached? To what degree is water being treated (removal ratio) at
treatment source nodes, and is there a treatment source node at which treatment
reaches its bound?), and head (Is there a node at which the head reaches its
bound? Are dummy valves being used within pumping stations?).
The paper of Coulbeck et at. [43] use a cost function incorporating pumping
and treatment costs (i.e. How much of each tank are you using?). Quadratic
programming is used to obtain optimal results. where costs in terms of money
are minimised. For inclusion in the optimisation, a monetary value was assigned
to water quality after treatment.

43
2.9 Algorithms and Simulation
Rossman and Boulos [125] compare different numerical methods for water qual-
ity modelling. They investigate the Eulerian finite-difference method (FD~I), the
Eulerian discrete volume element method (DV~I); the Lagrangian time-driven
method (TDM) and Lagrangian event-driven method (ED~I). It is concluded,
among other things, that TDM is the most versatile of the tested methods. Cur-
rently, DVM is used in EPANET. That method is discussed in some more detail
in Rossman et at. [126].
Boulos et at. [14] and Boulos et at. [15] introduce the event-driven algorithm
(EDM, see above) to model contaminant propagation in distribution systems.
The paper [15] is a more mathematical description of this algorithm with an
abstract example included, whereas [14] elaborates more on background. method,
and ideas without giving many equations.
The advantage of the event-driven method is that the discretisation of the
continuous model, which is always needed in computer applications, is now event-
oriented rather than uniform. Thus, the discretisation is now inherent to the
system and therefore exhibits better characteristics.
Chen and Brdys [16] use an approach employing bounds on variables for the
solution of a joint parameter and state estimation problem. Brdys et at. [29]
combine mixed integer programming and genetic algorithms to obtain an optimal
control schedule for a water distribution system, taking into account both the
hydraulic conditions (quantity) and quality in terms of propagation of substances
(normally free chlorine).
The EPANET User's guide [124] by Rossman includes an introduction to the
relevant theory. The most important equations are also given by Rossman in
[123]. In addition, the manual includes a short derivation of the relationship of
the decay constant with the bulk water and wall chlorine demands as given in
eq. (2.6).
GINAS [42] is a hydraulic simulator for water svsterns, A suite of software
IS available from \ rater Software Systems which includes network simulation.
operational scheduling, demand prediction, network model simplification and a
water quality simulator.
The description of WASMACS (Water Systems Monitoring and Control Soft-
ware) by R. S. Powell [119J contains an overview of the suite of software developed
by Brunel University. It includes among other things a graphical user interface,
a relational database, a quasi-dynamic and a static network simulator, a state
estimator using weighted least squares, the possibility to detect and locate leaks
and predict demands, a pump optimiser, and an optimised pressure controller.
Piccolo [127J is another hydraulic simualtion package. It contains also a Qual-
ity Module that covers chlorine kinetics. Recently, an extension of Piccolo for
bacteriological water quality, Piccobio, was presented [117J.
A paper by Cohen [39J describes QUALI, a sub program to DYNASIM. DYNA
is described in [38J. Both are developed in Fortran 77 on VMS.

2.10 System Identification and Physical Mod-

elling
Probably one of the most influential textbooks in system identification is
Ljung's "System Identification: Theory for the User" [89]. Other important text-
books in the field are Soderstrom and Stoica [135], Unbehauen [146] (in German),
or more recently Johansson [77], Ljung and Glad [90], and Walter and Pronzato

[155J.
All these books are written by authors belonging to, what could be termed,
the control community. Within this community system identification is differen-
tiated from 'mathematical modeling' [135] or 'physical modeling' [90]. Gold [55],
working on modelling in biological systems, argues along similar lines, when he
distinguishes between 'correlative' models and 'explanatory' models. To follow
Soderstrom and Stoic a [135, pA], 'mathematical modeling' is an analytical ap-
proach to find a model from basic laws of physics (or biology etc.), whereas system
identification is an experimental approach, where a model is fitted to recorded

-t5
Table 2.3: Comparison between physical modelling, system identification and
"hybrid" approaches.

name "Physical "Hybrid" System

Modelling" Methods, Identification
e.g. ..Empirical
Modelling"
based on physical insight both data
structure Tailor-Made "Grey-Box" Black-Box
interpretation of yes partially no
parameters

data by assigning suitable numerical values to its parameters.

The standard linear model structures that are used in system identification
contain generally an autoregressive and a moving average part of a suitable order
with an exogenous input (ARMAX models). In particular, differences in the
model of the influence of the noise on the output signal gives rise to variations of
that model, see [89, 155, 77].
In table 2.3 these "pure" approaches, system identification and physical mod-
elling, are compared with "hybrid" methods, e.g. "empirical modelling" as used
by Johansen and Foss [76] or semi-physical modelling as discussed by Lindskog
and Ljung [87]. Lindskog and Ljung write that "[by] semi-physical modeling we
will mean the process to take physical insight about the behavior of the system
into account, to use that insight to find adequate nonlinear transformations of
the raw measurements so that the new variables - the new inputs and outputs
- stand a better chance to describe the true system when they are subjected
to standard model structures (typically linear in the new variables)" (p. 1199 of
[87]). Similarly, Johansen and Foss [76] combine four linear models to one non-
linear model based on operating regimes. Thus, standard model structures are
combined based on physical insight. A similar approach is taken by \'elles et al.
[108] or Babuska and Verbruggen [5], where Fuzzy weighing functions are used for
the interpolation of local linear models. Other "hybrid" approaches of modelling

46
are possible.
Table 2.3 indicates the basis, type of structure and interpretability of pa-
rameters for physical modelling, system identification and "hybrid" approaches.
Whereas a physical model is based on physical insight with a structure that is
tailored to the particular problem, system identification models are based on
experimental data and a standard model structure (Black Box) is assumed. "Hy-
brid" approaches include some physical insight and some conclusions from data
and therefore the model structures are frequently termed "Grey-Box" models.
This reflects also on the physical meaningfulness of the model parameters. While
a physical model contains only meaningful parameters, it is generally impossible
to interpret the parameters of a Black-Box model as obtained through system
identification. Semi-physical or empirical modelling is again a mixture of the two
extremes, because some of the parameters may carry a physical interpretation,
but not all of them.
The most critical issue in system identification and "hybrid" approaches is
the determination of a nonlinear model structure. Several methods for selecting
between several possible structures have been published, e.g. Billings [9], Haber
and Unbehauen [64] or Unbehauen [147]. Standard nonlinear model structure
include Volterra series, Wiener and Hammerstein models.
,
Sjoberg et al. [134]
discuss feed-forward neural networks as possible nonlinear black-box models.

2.11 The Approach of This Thesis

In this thesis, the dynamics of suspended particles, the impact of biofilm on
particles and monochloramine as disinfectant in the syst~m are considered. Two
different types of published chlorine models are compared and a new particle
counts model is developed.
The chlorine decay models contain either a single decay coefficient [22, 24, 51,
123, 154] or a combined decay coefficient as proposed by Rossman [123, 124, 92]
(section 2.6.2.1). In particular, the variablity of decay coefficients as discussed in
section 2.6.2.3 is investigated in chapter 7 [67, 92, 123, 128. 160].
Chapter 6 of this thesis investigates particle counts (section 2.5.3) as an in-

47
dication of suspended biomass (section 2.5.1). The relationship between them is
discussed in [110, 98] and [65, 140, 159] elaborate on the use of particle counters
for similar purposes.
This thesis focuses on the contribution of biofilm to the suspended biomass
(after a flow increase). Therefore, different biofilm structures [151, 105, 138, 80]
(section 2.4.1) are discussed briefly and the effect of a disinfectant in the system
[85, 112] (section 2.4.4) is shown. The dynamics involved in biofilm detachment
are generally reduced to a linear time dependency in the literature [137, 115,
122, 28, 61, 6] (section 2.4.3). However, in the case considered in this thesis the
transient shear-off did have an important impact on the biofilm in the pipe rig
and therefore a new model is developed. This model allows a shedding profile
which is related to biofilm thickness to be obtained. The biofilm thickness is
difficult to measure [149, 150] (section 2.4.5).
The approach taken in this thesis to the development of the particle counter
model structure from data is novel. It was initiated by the author with simpler
experimental data in the work for his Master of Science [94, 95], but it was never
used for such a complicated system. To date, there is (to the author's knowledge)
no procedure available for finding a nonlinear system structure directly from data,
however, methods for selecting between several possible structures have been
published [9, 64, 147].
In the case discussed in [94, 95], some prior knowledge (i.e. a physically mean-
ingful parameterisation and the model for the limits of one of the parameters) was
available. This knowledge of a non-trivial model for a partially vanishing param-
eterisation implies knowledge of system structure. The full model incorporates
that knowledge together with specifications derived from the data.
For the particle counts model presented in this thesis there is less prior knowl-
edge available. Again specifications derived from data are developed. These
specifications have to be fulfilled if a model is to fit the data. Due to little prior
knowledge, assumptions about the system structure have to be made, which in
turn are investigated for their meaningfulness by numerical modelling and pa-
rameter identification. Results and the interpretation of them form, as in the

48
earlier work [94, 95], an important part of the model validation.

2.12 Conclusions

This review covered literature in the field of water quality modelling for water
distribution systems. The focus of this review lay on biofilms, suspended bacteria
and inanimate particles and various species of chlorine as disinfectant.
This chapter is not intended as an exhaustive review of the literature in the
field of water quality modelling, but it attempted to present a representative
selection of important publications. It gives an overview of the state of the art in
water quality modelling for distribution networks and of some relevant literature
not directly in that field.
Part II

Data

50
Chapter 3

Experimental Work

3.1 Experimental Set-Up

3.1.1 The Pipe Rig

Between January 1993 and January 1994 Thames Water Utilities Ltd. designed
and commissioned the "TORUS" pipe test distribution system at Kempton Park
Water Treatment Works. It was intended to mimic as closely as possible the
conditions found in a real distribution system.
The first large-scale pilot distribution system was built in 1986 by the In-
ternational Water Research Centre at Nancy, France (NAN.C.LE.) [41, 36]. It
consists of two sets of three pipe loops in series. The individual loops provide
mixed reactor conditions (as opposed to a once-through system). Both Lyonnaise
des Eaux [116] and the U.S. Environmental Protection Agency [34] built facilities
based on a similar layout (i.e. looped systems).
The "TORUS" pipe rig, however, is the only once-through large-scale pilot
facility. A schematic drawing of this pipe rig is provided in figure 3.1. The
pipe rig consists of a 1.3 km, 110 mm inside diameter pipe distribution system
built of approved construction materials. It is mainly made of medium density
polyethylene (MDPE), but contains about 1-2 % exposed or coated iron pipes.
It is divided into three sections of approximate length 500 m, 400 m and 400 m,
respectively, which are buried 0.5-1.0 m underground. After each buried section

51
 Inlet
SP 3 meters
Shed

Inlet (l)- Outlet

meters, +

Piperig length: ',3km

Figure 3.1: Schematic drawing of the "TORUS" pipe test rig at Kempton Park.
The rig has a total length of 1.3 km, most of it buried under ground apart from
the initial and final parts and three sections of 24 m in the portacabin. Four
meter locations at the inlet ('Inlet meters ') and sample points 1 ('SP 1 meters '),
2 ('SP 2 meters ') and 3 ('SP 3 meters ') are shown. This figure is reproduced with
permission after a drawing by A. Delanoue.

the water passes through a 24 m experimental testing station with facilities for
on-line monitoring, water sampling and for studying the performance of pipe
materials. In addition, there is a sampling station with on-line meters at the pipe
rig inlet. Measurements were taken at the inlet (i.e. at the balancing tank situated
there or shortly thereafter) and at the three sample points along the rig. Both
grab samples and on-line meters were employed. The design and construction of
the "TORUS" test distribution system are described in [69, 44].
During the time of the 1996 experiments, the rig was fed with water from
the Advanced Water Treatment Centre (AWTC) at Kempton Park. Thus the
treatment of the lowland surface water used as a source to the rig consisted
of pre-ozonation, coagulation and mixing, flocculation, rapid gravity filtration,
main ozonation, granular activated carbon filtration, slow sand filtration and
disinfection. Before the water enters the rig. bisulphite (S02) is used to reduce

52
the disinfectant residual to (generally) 0.25 mg/l of free chlorine. Subsequently
ammonia is added to transform, in this case, all free chlorine (Ch or HOCI) into
monochloramine (NHCh). Therefore, after a balancing tank situated just in front
of the pipe rig inlet, a total chlorine residual of about 0.35-0.4 mg/l is obtained.
The inlet chlorine concentration was changed to higher values for short periods
within the experiments.
In later experiments (in 1997) a different feed of water was used: the main
treatment works at Kempton Park (rather than the AWTC). The treatment pro-
cess is basically the same as that of the AWTC, however the mono chloramine dose
was less stable and generally varied around the higher value of about 0.5 mg/L
Measurements and instrumentation include the following.

• Heterotrophic plate counts

• Epifluorescence microscopy counts

• Particle counters

• Total chlorine meters and bench titrations

• Volumetric flow meter

• pH and temperature meters

They are described in some detail in the following section (section 3.2).
Results of experiments done with the "TORUS" pipe rig, which are only
indirectly related to this work, are published in [159, 160, 45], some initial results
of these experiments can be found in [98, 104].

3.1.2 The Pipe Rig as Input-Output System

If a "systems view" as introduced in section 1.4 is applied, the "Torus" pipe rig
is considered to be an input-output system, i.e. through complete knowledge of
the input-output behaviour the system is completely described. In the present
case this approach has to be taken cautiously since the output depends on past

53
 flow rate
--+
[CI] [CI] [CI] [CI]3
0--+ first --.l. second ---4 third -=+
section section section
pel pe2 pe3
peO--+ ~ --+- --+-
(particle counts)
bacto I bacto 2 bacto 3
bactoO--+ --+- --+- --+-
temp --+-

BDOC
--+
source
water --+-
quality

r
biofilm
l
r
biofilm 2
r
biofilm
3

Figure 3.2: The pipe rig as an input-output system

input (because of the biofilm) and since metabolic processes (growth, death, and
respiration) are to some extent independent of the input. This way of viewing
the pipe rig is illustrated in fig. 3.2.
The inputs of the investigated system are flow, monochloramine concentra-
tion at the pipe inlet, amount of bacteria entering the pipe rig (found through
culture assays), initial amount of biofilm present, amount of particles counted by
the particle counter at the pipe inlet, amount of nutrients entering the rig (of-
ten measured in terms of Biodegradable Dissolved Organic Carbon, BDOe) and
source water quality. The outputs of the system are monochloramine (or total
chlorine) concentration along the rig, particle counts, amounts of suspended bac-
teria (again found through culture assays), and amount of biofilm present. One
of the most important parameters which has to be taken into account is temper-
ature. Most of these input variables cannot be chosen, only the monochloramine
concentration and the flow are independent variables.

3.2 Measurement Method

In what follows, both grab samples and on-line meters which were used during
t he experiments, are described briefly. All meters were connected to a SC ADA
svst ern via --1- 20 mA cables. Data include readings in time intervals (sampling
times) of 5 min or 2 min from particle counters, chlorine meters, temperature me-
ters and pH meters. Grab samples of heterotrophic plate counts, epifluorescence
microscopy counts and chlorine titrations were taken at irregular intervals (the
majority during step flow trials).

3.2.1 Heterotrophic Plate Counts

Since water utilities generally use plate counts or epifluorescence microscopy

counts as indications of biofilm growth and detachment both these methods were
used in this study. Grab samples for heterotrophic plate counts (HPC) were taken
three times a week and at a higher frequency during step flow trials. Counts of
colony forming units (cfu) were performed on R 2A agar after 7 days of incuba-
tion at 22°C by the Thames Water labs. Standard procedures were applied when
taking samples, which include flaming the sample taps and using the standard
lab bottles with bisulphite tablets for the neutralisation of chlorine.
Note that HPC count only the number of bacteria or bacteria accumulations
capable of proliferation (e.g. forming a colony), not injured or dead bacteria. In
addition an accumulation ('clump') of bacteria will be counted as only one cfu.

3.2.2 Bottled Samples for Heterotrophic Plate Counts

During September and October 1996 fourteen sets of between 12 and 18 bottles
were taken at 10 time instants. Bottles were made either of plastic or glass. On
four occasions plastic and glass sets were taken simultaneously. Three replicate
bottles were evaluated immediately and after time intervals of 1 to 10 days (using
a different set of three bottles at each time).
Plastic bottles used were the standard Thames Water grab sample bottles,
which are, however, not tested for as long periods of time as were necessary here.
The glass bottles were burnt at 555°C and sealed by the Thames Water labs.
When taking the bottles, first the tap was flamed, then they were half filled,
rinsed, and subsequently filled completely. At the time of sending them off for
evaluation. bisulphite tablets were added to chlorinated water. Thus, before the

55
evaluation of bottles any disinfectant present was uninhibited in its action.
Water samples of unchlorinated water and water containing monochloramine
were taken at the inlet tank tap (during the corresponding disinfection regimes
in the rig). Water after disinfection with free chlorine was obtained from the tap
before ammonia dosing within the disinfection stage of the treatment process.
Disinfection is the last stage of water treatment. In it, the treated water is
brought into contact with a high free chlorine residual (in this case about 0.8-
0.9 mg/I after 5 min contact) for about 30 min. Afterwards the free chlorine
residual is reduced by dosing bisulphite (to about 0.25 rng/I). Ammonia dosing
(to turn free chlorine into mono chloramine) is the next (and last) step of this
stage. At the pipe rig inlet tank total chlorine residuals of about 0.4 mg/l were
measured.
For the duration of the bottle experiment the samples were kept in cardboard
boxes, inside the pipe rig portacabin. This ensured that the temperature of the
bottles was similar to the temperature in the pipe rig.

3.2.3 Epifuorescence Microscopy

Epifuorescence microscopy counts using acridine orange as staining agent were

performed on grab samples throughout the experiments and especially during
step flow trials. The method was adopted by S.M. McMath and A. Delanoue for
this particular application [103, 104]. This method makes it possible to count all
bacteria, even if injured or dead, and allows the investigation of bacteria sizes
and accumulations. This implies that if they occur in 'clumps' each bacterium
can be counted separately, which is impossible with heterotrophic plate counts.

3.2.4 Particle Counters

Particle Counters ("Versacount" from Hiac/Royco) were installed at four loca-

tions along the pipe rig, i.e. close to the start of the rig, named 'Inlet', and at the
before mentioned three sample locations, termed 'Sample Point (SP) 1, 2. and
3'. The inlet particle counter was not present at the start of the experiments (see

56
section 3.3).
Each of these counters extracts 60 mljmin of water from the pipe rig and
measures particles by leading this sample through a sensor. The sensor obtained
a reading for each particle proportional to the size of the shadow cast by it in a
laser beam. The number which is referred to as 'particle count' is the number
of particles which pass through the sensor divided by the volume of water that
passed through the counter averaged over one minute, thus, it is the number of
particles within one ml of sample water. Particle counts that are not divided by
the volume of sample are generally referred to as 'cumulative' particle counts (i.e.
particles per time unit). They are not used in this thesis.
The size of each particle is information additional to the (plain) particle count.
Particles were sorted in different bins according to the size of their shadow (di-
ameters between 2 uu: and 5 iuu, 5 iut: and 10 um, 10 uiu and 15 tut», 15 uu: and
20 uus, 20 uu: and 25 uiu, 25 uu: and 50 ust»; 50 uu: and 100 uiu, and 100 utt: and
150 /-Lm). The counters were calibrated in six monthly intervals with latex spheres
for recognition of these sizes. Therefore, particle counts can be distinguished ac-
cording to the size range they cover. The total counts of all particles > 2/-Lm are
referred to as 'total particle counts', any reference to counts within one bin (e.g.
counts of particles with diameter between 2 iuu and 5 /-Lm) are generally termed
difference counts. In this work size distributions of particle counts are also used.
The size distribution contains the amount of particles in each size range (bin)
divided by the total particle count. In other words, it contains the percentage of
all particles which fall into a certain size range (or bin).
The fact that the counters were calibrated with latex spheres introduces dis-
crepancies between the readings of different counters for not-spherical particles.
This issue is addressed in section 5.2.

3.2.5 Chlorine Meters

Chlorine meters C'l\Iedwa~'" from Portacel) were installed at the three sample
points and t he inlet of the pipe rig. They are capable of measuring either free

57
chlorine of total chlorine. All meters extract a flow of less than 200 ml/min.
The water sample passes through a cell containing two electrodes of dissimilar
metals, i.e. one of gold, the other of copper. This sets up an electrolytic action,
generating an electric current that is proportional to the concentration of free or
total chlorine in the water. Addition of Potassium Iodide to the normal buffer
solution, which is circulated in the cell in addition to the sample, changes the
obtained measurement from free to total chlorine. A range of 0 to 2 mg/l of free
or total chlorine was used during the experiments. The manufacturer's manual
gives the accuracy of the meters as ±0.05 mg/l at full scale deflection.
Chlorine meters have to be recalibrated, both in zero position and span, in
regular intervals, since they drift slowly out of calibration. If they are allowed to
drift too far from the actual value, the meter may leave the linear region of its
scale, which would introduce errors that are very difficult to correct. The problem
of meter drift is addressed in section 5.3.

3.2.6 Chlorine Titrations

Chlorine titrations allow readings for free chlorine, monochloramine and dichlo-
ramine to be obtained. The combination of all these is known as total chlorine.
Titrations are a bench analysis performed on a bottled sample of water. They
were done on site by members of the pipe rig team (including the author of
this thesis). The method employed during this work uses diethyl-p-phenylene
diamine (DPD) as an indicator to produce pink/red colour in a sample prepared
with an appropriate buffer and Potassium Iodide (for monochloramine and dichlo-
ramine). This solution is titrated against Ferrous Ammonium Sulphate. Some
samples were double-checked with the Dr. Lange analysis system, which is based

on a spectrophotometric method.

3.2.7 Bottled Samples for Chlorine Titrations

During September to ='\ovember 1996 eight sets of between 3 and 10 bottles were
taken at eight time instants. All bottles were Blade of glass except for those

58
used for the immediate titrations. Five of the eight bottles for the immediate
titrations were made of plastic, the others were glass bottles. Three or more
replicate titrations were done per bottle. They were done immediately and after
time intervals of 0.5 hours to 7 days (using a different bottle at each time).
The glass bottles were cleaned by filling them with nitric acid for at least
twelve hours. Subsequently they were rinsed with sample water. Samples of wa-
ter containing monochloramine were taken at the inlet tank tap (at the time when
monochloramine was in the rig). Water after disinfection with free chlorine was
obtained from the tap before ammonia dosing within the last stage of the treat-
ment process. The bottles were kept, like the bacteriological ones. in cardbord
boxes in the pipe rig portacabin.

3.2.8 Volumetric Flow Meter

An electromagnetic flow meter (Krohne) at the inlet (i.e. before all meter sta-
tions) obtained volumetric flow readings in litre/second by measuring the voltage
induced by the water flow in a magnetic field.

3.2.9 Temperature, pH and RedOx Potential

Temperature, pH and Reduction-Oxidation Potential, 'ReDox', were all mea-

sured with the same meter ("Meredian" from Leeds & Northrup, with three
probes). Meters were installed at the three sample points along the rig and ex-
tract about 200 ml/min. For the present investigation RedOx was disregarded,
and pH was only monitored (the pH value is generally stable at around 8) but not
taken further into account. The pH measurement is temperature compensated.
Temperature is measured by an electronic temperature sensor. Only temperature
was considered in the further analysis.

59
 1.5 1

H..OW

!
0.5 ~
I

r t
o
"'"
~
-
"

et REGIME I REG.2 REG.3

E

-
~0.5

~ -0.5
:Eu
CHLORINE

o I
03-lul 22-Aug II-Oct 30-Nov
TIME

Figure 3.3: Experimental Schedule showing flow and total chlorine at the pipe
rig inlet, three baseline regimes of Experiment 1 are indicated

3.3 Experiments

3.3.1 Pipe Rig Experiments During 1996

The 1996 experiments took about five months of preparation and seven months of
,
actual data gathering. They consisted of 12 sharp flow increases lasting 12 hours
to three days and subsequent decreases to previous flow values (termed 'step flow
trial', as opposed to changes in baseline flow) and four step chlorine trials. The
experimental schedule is depicted in fig. 3.3.
The experiments can be divided into two parts, Experiment 1 consisting of
the step flow trials 1 to 9 and Experiment 2 consisting of the remaining three
step flow trials and four step chlorine trials. Experiment 1 is aimed at obtaining
some information about the effect of biofilm on water quality, Experiment 2 was
designed to result in some information about changes in chlorine decay due to
changes in flow or inlet chlorine concentration.
Three main baseline regimes in terms of flow and disinfectant regime can be
distinguished in Experiment 1 (as marked in fig. 3.3).

regime 1: laminar baseline flow (0.08 lis. 0.0084 tn]«. Reynolds number 920).

60
 no disinfectant residual in the rig
regime 2: laminar baseline flow (0.08 lis, 0.0084 mis, Reynolds number 920),
disinfection of about 0.35-0.4 mg/I total chlorine at the inlet of the rig
regime 3: turbulent baseline flow (0.3 lis, 0.032 ta]«, Reynolds number 3500),
disinfection of about 0.35-0.4 mgll total chlorine at the inlet of the rig

Within each of the considered regimes, at least two step flow trials took place
to provide an indirect measurement of the biofilm present. Evidence for increased
biofilm detachment due to a flow increase within the "TORUS" pipe rig has
been found previously [159J. The higher flow rates during step flow trials were
0.6 lis (0.063 mis, Reynolds number 6900), 0.9 lis (0.095 tn]», Reynolds number
10400) and 0.45 lis (0.047 mis, Reynolds number 5200). Since the transition
from laminar to turbulent flow occurs generally between Reynolds numbers of
2000 and 4000 [102], all high flow rates are turbulent. For a graph of Reynolds
numbers versus flow rate see Appendix D.
In particular, Experiment 1 aims at identifying the relationships between the
amount of biofilm present and flow rate or total chlorine concentration. The first
part of this experiment aims to establish the relationship between biofilm accu-
mulation and flow rate, without any chlorine present (i.e. only one independent
variable). In the second part it was anticipated to establish a relationship between
biofilm detachment and chlorine concentration for two baseline flow rates.
Throughout the different regimes of this experiment several step flow trials
took place. The idea behind these trials is that the increased shear off from the
biofilm might give some indication about how much biofilm is present without
removing all of it (due to the short duration). The baseline flow period between
step flow trials may help to establish if conditions are still similar.
Experiment 2 is concerned with the relationship between chlorine concentra-
tion within the pipe and initial chlorine concentration or flow rate. To obtain a
maximal amount of chlorine decay data, both of spatial and temporal decay: the
transient response to step input changes (step in inlet total chlorine or step in
flow rate) is considered and not only the (short-term) steady state. Furthermore,

61
for each pair of independent variables a step flow trial with incorporated step
chlorine trial of roughly 2-3 days duration was made.
The time scale in Experiment 2 is much shorter compared to the first part
of the experiments, thus it was anticipated that each step trial is long enough
to establish changes in the mono chloramine decay rate. This short duration of
these step flow trials gives the possibility to assume that the effect of biofilm is
the same throughout the step flow trial and the usage of a reference flow (and
chlorine concentration) in between each step trial allows to calculate possible
long-term changes in chlorine decay. These long-term changes are to date merely
side effects of this experiment, however, they may be considered in future work.

3.3.2 Bottle Experiments During 1996

During September and October 1996 fourteen sets of between 12 and 18 bottles
for bacteriological analysis were taken at 10 time instants as already described in
section 3.2.2, which focuses on the measurement method. Three replicate bottles
were evaluated immediately and after time intervals of 1 to 10 days (using a
different set of three bottles at each time).
The schedule for this experiment is given in table 3.1. There the date and
time when each set was taken, the material of the bottles, the type of disinfection
present in the bottles, the number of bottles in a set and the time until they were
send for evaluation are given. The material of the first three bottles of all sets
was plastic, since these samples were within the limits of the standard procedures
which use bottles made of that material.
During the period of September, October and November 1996 eight sets of
between 3 and 10 chlorine bottles were taken at eight time instants as already
described in section 3.2.7 (again chiefly dealing with the measurement method).
Three or more replicate titrations were done per bottle. They were done imme-
diately and after time intervals of 0.5 hours to 7 days (using one new bottle for
all replicas each time).
The schedule for this experiment is given in table 3.2. There the date and

62
Table 3.1: Bottles taken for bacteriological analysis (heterotrophic plate counts)
as part of the bottles experiment of September/October 1996

Set Taken Material Disinfection Number of Evaluation

number at regime bottles after [days]
1 3-9-96 14:15 plastic no chlorine 15 0,L2,3,7
2 4-9-96 16:00 plastic no chlorine 15 0,1,2,4,7
3 4-9-96 16:00 glass no chlorine 12 0,1,2,7
4 5-9-96 15:00 plastic no chlorine 15 0,1,2,4,7
5 5-9-96 15:00 glass no chlorine 14 0,1,2,4,7
6 12-9-96 14:45 plastic monochloramine 15 0,1,3,4,7
7 12-9-96 14:45 glass monochloramine 14 0,1,3,4,7
8 13-9-96 13:00 plastic monochloramine 15 0,L3,4,7
9 13-9-96 13:00 glass monochloramine 14 0,1,3,4,7
10 17-9-96 15:00 glass monochloramine 18 0,1,2,3,4,7
11 23-9-96 16:00 glass free chlorine 15 0,1,2,3,7
12 24-9-96 14:00 glass free chlorine 14 0,1,2,5,7
13 30-9-96 glass free chlorine 18 0,1,3,7,10
14 4-10-96 glass free chlorine 11 0,3,4,7

time, disinfection regime and number of bottles in the respective set are given.
In addition the time in days that elapsed until the last bottle was titrated is
provided. Intermediate bottles were evaluated at various times, roughly after
o hrs, 0.5 hrs, 1 hr, 4 hrs, 6 hrs, 24 hrs, 30 hrs, 48 hrs, 72 hrs, 5-7 days. One or
more of these times were generally omitted to allow for long time investigation
(i.e more than 3 days).

3.4 Problems Encountered

There are a number of problems which did occur during the data collection period
of 1996. Some are given in the following.

• meter drift (in particular chlorine meters)

• limits of accuracy of meters

• equipment failures (e.g. one of the circuit boards on the inlet chlorine meter)
and deliverv delays (inlet particle counter)

63
Table 3.2: Bottles taken for chlorine titrations as part of the bottles experiment
of September/November 1996

Set Taken Disinfection );umber of Last eval uation

number at regime bottles [days]
1 12-9-96 15:30 monochloramine 9 5
2 13-9-96 1.5:20 monochloramine 9 6
3 17-9-96 15:50 monochloramine 9 7
4 18-9-96 16:00 monochloramine ,
,....
7
5 25-9-96 13:00 free chlorine 9 12
6 2-10-96 17:50 free chlorine 3 1
7 4-11-96 11:20 free chlorine 10 'J

8 5-11-96 11:40 free chlorine 10 6

• source water quality changes and treatment plant failures

• particle numbers entering the rig vary significantly

• temperature is uncontrolled and will have some effect on bacteria

• time scale of some effects could be several months

• no meter to measure the amount of biofilm in the rig is available

• heterogeneity of biofilm may cause effects of biofilm to change over time

For some of these problems remedies can be found, e.g. frequent titrations are
used to check on chlorine meters and thus adjust for meter drifts. Other possible
problems, like the influence of temperature and source water quality have to be
taken into account in the model. The heterogeneity of the biofilm and its possible

effects are neglected.

3.5 Summary
This chapter introduced the experimental set-up used for the work of this thesis
and gave an outline of the experiments performed to obtain the data.
The "TaReS" pipe rig provided by Thames \rater Utilities Ltd. was used
as set-up. It is a large scale once-through experimental distribution system. A

6--1
number of meters and grab samples were employed for these experiments, includ-
ing a flow meter, particle counters, total chlorine meters, titrations, heterotrophic
plate counts and epifluorescence microscopy.
The main element of almost all pipe rig experiments is the step flow trial
(SFT). Chiefly two sets of experiments were performed in 1996, the first aiming
mainly at biofilm and particle counts, the second at total chlorine. Bottle ex-
periments, for bacteriological analysis and for evaluation of chlorine, were also
performed using the same water as in the pipe rig. Schedules of all experiments
are provided in this chapter.
The following chapters give an overview of the collected data and an in-depth
analysis of them.

65
Chapter 4

Qualitative Characterisation of
the Data

The first step towards obtaining a model from the experiment results is to
clarify what can be learnt from the data. In the experiment described in this
work heterotrophic plate counts (HPC) and epifluorescence microscopy counts
are considered, which are frequently used within the water industry as indications
of biofilm growth and detachment. In addition particle counters were used for
the same purpose. Furthermore, monochloramine, which is used by various water
utilities to control bacterial populations, is investigated. For a description of the
measurement methods used to obtain readings of these variables see section 3.2.
Since the goal of this work is to obtain information about the behaviour of
the investigated aspects of water quality, amongst the total duration of the ex-
periments of seven months, the focus of attention lies on periods of input variable
(i.e. flow) changes. These were known as the step flow trials (SFTs).
Together with examples, a characterisation of the data will be given in this
chapter. In addition, hypotheses for the underlying phenomena or relationships
with other aspects of water quality are included. Modelling will be used in the
following chapters to decide on the best hypothesis. The aim of this chapter is to
set out the basis of the models developed or investigated in part III of this thesis.
Particle counts and monochloramine decay will be the subject of the subse-
quent chapters 6 and 'i. respectively. Only the bacteriological data (including the
HPC bottles) feature little in further chapters of this thesis, since the inaccura-

66
des arising from this type of analysis is too great for analytical modelling them.
Therefore, some emphasis is put on these results in this chapter. In particular.
section 4.5.3 summarises the results obtained from heterotrophic plate counts and
epifluorescence microscopy data. Due to the relationships of particle counts with
bacteriological results (cf. section 4.5.3), the models of particle counts are an
indication of processes involving bacteria.

4.1 Total Particle Counts

A full description of the particle counters used for these experiments is provided in
section 3.2.4 on page 56. Briefly, these counters measure the number of particles
per volume of water (i.e. per ml) and the size of the respective particles. This
section is concerned with the total particle counts, which are the number of
all measured particles per ml (i.e. all particles with diameter> 2/-lm). The
observations of this section are, however, in principle equally valid for counts
of larger particles (e.g. > Sus», > 10/-lm etc.) or even difference counts. Size
distributions of particles are the object of section 4.2.
Two typical examples of total particle behaviour during a SFT are given in
figs. 4.1 and 4.2. The first figure depicts the total particle counts during step flow
trial 5 which took place on 28 August 1996, the second figure depicts the same
data during SFT 8 of 24 October 1996. Volumetric flow rate, measured at the
inlet, and total particle counts for three sample points along the rig (SP 1, SP 2,
SP 3) are given on the same y-axis on a logarithmic scale. In fig. 4.2 inlet particle
data is also included. Note, for comparison, that fig. 4.1 is the same SFT as the
example of heterotrophic plate counts given in fig. 4.9.
Since the observations of this section are based on all data, additional relevant
portions of the total particle counts data are provided in appendix A.l. There
the particle counts of all twelve step flow trials of the 1996 experiment can be
found (figs. A.l to A.12 on pages 233-244). Within this appendix only one sample
point is given per graph, which is a different way of presentation compared to the
examples provided in this chapter.

67
 FLOW

12:00 29-08-96 12:00 30-08-96

TIME

Figure 4.1: Step flow trial 5 as first example of total particle counts. Depicted
are three sample points along the rig (SP 1, SP 2, SP 3) and inlet flow.

4.1.1 Observations

From the examples given in figs. 4.1 and 4.2 some observations which are typical
for the complete data can be made.

• The particles move through the rig, i.e, a behaviour of particles similar
to the one observed upstream is measured downstream after a time inter-
val. This time interval agrees approximately with the predicted travel time
between the respective points (cf. section 5.4).

• Sharp peaks occur with flow increase. These peaks propagate almost intact
through the rig. They occur only within less than about one travel time to
the respective sample point.

• A pattern of behaviour which occurs simultaneously throughout the rig is

observed: particle counts increase slowly at high flow, reaching a maximum
after about one full travel time to that particular sample point and decline
subsequently. The counter signal settles after a short period at high flow

68
 ~
~ 103
~
<
8 102
........
.
til
~
u
'E
~ 101 ~'"-"""~
en
~ ::
S 100 :.: · · ·: ~?W .
...... ......................................: : : : :: : : : : :::::: ::: : : : : : : : : : : : : : : : : : : . : .
~
.

tJ.1 .. ..... -_ :::::::: : ::: : :: :: :: : :: : :: :::: ::::: :: :::::::: ::: :::: .

-l
U ~--_ -----------~

f=:
~i2=~'--0--96---------'----_---l.-_--_ ---l
12:00 25- 10-96 12:00
TIME

Figure 4.2: Step flo w trial 8 as second example of t ota l particle counts. Depicted
are inlet particle counts, counts at all sample points (SP 1, SP 2 and SP 3), and
flo w measured at th e pipe rig inlet.

to a value t hat is generally high er t han t he low flow (baseline) value. This
will be referred to as t he 'slow signal '.

The decline is generally st rict ly monotonic, however , if all t he relevant data

provided in t his work (figs. A.l to A.12 on pages '233- 244) is considered,
notable exceptions emerge. In par ti cular, "humps" sometimes occur wit hin
t he decline, e.g. in all sa mple points at 8FT 3 (fig. A.3 on page 235) and
only sample point 3 of 8FT 8 (fig. 4.2 or fig. A.8). Also, as t he experiment
progresses t his signa l becomes less prominent. E.g. in 8FTs 8 and 9 (figs. 4.2
and A.9) the declin e almost degenerates into a straight line.

• Occasionally parti cles increase at sample point 1 with flow and stay at t he
higher value until flow decrease (e.g. in 8FT 8, fig. 4.2).

• On a few occasions a sharp peak occurs at sa mple point 1 wit h flow decrease
(e.g. in 8FT 5, fig. 4.1).

• Afte r step down in fl ow t he total particle counts decrease slowly. They ettle

69
 down to a potentially different value than before the 8FT within about one
(low flow) travel time.

• The inlet particle counter data as given in fig. 4.2 exhibits a large number
of apparently random peaks throughout the high flow period. These peaks
do not propagate through the pipe rig. A decaying signal is additively
underlying the random peaks.

4.1.2 Discussion and Hypotheses

As indicated by the bullet points above, the particle counter data is split into
several parts, which might be quite independent of each other. This hypothesis
is expanded further in this section, and, if appropriate and possible, underlying
explanations are put forward. Based on the observations established in the pre-
vious section, three different particle counter signals originating from the step
increase in flow will be considered: peaks (mainly at increase of flow), a step in
particles and a slow signal.
From the point of view of usefulness of the data for the construction of a
model, signals that have only 'face value' have to be .distinguished from those
with observable underlying effects. Only the latter are suitable as a starting
point for a model. Both types of signals occur in the data described above.

4.1.2.1 l?ecUks

Peaks occur either directly with increase of flow or (but only in section 1) with
flow decrease. They occur only at well defined points along the rig which are
identical to the sample points (and the inlet) or, considering the limited accuracy
of the data, in close proximity to them. The accuracy of data is limited by the
sampling time, which was initially 5 min (that corresponds, at a flow rate of
0.6 lis, to a distance of about 19 m in a 110 mm diameter pipe, at 0.9 lis to
28.4 m) and later 2 min (equivalent to 7.6 m at 0.61/s and 11.4 m at 0.9 lis).
These observations indicate that the peaks originate from valves or bends
which are situated close to the sampling points. There are two main valves which

70
have to be opened or closed for a flow increase or decrease: on the one hand there
is the pipe rig inlet valve, which marks the beginning of the rig, and on the other
hand there is a gate valve for fine adjustments just in front of sample point 1.
Peaks sometimes occur at sample points 2 and 3, in particular only in SFTs 1
and 7. The fact that there are, however, always initial peaks originating from
close to the inlet and sample point 1 is very strong evidence for the hypothesis
that the valves are the origin of these particles. Very seldom there is a peak at
step down, but again when that does occur it is only observed at sample point 1.
Thus, this leades to the hypothesis: closing the valve near sample point 1 may
sometimes lead to a release of particles.
The narrow width of the peaks indicates that the release of particles is not
only narrowly localised but also that it happens quickly. If particles were released
over a longer period of time, broader peaks would be observed, the same would
be true if the point of release covered a large length of pipe. The short release
period indicates that these particle do not have a strong adherence to the pipe
wall. All material which has the potential of becoming entrained is lifted into
suspension (almost) at once. This behaviour is different from the supposed phe-
nomena underlying the slow part of the particle counter signal as discussed in
section 4.1.2.3. Thus, it is proposed that these particles are loose deposits which
are collected mainly in the valves.
The hypothesis, that in the particles in peaks are loose deposits entrained into
suspension form localised 'hot spots' due to a sudden flow increase, was already
put forward in [159].
Within this work, peaks are considered to be 'face value' signals. Not because
no relevant underlying effects are expected, but because the exact shape of the
peaks is unobservable since the sampling time is too large for a detailed picture
of their structure. However, because the main interest of this work is in the
biofilm coating of the distribution system and it is assumed that the origin of
these particles is unrelated to it, there is no need to gain a deeper understanding
of the behaviour of the particles in peaks.
Finally, since the peaks propagate with about the theoretical travel time

71
through the pipe rig, they can be used to improve the estimate of the travel
time, cf. section 5.4.

4.1.2.2 Occasional Particle Step

A second 'face-value' signal is the occasional particle step as described in sec-

tion 4.1.1. It occurs only in some SFTs and, similar to some of the peaks de-
scribed in section 4.1.2.1, this signal originates always from sample point 1. In
particular, its origin is not further than 12 m upstream of particle counter 1,
since it was observed at a flow of 0.9 lis only. This indicates that the origin of
this signal is again the valve situated in front of sample point 1. The shape of
this signal suggests a near constant release of particles into suspension, which are
propagated through the rig. A certain size of the opening of the valve could cause
this effect, for instance by inducing eddies which lift an almost constant amount
of loose particles into suspension whilst the flow is high.

4.1.2.3 Slow Signal

The most interesting part of the particle counter response to a step increase of
flow rate, from the modelling point of view, is the slow signal as described in
section 4.1.1. Although the maximum values of this part of the particle counter
response are reached after the travel times to the respective sample points, this
signal is not merely propagating through the rig: at all sample point a simulta-
neous increase of counts that starts with the increase of flow is seen. The fact
that the counts are first increasing and then decreasing rules out the hypothesis
of a mere artefact produced in the counters due to the variation in flow, since
this would result in fairly constant higher counts for a higher flow in the pipe
rig. Thus, this is a generic effect occurring over the whole length of the pipe rig
simultaneously.
Possible explanations of the underlying processes include an increased number
of patches of biofilm (e.g. a biofilm consisting mainly of streamers) which are
sheared off the pipe walls with a certain probability at high flow or repeated,
random sloughing (a continued supply of particles) due to a thick, rich layer of

72
biofilm. Simulation results presented in section 6.3.1 on page 127 imply that
the latter hypothesis better matches the observed data. The hypothesis of the
particle counter response being due to a continuous shear-off which is decreasing
in size was mentioned earlier by the author et al. in [98].
Since this signal is slow, the particles considered here adhere better to the
wall than those particles that form the peaks, for the material is not coming off
the wall at once. Therefore, it is conclude that these particles were attached
particles.
The concept of a local shear-off density (with units p~~les per metre) will be
introduced in section 6.4.1.3. This concept allows the development of an equation
(i.e. a parametric model) that describes the slow signal. It follows the hypothesis
of repeated shear-off given above. Again, the magnitude of the shear-off has to
decrease over time.
If assumption 1 of section 6.4.1.3 is followed, then the shape of the decline of
the slow signal reflects the shape of the local shear-off function. If, in addition,
assumption 2 of section 6.4.1.3 is considered to be valid, then a major influence
on the increase of the slow signal is the biofilm shedding profile, which is related
to the biofilm thickness.

4.1.2.4 Inlet Particle Counter Signal

The inlet particle counter exhibits a large number of apparently random peaks
or spikes during high flow. These peaks do not appear after the flow is decreased
and do not propagate through the pipe rig (i.e. they do not appear at the other
sample points). In particular the latter argument gives rise to the conclusion that
these are artefacts, in this case generated by, or in the line to, the particle counter.
It therefore does not reproduce the conditions in the pipe rig. The inlet particle
counter is the same type of counter as the others, however, it was installed only
during the experiments of 1996 and might be slightly different because of its later
manufacturing date. In addition, it does not use a pump to maintain the flow at
a constant level but an overflow mechanism. However, this does not imply that
the flow through the counter changed during high flow, since the counter flow was

73
monitored and found to be within a few ml/min of the target flow of 60 ml/rnin.
In summary, it is assumed that the spikes or peaks in the inlet particle counter
signal are artefacts but the reason for their appearance is not known.
Underlying these spikes is a decaying function. Since the inlet particle counter
is very close to the inlet (within 5 m downstream of the inlet valve) this signal
corresponds reasonably well to the hypothesis of a local shear-off function as
discussed above.

4.1.2.5 Particle Counts after Decrease of Flow

Particle counts decrease at the end of a step flow trial. An immediate end of the
additional shear-off introduced by the increase of flow is proposed to explain this
observation. Following that assumption the particle counts should decrease within
one travel time. This coincides with the data. The counts do not necessarily settle
down to the same level as before the step flow trial, however, a definite decrease
occurs. Given the low accuracy of particle counts in terms of absolute numbers,
the trend towards lower numbers is considered more conclusive than the actual
final value.

4.2 Particle Counter Size Distributions

In this section, the size distributions of particle counts as introduced in sec-
tion 3.2.4 on page 56 are investigated. Briefly, a size distribution contains the
percentages of total particles which fall into the available size ranges.
Fig. 4.3 depicts, on the left y-axis, the percentages of all counted particles
which fall within the size ranges 2-5 uu: diameter, 5-10 tut: and 10-15 uu: (for
the same step flow trial as the epifluorescence data in fig. 4.14). Only sample
point 1 is depicted. Inlet flow is given on the right y-axis.
The size distributions of sample points 2 and 3 are similar to each other, and
the shape of the size distribution of the particle counter at the inlet (downstream
of the tank) is also similar but here the signal amplitudes are reduced to very
small values.

74
 The size distributions of all sample points and the inlet and of all step flow
trials of 1996 are given in appendix :-\..2.

12
~
0
~ 80

"
~

-"
..J
U
Eo-<

<t:
60

0..
~
40 6,.--
:/l
0 5-10 urn ~
~
0 ~
<t: 40
Eo-<
Z ~
~
U
0
"
~
0..
FLOW

-31 0-Oct 12:00 16-0ct 12:00

0
TIME

Figure 4.3: Particle Counter Size Distribution of sample point 1 (as percent-
ages), particle diameters are within 2-5 urn, 5-10 urn, and 10-15 urn; Inlet flow
values are given on the right y-axis.

4.2.1 Observations

It is useful to distinguish between small particles (2-5 pm) and large particles (5-
10 usu, 10-15 iut: etc.), since these groups of size ranges exhibit a very different
behaviour. From fig. 4.3 several patterns of behaviour can be observed.

• There are peaks within the initial period of the step flow trial. They are
negative for small particles and positive for large particles. A comparison
with total particle counts shows that the peaks coincide .

• \\'ithin the initial period of the 8FT something like a dip in small particles
and a hump in large particles takes place. This pattern does not alwavs
contain a clear minimunr/maximum value; it consists. during some 8FTs
(e.g. 8FTs 5 and 9). purely of an overall decrease/increase. It finishes
shortly after one travel time.

( .J
 • If a step in particles occurs (cf. section 4.1.2.2), the percentage of large par-
ticles is significantly higher. This can be seen particularly well in 8FT 8,
when the propagation of particles released at sample point 1 by this phe-
nomenon during high flow is observed in sample points 2 and 3 after decrease
of flow.

• At a little more than one travel time past the start of the 8FT the per-
centages settle to a fairly constant level. For 8FTs to 0.9 lis, this level is
different from the values before the flow increase: the percentages of large
particles are larger, the percentage of small particles is smaller than at low
flow. Thus, the relative amount of large particles increases significantly
during high flow. For 8FTs to 0.6 lis (8FT 1, 2 and 3) the percentage
levels settle down to similar values as before the flow increase.

• After the decrease of flow, the percentages return slowly to values that are
close to those observed before the 8FT. This occurs within about one (low
flow) travel time.

4.2.2 Discussion and Hypotheses

The particle counter size distributions (fig. 4.3) show a similar phenomenon as
the epifluorescence counts given in fig. 4.14: there increased numbers of particles
('clumps') with flow increase are observed, in the size distributions an increase of
the percentage of large particles within the total number of particles is seen.
This could be due to a different size distribution of attached as opposed to
suspended bacteria. The attached bacteria are predominately larger. With a
step increase of flow, attached bacteria are sheared-off from the biofilm and,
thus, change the size distribution of suspended particles. This hypothesis is
investigated in some more detail in section 6.3.2.

76
 Chlorine Decay Along the Rig, 15.10.96
':" " 0
..

. .
-:'" . .
. "

0.5

::::
en
oS 0.4
(ij
::J
~0.3
Q)
a:
Q)

.§ 0.2 .. .....
o
z '. " .
o . .. ..: ' " "., . ..
19:11
~ 0.1 .. :
o ' "

16:48
I- . . . ".-: .
14:24
o 12:00
o 09:35
2 07:11
04:48 Time at Inlet
4
Travel Time [h]

Figure 4.4: Total chlorin e vs. travel time an d time at the in let, 15.1 0.1996 (step
flow tri al 7). Th e asterisks t»') are titrati on data (see text) .

4.3 Disinfection

During t he expe rimental campaign of 1996, aft er fou r mo nt hs of no chlorine,

monochloramine was used as disinfectant in t he pipe rig . In t his section some
exam ples of chlorine data, ob t ained from t he pipe rig and from bottled samples,
are provid ed. A qu an ti t a ti ve treatme nt of t hese dat a can be found in section 5.3
and chapter 7.

4.3.1 Observations in the Pipe Rig

F igs. 4.4 and 4.5 show plots of t otal chlorine versus both t ime and t ravel t ime .
T he da t a for step flow t rial 7 and t he step chlor ine t ria l are dep ict ed . The t rave l
tim e is t he t ime it will take for a small volume of wat er (an aliquot ) to reach a
respect ive sam ple tap from t he inlet shed whereas t he t ime at Inl et is t he actual
clock t ime when a considered aliquot of water started at t he inl et .

77
 Chlorine Decay Along the Rig, 26.11.-29.11.96
.:" " ..

. ' " .
.. ...

. .

. . " . " .. " . .

0.9
......
~0.8
Cl

.s0.7
Cii - .. :- . '"

~
:Q 0.6
III ... ..
(])

a:
(])
0.5
c
.~ 0.4
:c
U
_ 0.3 29-11-9
.so 12:00
f- 0.2

0.1
o

50 Time at Inlet

Travel Time [h]

Figure 4.5: Total chlorin e vs . travel time and time at the in let, 26.11-29.11.1996
(step chlorine trial 1). Th e asterisks ('* ') are titration data (see text) .

On the figures , to t al chlorine is given on the y-axis (vert ical), the travel time
is given on the x-axis (horizontal) and the inlet clock time for the start of this
aliquot is depicted on the z-axis (into the plane of the paper). The sample taps
are represented by the t ravel time . The travel t imes are calculated for t he t hree
sample points along the rig and t he plots along the x-axis are formed by connect-
ing these calculated values t hrough straight lines. Three lines along t he z-axis
(into the plane of the paper ) show when the sample points 1, 2 and 3 were reached
by t he aliquots that started at t he time given on t he z-axis .
It is seen from t hese graphs , how total chlorine decreases over t he length of
the rig by following t he lines of data along t he x-axis (t he t ravel time-axis). Each
such line shows how t he chlorine residual in one aliquot of water decreased over
time and distance wh ile t ravelling t hrough t he pipe rig. Going along the z-axis
it is seen in fig. 4.4 t hat t he time for t he water aliquots suddenly become much
larger t owards t he end of t he hown por ti on of data. T he reason for t hi i t hat

78
the SFT also comes to an end towards that point and thus the aliquots which
are shown travel only part of there way at high flow, which increases the overall
travel time.

Note that there is almost certainly a problem with the meter at sample point 2
in fig. 4.4. It shows a higher chlorine residual as the meter at sample point 1.
This is quite obviously impossible, since it would imply that in some aliquots of
water the chlorine residual while travelling down the pipe rig. This problem is
addressed in section 5.3.
In fig. 4.5 the travel time does not change over the span of the depicted data,
since this graphs shows a pure step chlorine trial with no change of flow. It is
seen that the chlorine residual increases at the beginning of the data set, forming
something like a 'hill' in the third dimension. However, the decay of chlorine
along the rig (along the travel time axis) is still observed.
On both figures, figs. 4.4 and 4.5, mainly meter data is given. Due to problems
with meters (as discussed in section 5.3) titrations are more accurate. They are
also included in these graphs, as single stars or asterisks ('*'), however, it is
difficult to discern the titration data within these graphs. Figs. 5.3 and 5.4 on
page 104 and page 105 show the difference between meter data and titration.

4.3.2 Observations in Bottled Samples

In figs. 4.6 and 4.7 results of total chlorine titrations of bottled samples are
presented. Three or more replicate titrations were done per sampling instant.
These replicates are shown as crosses '+', their average as connected squares. For
the detail on the measurement method see sections 3.2.7 and 3.3.2.
In fig. 4.6, four sets of bottles with free chlorine as disinfectant are depicted.
The four sets shown in fig. 4.7 show bottles with monochloramine. Since both
figures have the same axis ranges, visual comparison of the two graphs shows that
free chlorine decays faster than monochloramine, as expected. Decay coefficients
for these data will be calculated in section 7.5 on page 195. The aim of this part
of the experiments is to compare the decay coefficients obtained from bottled

79
 0.45

0.4
.........
0iJ0.35
E
........

-
~ 0.3
~
00.25
~
u 0.2
~
5°·15
Eo-<
0.1 •
•
0.05 . . -. .
•
~

° 2 468 10 12
TIME [days]

Figure 4.6: Bottle samples with free chlorine: titrations of total chlorine of glass
bottles. Lines with squares show average titration results, results of replicas are
given as unconnected crosses '+'

.........
~0.35
E
........

-
~ 0.3

~0.25
~
U 0.2
•
~
5°·15
Eo-<
0.1 •

0.05
-
')
4 6 8 10 12
° TIME [days]

Figure 4.7: Bottle samples with monochloramine: titrations of total chlori ne of

glass bottles. Lines with squares show average titration results. results of replicas
arc given as unconnected crosses '+'

80
samples with those obtained in the pipe rig. The difference of the two is expected
to be chiefly due to the biofilm in the rig. This question is addressed in chapter 7.
Note that in fig. 4.6 the total chlorine residual reduces to a near constant
value which remains even after 12 days. As well as free chlorine, total chlorine
generally contains also mono chloramine and dichloramine. In these bottles no
monochloramine was introduced, but due to interaction with organic substances
in the treatment process, organic chloramines are expected to be present in the
sample water [52]. In titrations they generally show up as dichloramine. Since
organic chloramines are almost non-reactive, they stay close to constant over a
period of 12 days (cf. section 7.2).

4.4 Temperature

Fig. 4.8 shows the change in temperature during the whole seven months of exper-
iments. Temperature is given on the left y-axis, inlet flow is provided as reference
on the right y-axis. The short spells of zero temperature in the measurements de-
picted in fig. 4.8 are artefacts which originate generally from periods of zero flow
into the pipe rig. The flow values show short spikes at the beginning of several
step flow trials. These are the initial overshoot which occurred occasionally at
flow increase. From fig. 4.8 it is seen that the temperature changed from about
22°C to roughly 8°C within the duration of the experiments. This correlates quite
well with the heterotrophic plate counts given in fig. 4.10 (see section 4.5.1.2).

4.5 Bacteriological Data

4.5.1 Heterotrophic Plate Counts

In fig. 4.9 typical results of heterotrophic plate counts, in this case for step flow
trial 5, are given. A logarithmic scale is used for better overview. In this particular
example no chlorine was present in the rig and the flow increase was 0.08 l/s to
0.9 lis (regime 1). Flow at inlet and plate counts for the pipe-rig inlet tank and

81
 TEMPERATURE AND FLOW
r------------------------,4
30
3.5
25
3

2.5

2
-
~
v.:

o......:J
CJ...
1.5

1
5

03-Jul 22-Aug II-Oct 30-Nov

DATE

Figure 4.8: Temperature on the left y-axis over the full experimental time pe-
riod with inlet flow on the right y-axis for comparison with the schedule of the
experiments

the three sample points along the rig (INLET, SP 1, SP 2, SP 3) are given on the
same v-axis on a logarithmic scale. Total particle counts for the same step flow
trial are given in fig. 4.1 on page 68.

4.5.1.1 Observations in the Pipe Rig During SFTs

The following observations can be made.

• Plate counts at the inlet of the pipe rig (tank) are significantly lower than
the counts elsewhere in the rig.

• Plate counts are in general before the start of the step flow trial higher at
the three sample points than at the inlet.

• Plate counts increase at the verv start of the step flow trial; the magnitude

82
 5
10 r - - - - - - - - , - - - - - , - - - - - - - - - r - - - - - .

~~ oSP 1
~_~.~~
-_X:
... t ..
.. ~.lI:.a - - ..,_.
"'_'w, -- -__
....... ~ ...... 4t.,_.
._._._ ..,.
::c+··x SP
.. .. SP 2
3

..,;-~- INLET
..
1,
,

FLOW

1
12:00 12:00 30-08-96

Figure 4.9: Heterotrophic plate counts of inlet tank (INLET) and three sample
points along the rig (SP 1, SP 2, SP 3) and inlet flow.

of this increase varies for different trials.

• Plate counts decrease generally within roughly one travel time, i.e. the time
it takes for the aliquot of water which was first to pass through the rig at
the higher flow rate to go all the way through the rig. Due to the fairly low
sampling frequency of grab samples (compared to on-line instruments), it
can not be said for certain, exactly how long it takes until the HPC settle
down.

• Counts generally stabilise within the final stages of the step flow trial.
Again, the time of settling down varies from step trial to step trial and
occasionally the decrease occurs throughout the high flow period.

• After the step flow trial there is initially no change in plate counts.

• Plate counts increase slowly at baseline flow. This can be seen from the
fact that immediately before the flow increase counts are always higher
than directly after t he previous 8FT.

83
 8 HETEROTROPHIC PLATE COUNTS AND FLOW
I0 r -- - - - - - - - - - - - - - - - ---.4

3.5

3
HPC ofSP2
2.5

0.5

03-Jul 22-Aug II-Oct 30-Nov

DATE

Figure 4.10: Heterotrophic plate counts of sample point 2 (HPC of SP 2). inlet
tank (HPC of INLET) and inlet flow. Plate counts are given on left y-axis, flow
on right y-axis.

4.5.1.2 Long-Term Observations in the Pipe Rig

In fig. 4.10 HPC of sample point 2 and the inlet tank are presented. Plate counts
are given on the left y-axis. The right y-axis shows the flow values which are
included for reference. As can be seen from the time given on the x-axis this
graph shows the HPC for the whole experimental period of 1996 (about seven
months).
From these data it can be seen that the plate counts in the incoming water
(inlet tank) stay low throughout the experimental period. However, the het-
erotrophs in the pipe rig decrease with successive increases of flow, the addition
of monochloramine and the decrease of temperature. ~ote that the decrease of
HPC in September 1996 coincides roughly with a decrease of temperature (cf.
fig. -l.S). There is. however. no one-to-one match between temperature and HPC.

84
 2'..J
~

CI')
~

s:
~
CI')
o .
~ 10'
0
~
~
0
U
§:
1
10
0 1 2 3 4 5 6 7
TIME [days]

Figure 4.11: Bottle samples without chlorine: logarithmic plot of two sets of
heterotrophic plate counts of glass bottles. Lines with diamonds show average
HPC, total range of replicate results is shown by bars terminated with 'x'.

which implies that the introduction of chlorine (also in September) and the flow
changes (starting again in October) influenced the HPC.

4.5.1.3 Observations in Bottled Samples

Some results of heterotrophic plate counts (HPC) of bottled samples are given
in figs. 4.11-4.13. Depicted are the colony forming units (cfu) found after the
time given on the abscissa. A. logarithmic scale is used for the HPC. Several sets
of three replicate bottles per time instant were averaged for the figures and are
given as a connected line in the graphs. At each time instant a range bar shows
in addition the spread of results (minimum and maximum HPC of all replicate
bottles evaluated at that time). Due to the different number of sets, a maximum
of six, twelve and nine replicate bottles are the basis of each data point (with
range bar) in figs. .,l.1L .,l.12 and .,l.13, respectively. (:\ote that not always all
bottles could be cvaluat cd.]
Depicted are all sets made with glass bottles, giving samples without chlorine

85
 -=
-

-o
,.

v
-
c,
.....

2 4 6 8 10
TIME [days]

Figure 4.12: Bottle samples with free chlorine: logarithmic plot of four sets
of heterotrophic plate counts of glass bottles. Lines with diamonds show average
HPC, total range of replicate results is shown by bars terminated with 'x'. The
triangle at the time 10 days indicates that the analysis of two bottles showed a
HPC of more than 300000, without finding the exact number.

(fig. 4.11). with free chlorine (fig. 4.12) and with monochloramine (fig. 4.13) as
disinfectant in the bottles.
The aim of this part of the experiments was to establish the biological stability
of the water as it would normally enter the distribution system. but in the absence
of any influence of the system (in particular the biofilm). Xote that the maximum
retention time of the pipe rig (between inlet and sample point 3) is about 48 hours
(i.e. an inlet flow of 0.071/s).
The obtained results of HPC in water without chlorine in fig. 4.11 show lower
counts as expected. Counts increase within the first two days only little and after
seven days the range of obtained values is very large. which implies that although
in some bottles HPC rose sharply there are other bottles with very little increase
of counts over the seven day period of investigation. Results of plastic bottles
show a drastic increase already at the second day. Since it is. however. not clear
if the material of the bottles is related to this observation. these results are not

86
 I02rr---------.---r---.---. _

o 1 2 3 4 5 6 7
TIME [days]

Figure 4.13: Bottle samples with monochloramine: logartithmic plot of three sets
of heterotrophic plate counts of glass bottles. Lines with diamonds show average
HPC, total range of replicate results is shown by bars terminated with 'x'.

included.
Considering again the pipe rig retention time of two days, these results indi-
cate that the water is biologically stable within the pipe rig, especially since the
retention time is reduced to less then four hours during SFTs (at 0.9 lis inlet
flow).
The graph of HPC of bottles with free chlorine given in fig. 4.12 shows a more
prominent increase of counts in the average. In one set the counts increase about
10 fold between day 3 and day 7, while other sets a fairly stable over 7 days.
The three bottles evaluated after 10 days show very high counts with two bottles
higher than the highest investigated count of 300000 (the exact HPC is unknown).
An increase in counts was expected since free chlorine reacts with organic and
inorganic particles in the water and thereby loses its disinfecting power. From
fig. 4.6 on page 80 it is seen that after about three days total chlorine is reduced
to about 0.1 mg/I, which represents the fraction of organic chloramines in the
water (see section 4.3.2). Since they are non-reactive, the sample water is at that

87
stage disinfectant free and the same behaviour of heterotrophs as with no chlorine
is observed, as would be expected. Again counts remain fairly low within a pipe
rig retention time.

The graph of HPC in bottles with mono chloramine is given in fig. -1.13. It
shows that counts remain approximately stable and low throughout the seven
day period. This reflects well the expected slower decay of mono chloramine (see
section 4.3.2).
In summary, it is seen that the disinfectants react as expected, i.e. free chlorine
is used up after a few days and mono chloramine lasts throughout a whole week.
The feed water of the pipe rig is within the maximum retention time of the system
biologically approximately stable. Any increase of bacteria in the pipe rig is likely
to originate in shear-off from the pipe walls.

4.5.2 Epifluorescence Microscopy Counts

Epifluorescence microscopy counts (as adopted by S.M. Mclvlath for this partic-
ular application [103, 104]) give the possibility to investigate in more detail if
bacteria are present in 'clumps' or not. All the epifluorescence data used in this
thesis were collected by S.:"1. McMath and A. Delanoue. In fig. 4.14 the number
of large. 'clumps' per ml (i.e. 'clumps' containing 10 or more bacteria) for step
flow trial seven is depicted. Data for the four sample points along the rig (on the
left y-axis) and the corresponding inlet flow values (on the right y-axis) are given.
For this step flow trial the flow rate was increased from 0.08 lis to 0.9 lis with
monochloramine present (regime 2). Fig. 4.15 depicts the same data for SFT 8.
Note that in this figure the example with the largest variability of counts at high
flow is given.
Fig. 4.14 is a typical example of epifluorescence results for large 'clumps'. A
less typical result is presented in fig. 4.15. There the data for step flow trial 8 is
depicted. It is the only step flow trial with such a marked variability at high flow
and therefore it is considered to be the most variable epifluorescence data obtained
during the 1996 experimental campaign. The other data also show variability at

88
 7000r------------------,
-.... 8
~6000
c, SP 1
~ SP3
a
~ 7
2. 5000 ~ :
en 6
~~ 5
U 3000
t.t:l

ffi 2000
U
en 1000
~
~ 0
2

~-1O()(J
I
1
I FLOW
15-~O=c=t===L_.....L.-_~====~========~~d 0 ..
12:00 16-Oct 12:00
TIME

Figure 4.14: Epifluorescence microscopy counts of 'clumps' of 10 or more bac-

teria of step flow trial 7; Inlet tank (INLET) and three sample points along the
rig (SP 1, SP 2, SP 3) are shown; Inlet flow values are given on the right y-axis.

high flow, but never to the degree as shown in fig. 4.15. It is obvious from the
data of 8FT 8 that the variance of bacteriological data can be very large.
In these examples, and the vast majority of the step flow trials, large 'clumps'
were absent from the inlet tank samples. Prior to the step flow trial no large
'clumps' could be found in the entire rig, but as soon as the flow was increased,
a significant increase in 'clumps' was observed. Only after the step down in flow
was there a decline in the number of large 'clumps'. This comparison with particle
counter size distributions is striking (see section 4.5.3).
The increase of 'clumps' is generally sustained throughout the high flow pe-
riod. Fig. 4.15 is to some extend an exception to this rule, however, even in this
case counts are on average throughout high flow significantly higher than at low
flow.

89
 ....... l0000r------,-----r----r--~--____r_--___,
~ Tsp
,, 1
S"
U
::l ,
It
'-"

...It
~
::J
II

o 5000 . '11
II
.-.~ .. ~p 2 .
u
-
II : ,"
t.tJ
U
Z
t.tJ
"I · ;! \\"
,
<Il
:::::
U
CI}
t.tJ
e::::
o
· Pr;!.;rr
··

•
I
. .,I

:1, .•
SP 3 \
".

\ •
::J ............... .'...• rt'ft-' ":.~~ .•
u: 0 ..... ......• : -f,
.!I
_
11'_' \..
••••
.. .r .'"•
" .~."'~ 2
c:t.tJ INLET ...--.....---,
FLOW
1

23-0ct 12:00 24-0ct 12:00 25-0ct 12:00 26-8ct

TIME

Figure 4.15: Epifluorescence microscopy counts of 'clumps' of 10 or more bac-

teria of step flow trial 8; Inlet tank (INLET) and three sample points along the
rig (SP 1, SP 2, SP 3) are shown; Inlet flow values are given on the right y-axis.
This example shows the worst results of the 1996 experiments.

4.5.3 Discussion

4.5.3.1 Heterotrophic Plate Counts

During baseline periods heterotrophs increase within, roughly, a week by about

a factor of 5 to 10 (i.e. half a log to one log) at temperatures between 15°C and
22°C). Results from the bottled sample (fig. 4.11) do not sustain a hypothesis
of regrowth of suspended bacteria. The results of the bottle experiments would
indicate that the biofilm in the pipe rig is an important contributor to the increase
of HPC in suspension at low flow.
The short peak in heterotrophs after flow increase is very likely due to a short
and heavy shear off of biofilm and re-entrainment of settled deposits. Subse-
quently the heterotrophic counts settle down to a lower value within about one
travel time. This indicates that a large amount of the suspended bacteria which
are present at low flow is washed out of the pipe rig. The difference between
inlet tank HPC and plate counts in the rig occurring subsequent to the initial
wash-out must then originate from shear-off from the biofilm (note that the travel

90
 HETEROTROPHIC PLATE COUNTS AND FLOW
r-------------------.2

1.8

1.6

1.4

1.2-;;:
:::::

-'-1
1
.. - 1 ~
9
0.8 tl.
1
I
1
1
0.6
1
1
1
0.4
1
FLOW 1 0.2
-------- -----------
1
.. _----
o 5 10 I~
TIME [days]

Figure 4.16: Simplified pattern in heterotrophic plate counts; inlet flow values
are given on the right y-axis.

time through the rig is then only a few hours). After the return of the flow to
the lower baseline values, the HPC recover.
The pattern of heterotrophic plate counts as discussed above is presented in
a simplified form in fig. 4.16. Typical numbers of plate counts are given on the
left y-axis, inlet flow values are shown on the right y-axis. This figure shows the
expected "ideal" (and simplified) response of HPC to a SFT.

4.5.3.2 Epiftuorescence Microscopy Counts

Epifluorescence counts of 'clumps' with 10 or more bacteria show a sharp increase

in number during high flow. Similarly, particle counter size distributions show a
relative decrease in small particles and an increase in large ones. Thus, nothing
conflicts with the hypothesis of a good comparison between these sets of data and
a simplified pattern as depicted in fig. 4.17 is proposed. Similarly to fig. 4.16, typ-
ical numbers of epifluorescence counts and typical percentages of particle counter
size distributions are shown on the left y-axes, while the right y-axes give corre-

91
 EPIFLUORESCENCE COUNTS AND FLOW
r-- r--
4

EPIFLUOR-
ESCENCE
COUNTS
r -. -
.... _--
FLOW
--- • 1
5 ----------10 •• • ..- -1~
TIME ldavsl
PARTICLE SIZE DISTRIBUTIONS AND FLOW
100 - urn
~
0 4
<
E--
Z 50
~
U
l::I:::
~
c,
5-lOurn
--I

0
0
FLOW
-- •
1 •

5 10
.. 1~
TIME [days]

Figure 4.17: Comparison of patterns in epifluorescence counts of 'clumps' of 10

or more bacteria and the particle counter size distribution; inlet flow values are
given on the right y-axis.

sponding flow readings. Both sets of data indicate a significant increase of sheared
off material from the biofilm within the experimental distribution system.
The only major difference in shape between the curves shown in fig. 4.17 is
that after the 8FT it took about one travel time for the percentages in the particle
counter size distribution to come back to normal (baseline flow) values. This is,
however, most likely also the case in the epifluorescence data, since due to the less
frequent sampling (this particular period was during the night), the slow decrease
in large particles after flow decrease was probably not detected.

4.5.3.3 Summary

The combination of step flow trials and particle counters provide a valuable tech-
nique of characterising biofilm detachment in water distribution systems.
Particle counter size distributions compare well to the simplified pattern of
epifluorescence results of large 'clumps' . However, because of the much higher
logging frequency (compared to grab sampling) and higher accuracy of measure-

92
ments, the particle counts give a more detailed time history if large 'clumps' of
biofilm dominate (this condition has to be met since they do not count any par-
ticles smaller than 2 usi: in diameter). Thus, particle counters will be of better
use in a high shear stress environment (as e.g. during a step flow trial).
The results of this section provide evidence on the size distribution of the
biofilm detached under a range of shear stress conditions.

4.6 Long-Term Effects

Throughout the twelve step flow trials the disinfectant and flow regime had little
effect on the size distribution of particles and the epifluorescence microscopy
results. However, successive increases of flow, the presence of monochloramine
and the decrease of temperature caused an overall decrease of heterotrophs.

4.7 Conclusions

This chapter gave an overview of the data that will be used for modelling water
quality in the following part III of this thesis.
Particle counts, both total and in size distributions, monochloramine, temper-
ature, heterotrophic plate counts and epifluorescence microscopy counts of large
'clumps' were discussed. Especially the bacteriological counts will feature only
marginally in the rest of this thesis, whereas the focus will be on total particle
counts and monochloramine.
Important results from this chapter include the observations from total par-
ticle counts and their discussion in section 4.1.2. This section paves the way for
the specifications of the particle counts model given in section 6.2. Similarly, the
observations from the particle counter size distributions are instrumental for the
numerical statistical models of the size distributions in section 6.3.2.
The discussion of the bacteriological results in section 4.5.3 provides further
important conclusions. It was indicated that the biofilm in the pipe rig is an
important contributor to the heterotrophic plate counts of suspended bacteria,

93
that the epifluorescence microscopy counts compare well with the particle counter
size distributions and that particle counters will be of better use in a high shear
stress environment. The relationship between epifluorescence counts and particle
counter size distributions is especially relevant for the work of chapter 6 because
it adds to the significance of a particle counts model that will be developed there.

94
Part III

Model

95
Chapter 5

Preprocessing of Data

5.1 Introduction

Preprocessing is concerned with the extraction of desired information about the

measured physical quantities form meter output data. A meter transforms phys-
ical quantities into the available measured data, as illustrated in fig. 5.1.

Input ,Ir-----
~.SystemI
Physical
Quanti1y
- - - - Measured
.1 Meter I
Data •

Figure 5.1: Physical quantities and measured data

The chlorine meters that are used in the experiments discussed in this thesis
exhibit nonlinear dynamics of an order higher than one, i.e. a slow meter drift.
Rather than attempting to model the transfer function of these meters, mea-
sured data are preprocessed with the help of additional, independent data, i.e.
the chlorine titrations. The measurements of chlorine concentrations provided
by titrations are considered to be more accurate than the meter readings and,
therefore, the meter were recalibrated at irregular intervals (roughly every week
or fortnight) using the titration data.
Different particle counters do not produce the same reading if fed with the
same water (cf. "Switching the Particle Counters" below). Thus, additional in-

96
formation about particle counter accuracy was obtained through switching the
water feed between the counters.

5.2 Switching the Particle Counters

The particle counters do not necessarily measure the same amount of particles
even for exactly the same parcel (or aliquot) of water. These differences are
attributed to differences in sensors, i.e. since sensors are calibrated to produce
the same reading for spheres of a certain diameter they may react in a sensor
specific fashion to particles of other shapes and thus produce a difference in
counts.

To obtain some information about the difference between sensors, the particle
counters 1, 2 and 3 (not 0 or inlet) were switched in a circular fashion (cf. below)
once a week. Due to the physical distance between the inlet particle counter and
the other counters it is impossible to incorporate counter 0 in the switching.

5.2.1 Ratios of Particle Counters

Columns 2 to 4 of table 5.1 contain the ratios of the average particle counts of the
same water measured one hour before and after the switching by different particle
counters (as indicated in the first row). Columns 5 to 7 of this table show the
ratios of average particle counts of different water counted by, respectively, the
same counter one hour before and after the switching of the counters. Sharp
peaks or meter failures were excluded in both cases.
The last five rows give some statistics which describe the spread of data in
the corresponding columns of data. Since ratios are considered in this context it
would be inappropriate to calculate statistics based on the data as given. The
aim of the linearisation is that the ratio of 0.5 should have the same distance
from the target 1 as its inverse 2. Therefore, the data has to be normalised and
linearised first. The intended mean is shifted from one to zero after a negation
and inversion of all data points between zero and one. If the original stochastic
variable as given in rows 2 to 18 of the table is referred to as x and the new

97
 Table 5.1: The ratios of the particle counters (pc) before and after switching

Same Sample Point Same Particle Counter

pc1/pc2 pc1/pc3 pc2/pc3 pc1 pc2 pc3
15.07.96 1.509 0.908 0.931 2.200 0.583 1.208
22.07.96 1.703 1.710 1.229 0.769 1.427 1.115
29.07.96 1.337 0.988 0.977 1.283 1.283 0.804
05.08.96 2.333 1.578 0.756 1.059 1.116 0.946
12.08.96 2.150 1.530 0.761 0.978 1.089 1.004
19.08.96 1.615 1.279 0.707 1.218 0.873 0.839
27.08.96 1.700 1.162 0.784 1.016 1.021 1.106
03.09.96 1.712 1.135 0.696 1.024 0.845 1.214
09.09.96 2.046 1.244 0.525 1.145 0.817 0.924
17.09.96 1.768 1.251 0.561 1.291 0.621 0.989
30.09.96 2.259 15.246 7.008 1.016 1.027 0.995
07.10.96 5.777 23.057 2.400 1.154 2.350 0.222
14.10.96 2.089 13.487 6.830 0.669 0.714 2.213
22.10.96 1.275 1.157 0.745 0.926 1.068 0.831
28.10.96 1.734 1.130 0.589 1.093 0.640 1.290
07.11.96 1.295 17.789 13.300 1.733 0.842 0.663
08.11.96 1.417 12.719 9.025 0.611 1.707 0.954
Statistical Analysis of linearised, normalised data
Mean 0.957 4.728 1.718 0.1 0.004 -0.16
Median 0.712 0.279 -0.276 0.059 0.021 -0.011
Interquantile 0.707 11.76 2.928 0.271 0.426 0.333
Range
Standard 1.048 7.442 3.875 0.42 0.507 0.933
Deviation a
amin for 0.818 5.805 3.023 0.328 0.396 0.727
Normal Distr.

variable is y, it is

x for x > 0
Y = - l = sign(x - 1) . (xsign(X-l) - 1) . (5.1)
{ -~ +1 for x < 1

From eq. (5.1) it is obvious that y = 0 for x = 1 (normalisation) and that inverse
ratios will feature with the same absolute value and different sign (linearisation).
For a meaningful analysis, all statistics given in the last five rows are in y. and
not in x.

98
 Mean and median are provided in rows 19 and 20. The large difference of these
two values in the case of columns 3 and 4 indicates data skew. Rows 21 and 22
give two measures of spread: interquantile range and sample standard deviation.
The interquantile range is the difference between the 25th and the 75th percentile
of the sample and has the advantage that it is not affected by outliers (like the
standard deviation). From the table it is seen that the data spread is largest for
columns 3 and 4, followed by column 2 and 7, while columns 5 and 6 exhibit
small deviations. In column 7 only the standard deviation is large. while the
interquantile range is comparatively low, this implies outliers, like the 0.222 in
row 13. Without the transformation of eq. (5.1) this outlier would not have been
recognised as such.
The last row gives the minimum standard deviation for which the respective
column can be considered to be still normally distributed (with that standard
deviation). It is

(n - 1)a2
amin = 2 . (5.2)
XO.05,n-l

The X2 test statistic for an error probability of 0.05 and n - 1 = 16 degrees of

freedom (n is the sample size) is used to obtain that result. It is X5.o5,n-l = 26.3
(see [18, p.21]). This test statistic is routinely used for the test of variance
(cf. [139, 78]). amin has to be small if, as in this case, a small variation of
data is required. For instance, if it is required to have a standard deviation not
larger than 0.5, only columns 5 and 6 are normal distributions, while columns 2
to 4 and 7 are with 95 % probability (significance level of 0.05) not normally

distributed.
The fact that the values in columns 5 to 7 are fairly constant and close to y = 0
(x = 1) could indicate one of two things: either the water quality throughout
the rig is consistent (as far as the particle type is concerned) over the counter
switching periods or the counters produce always the same measurements. no
matter what water goes through them (specific only for the respective counter).
If the measurement of the counters were independent of the sensor and it is

99
assumed that the water before and after switching is roughly the same as far as
the type of particles present is concerned, columns 2 to 4 would always contain
values close to x = 1 (or y = 0). Alternatively, if the measurements were merely
sensor specific, constant ratios should be found (which are not necessarily one)
throughout each column. From the results discussed above it is seen that none
of these assumptions is valid. The spread of data in columns 3 and -1 is large.
This data can only be expressed in a normal distribution if standard deviations
of more than 3.8 are assumed. Only column 2 is less spread out.
From this discussion it is concluded that

1. the particle counts depend on both the respective counter sensor and the
type of particles going through them (possibly in terms of a predominance
of shapes of particles);

2. the water in the whole of the pipe rig contains particles of roughly the same
type at most times of counter switchings (apart from few exceptions like
7.10.96); and

3. the particle type varied significantly over time; even a period of only one
day (rows 17 and 18) was enough to introduce noticeable variation.

5.2.2 Original Order of Particle Counters

For the purpose of analysis, especially concerning travel times, the data of the
particle counters need to be switched back to its original particle counter sequence.
For this purpose a record of all the dates and times of switching the counters
was kept. This data was imported into a file and used directly to modify the
raw data such that the original order 1-2-3 of the particle counters is preserved
within the entire data set. In the vicinity of the points of switching possible errors
(i.e. wrong sequence) within a period of not more than one hour (generally much

lower) may occur.

100
5.3 Chlorine Meter Data

The chlorine meters used during these experiments measure total chlorine by
passing water through an electrolytic cell containing a gold and a copper elec-
trode. The current produced by the electrolytic action is proportional to the
concentration of chlorine in water. The manufacturer claims the accuracy of the
meters as ±0.05 mg/l at a Full Scale Deflection of 2 mg/I. Analysis of data
suggests that this performance is easily obtained in the short-term but that the
instruments were subject to some long-term drift. The chlorine residual measured
in the considered system is generally in the range 0.05 and 0.4 mg/l, i.e. fairly
close to the limit of the meter range.
Potential errors associated with meter drift can be quite significant with re-
spect to the measured signal. Therefore, it is important to correct this as far as
possible before attempting to calculate the decay coefficient(s). This was done by
comparing the meter readings against titrations, a bench analysis which yields
values for free chlorine, mono chloramine and dichloramine. The sum of these
measurements is called total chlorine. Although there are errors associated with
titrations, it is generally assumed that they are not systematic and may be consid-
ered to give the 'true' chlorine concentrations of the system. To minimise errors
as much as possible replicate titrations were performed (i.e. 3 to 5 titrations of
the same water sample). After discarding outliers the averages of the remaining
titration values are used for the following considerations.

5.3.1 Correction of Chlorine Meter Errors

It is assumed that the (averaged) titration values give the 'true' total chlorine
concentrations, whereas the meters are incorrect. It is further assumed that the
error in the meter readings is due to a relatively slow meter drift. The aim of
this correction procedure is to find this error signal by using titration results and
then adjust the meter reading values as appropriate. Since a titration is a bench
analysis, titration values are unequally spaced and relatively sparse (between one
value a day to five values a day, no values on weekends). However. if the error

101
 . . • • 6 •••••• · •• •• ••• ··.4 .

. . ·_ ._ .
· .
· .

10-3
-1~----:O~.8:----0~.6-=----~O.-4-_-OL.2-----LO--O......L.2--0.L4--0.L.6--0..J....8-~
FREQUENCY IN lIday

Figure 5.2: Discrete Fourier transform of the multiplicative error function i~~(t)
of sample point 3 before preprocessing (but after reverse-calibration) on a loga-
rithmic scale.

signal is only varying slowly the Nyquist theorem is still satisfied (i.e. there are
at least two samples per period of the error signal).
Evidence for the validity of the assumption given above is indicated in the
Fourier spectrum of a typical meter error function as depicted in fig. 5.2. In
this graph an example of the absolute value of the Fourier spectrum of the
equally spaced (interpolated) and reverse-calibrated signal i~(t) - see eqs. (5.7)
and (5.8) - is provided. The reverse-calibration removes the sharp changes that
were introduced into the complete data by recalibrations. These changes may oth-
erwise result in a high frequency error signal independent of meter drift. From
this figure it is obvious that the spectrum decreases roughly exponentially with
increasing frequency which shows a low high frequency content of the signal.
The multiplicative error function is calculated as the ratio between the titra-
tion data Cb(XO, td and the chlorine meter data em(xo, t). Here t, denotes the
unequally spaced time instants of the titrations, t represents, for brevity. the
equally and closely spaced time instants of the meter readings, and Xo is the

102
distance of the point of investigation from the inlet. Since this section is only
concerned with data specific to each sampling point, the dependence on the dis-
tance is irrelevant and will be omitted Xo in this section. The multiplicative error
function fM(t i ) (with the omission of the distance dependence) is obtainecd as

(5.3)

To reconstruct the error function between the time instants t i . fM (ti ) is at first
reverse-calibrated and then interpolated. For this, first the set W of recalibration
windows is defined,

(5A)

The w recalibration windows W j are of the form

(5.5)

where tj is the jth time of recalibration and <5 is a small positive pertubation.
From the above equation it is seen that these windows are rectangular and fixed.
thus, they are not the moving windows frequently used in signal processing. Fol-
lowing eq. (5.5), the reverse-calibration factors are

(5.6)

Thus, the reverse-calibrated error signal f~(ti) is

(5.7)

103
 0.8 r--- - ----r-- - - ,---- - ---,- - - -,-- - -.------- -----,
ORIG I AL METE R READING. S.P. I
CORRECTED METER READl G. S.P. I
0.7

~ 0.6
* TlTRATl O S. S.P. I

e.!J
E
....J
<: 0.5
::J
o
~ 0.4
0:::
UJ
z
Q2 0.3 . .. ~
- -___.......E
J". '
.. ..
o....J
:r:
UO.2

0. 1

12:00 18-1 2-96 12:00 19- 12-96 12:00 20-1 2- 96

TI ME

Figure 5. 3: Typical example of the correction attain ed by preprocessing. Cor-

rected and uncorrected chlorin e data of sample point 1.

F inally, t his error fun cti on is interpolated

m -l

E~.f ( t) = L Ef,[(td . ei(t ), (5.8)

i= l

where ei(t ) den ot es t he lin ear int erpola ti on functi on that is non-zero between t,
and t i+l and tri is t he number of t itration t imes tj in t he conside red da ta. Th e
supersc ript I in di ca t es t hat E~7 (t ) is a n interpolated signal, t he superscript R
refers to t he reverse-cali bration of eq . (5.7) . The reverse-calibrat ion is int ended to
inverse t he effect of t he recali bra t ion s such t hat the chlorine reading directl y after
reca libration (a t t = tic + 6") is t he same as t he readin g just befor e recali br ati on
(a t t = tic - 6") . Since t he recalibrations guarantee that t he met er is within t he
linea r region of its scale the resultin g error fun cti on E~[ (t i ) sho uld ad her to t he

low frequ en cy assumption made above.

Finally t he error fun ction E ~7 ( t ) is fil t ered , in accord an ce wit h t he a um pt ion
of slow varying error , to rem ove any high frequen cy noi e a nd moot h t he . i mal.

104
 0.8 r - - - - - - - , - - - , - - - - - - , - - - - - - , - - - -.----------..,
ORIGI AL METER READI G, S.P. 3
CORRECTED METER READI G, S.P. 3
0.7 TITRATIONS, S.P. 3

::::::: 0.6
* J
~o
::
...J
-c 0.5
::>
c
if3 0.4
0:::
u.J
Z
C2 0.3
o
...J
:c
UO. 2

0.1

o '--------'-----'------------l ---l.- L - -_ _---.J

17-12-96 12:00 19-1 2-96 12:00 20- 12-96

Figure 5.4: Typi cal example of the correcti on attained by preprocessing. Cor-
rected and uncorrected chlorine data of sample point 3.

Thus,

(5.9)

where * denot es the convolution sum of t he discret e Fourier t ransform and bs(t )
is the impulse response of a fifth order Bu tterworth filt er. In t his case , a zero-
ph ase distortion Bu t t erworth filt er (no t ime shift between fil t ered and unfilt ered
signal) 1 was chosen . The cut-off frequency of 0.3 1/ day was chose n for a remaining
abso lute value of t he Fourier t ransform of approximat ely 1-2 91c of its maximum
value (cf. fi g, 5.2) .
Sin ce t he sam pling frequency of t he met ers is adjust able, another interpolat ion
step mi ght be necessary at t his poin t , such t hat t he correct ed chlorine da ta c(t )
can be obtained by mul tiplicati on of t he met er dat a em (t) with t he final IT OI'

1 Matlab fun ction filtf i lt .

105
function flJ-F (t)

(5.10)

Typical examples of the correction obtained with this procedure are given in
figs. 5.3 and 5.4. These figures show original meter data (dashed line). corrected
data (dotted line) and titration results (large asterisks). In fig. 5.3 data of sample
point 1 of 8FT 12 is shown. Here the titration results were much higher than the
meter data. The corrected data contains accordingly larger values. Fig.- 5.-1 shows
data of sample point 3 of the same 8FT. In this case initial values of titrations and
meter readings match well and in the second half of the 8FT titrations are lower
than the meter reading (the other way round in sample point 1). In the total
investigated time period? the maximal absolute error is reduced from 0.28 mg/I
to 0.13 mg/I, the maximal mean square error from 0.075 mg/l to 0.029 mg/I,

5.3.2 Alternative Correction Procedures

In the previous section, a procedure for the correction of chlorine meter data was
demonstrated that utilises multiplicative error functions and reverse-calibration.
Instead of these, additive error functions or windowed data could be used in the
correction procedure.
The additive error function (or error residuals) fA(td are the difference be-
tween the titration and the computer logged meter data, analogous to eq. (5.3).
They are defined as

(5.11)

The main criteria for the decision between additive and multiplicative error
functions is which of them better fulfills the low frequency assumption. Thus, the
Fourier transform of the error function, as shown in figure 5.2. has to be small for

2In this thesis, titrations between 5.10.96 and 19.12.96 were considered with the exception of
data between 5.11.96 21:00 hours and 7.11.96 midnight due to a treatment plant failure. Xote
that this time window is different from the one used in [97].

106
large frequencies. In this case. frequencies that have periods up to a few days are
considered to be large. Keeping that in mind, the cut-off frequency for the filter
was chosen to 0.3 l/day (period of 3 days). Accordingly. the percentage PFTO.3 of
the maximal value of the Fourier Transform at f = 0.3 l/day is used as the most
important indicator of a low frequency signal.
If FT(f) is the discrete Fourier transform of a considered error function at
the frequency f in 1/day, then

FT(0.3)
PFTO.3 = max (FT(f)) . 100%. (:).12)
f

Table 5.2: Percentage PFTO.3 of the maximal value of the Fourier transform at
f = 0.3d~Y·

Sample Percentages of Fourier transform of error function ...

fIW fIR fIW fIR 1\\",0 IR,O
Point A A M M f M f M
Inlet 49.86 % 3.24 % 0.81 % 1.32 % 43.68 % 14.31 %
1 10.11 % 6.47 % 0.35 % 0.93 % 3.11 % 3.44 %
2 20.88 % 10.92 % 1.20 % 1.84 % 22.35 % 6.51 %
3 8.96 % 8.85 % 1.46 % 1.39 % 9.22 % 2.72 %

Table 5.2 shows the percentage PFTO.3 for several error functions. Note that
only interpolated error signal are transformed, since the standard discrete Fourier
transform requires equidistant data. Columns 2 and 3 show results for additive
errors, columns 4 and 5 show multiplicative errors. In both cases, first data
that was interpolated within the recalibration windows (columns 2 and 4) and
then reverse-calibrated data (columns 3 and 5) are given. The windowed data
preserves the recalibrations, which naturally introduces higher frequencies into
the data as can be seen from comparison of f~W and f~R. This is not apparent
from comparison of f~\' and f~7 which is probably due to the shift of mean
because of reverse-calibration which is significant in multiplicative errors but not
in additive errors. The mean value (zero-frequency value or DC offset) is removed
from the multiplicative error residuals in columns 6 and 7. Here. again, an overall
larger percentage in windowed data is observed.

107
 Table 5.3: Additive error residuals in mg/l.

Type of
error
Sample
Point
Before
preprocessing
Recalibration
windows
Flow Reverse- 1
windows calibration I
Maximal Inlet 0.2124 0.1911 0.2195 0.0860
Error 1 0.2848 0.1714 0.1969 0.0664
2 0.2615 0.1406 0.2226 0.1307
3 0.1010 0.1027 0.1755 0.0975
Mean Inlet 0.0523 0.0481 0.0885 0.0258
Square 1 0.0754 0.0468 0.0712 0.0229
Error 2 0.0639 0.0392 0.0639 0.0286
3 0.0404 0.0315 0.0667 o. o'r')
_0_
I

Thus, the most important columns in table 5.2 are the ones concerning reverse-
calibrated data, i.e. columns 3, 5, and 7. Here a significantly higher reduction for
multiplicative error functions fIJ- to 1-2 % is seen, but even if the zero frequency
is not considered, i.e. if f~'O is investigated, the overall reduction in the Fourier
spectrum of the multiplicative errors is still better than with additive errors.
Therefore, a multiplicative error is chosen for the correction procedure.

There are still several alternatives concerning the windowing of the data. In
particular, recalibration windows, flow windows or pure reverse-calibration are
investigated. Only the reverse-calibration does not need any windows. however.
windowing does not interfere with the low frequency assumption as long as both
the interpolation and the filtering is done separately in each window. In general.
windows may help to contain problematic or erroneous data but the amount
of data is smaller than for the non-windowed case. In particular, recalibration
windows permit use of data in its raw form. and flow windows may allow us to
take account of flow dependence of meter errors. But especially the last case
does not give good results since some of the flow window contain only very few
titrations (one or two). Note that for the use of flow windows a reverse-calibrar ion

is necessary (to remove the effect of recalibrations).

Table 5.3 shows maximum and mean square of additive errors fA before and
after preprocessing. :\ote t ha t t his error alone is insufficient to judge the pre-

108
processing procedure; the preservation of the high frequency meter signal. as
guranteed by the filtering and tested by the Fourier analysis above, is the other
important criteria. It is therefore not appropriate to simply use a non-filtered lin-
ear interpolation of the titration signal, which would reduce all residuals to zero.
Thus, a residual error which reflects problems in titrations, remains. However,
this final error of the corrected data should be small.
The maximum error given in table 5.3 is the maximum fA in the invest igated
time period, the mean square error is the l2-norm divided by the square root of
the number of error values, i.e,

MSE= (5.13)

From table 5.3 it is obvious that a pure reverse-calibration (column 6) leads to

the best reduction in error compared to the data before preprocessing (column 3).
Therefore this method was adopted.

5.3.3 Summary

Based on the assumptions that the chlorine titrations are more accurate than
the meter data in the long term (low frequency) and that the meter data is
more accurate in the short term dynamics (high frequency), a procedure for the
correction of chlorine meter data was developed and tested in this section.
It uses multiplicative errors between titrations and meters, interpolates these
errors after performing a reverse-calibration and filters the resulting signal before
multiplication with the original (raw) data for correction.

5.4 Travel Times (Delay Times)

Travel time refers to the time the water will take or has taken to cover a certain
part of the pipe rig starting at the inlet shed. Depending on which place along
rig is to be reached and which (time dependent) flow is in the rig at that period

109
of time, the travel times will vary. In the rest of this thesis. the travel time
variable will be referred to as T and the travel time to a fixed point along the rig
(normally one of the three sample points) as T, such that T = T(Xo), if Xo is that
fixed point. Naturally, if the flow is constant, T is constant.
The calculation of the travel time is not part of the preprocessing of da ta. The
travel time is, however, a fundamental property of the investigated system and
it appears in various guises throughout this thesis. The knowledge of this time
is of particular importance for work related to chlorine decay (chapter I). since
there it is attempted to follow small aliquots of water on their journey through
the pipe rig, but it also features in the particle counter work of chapter 6.

5.4.1 Calculation of the Travel Time

The travel time is calculated indirectly. It is

v= iT f(t) dt (5.14)

with T = travel time, V = total volume to be covered within T, f(t) = time-

dependent volumetric flow rate, t = time. The zero value of time (t = 0) has to
be normalised such that t = 0 when the considered aliquot of water starts at the
inlet shed of the pipe rig.
In this case, the system is logged in discrete time and thus eq. (5.14) becomes

n
V L i, llt i (5.15)
i=l
with
n
T L llt
i=l
i
(5.16)

The index i of Ii and 6ti indicates the time dependance of both variables. If
the sampling time is uniform, eqs. (5.15) and (5.16) will be simplified by ~ti = ~t
for all 0 < i < n. The time dependence of Ii (or f(t)) can normally not be
simplified since step flow trials are considered, which implies that the aliquot s of

110
water of which the travel time is desired may have started at low flow or may
reach the considered sample point after the decrease of flow.
Equations (5.15) and (5.16) are used to calculate the travel time. Xote that
if t = 0 is chosen as suggested above, the travel time obtained in this way is the
time it took the aliquot of water that is at the point of investigation at t = T to
reach that point from the inlet, starting at t = 0 from there. In other words. T
is the time it will take to reach the point of investigation if the parcel of water
starts at the inlet at t = O.

5.4.2 Possible Errors in Calculating the Travel Time

The calculation of the travel time in eq. (5.14) or (5.15) depends on the covered
volume V and the flow f(t). The volume is calculated from length and diameters
of the respective section of the pipe rig; the flow rate is the measured inlet flow
minus extractions of meters and grab samples.
The basic (approximate) settings of these parameters as known from the con-
struction of the pipe rig are given below.

• The pipe rig length is roughly 500 m between inlet and sample point 1,
400 m between sample points 1 and 2 and 400 m between sample points :2
and 3.

• The meter extraction flows are about 0.007 lis at each sample point.

• The pipe rig consists mainly of 110 mm inside diameter pipe, but contains
two sections of 3 m of 50 mm inside diameter pipes and two sections of -l m
of 25 mm inside diameter pipes (one section each between sample points 1
and 2 and sample points 2 and 3, respectively).

• The extraction of water due to grab sampling is not taken into account. it
amounts to approximately 1.5 to 2 litre/day per sample point 011 routine
days and can increase to a maximum of 10 litre/day per sample point 011 a
dav of extensive sampling (e.g. step flow trial day).

111
 • The flow meter reading is directly logged and used as such without correc-
tion.

Since these data are frequently only approximately known (in particular the
pipe rig length), they may cause errors in the calculation of the travel time.
To check on the accuracy of the travel time calculation, the particle counter
data as discussed in section 4.1.2.1 on page 70 is used. If we assume plug flow
(turbulent flow conditions), peaks should be spaced in travel time intervals (travel
time between sample points and inlet). Aligning these peaks will be used in the
following as a means of fine tuning the parameters that effect travel time.
Errors in calculating the travel time may come about because of inaccuracies
in any of the following:

• the pipe rig length,

• the flow rate of meter extractions,

• the inside diameter of pipes,

• taking into account the smaller diameter sections,

• the extraction of water due to sampling, or

• the meter reading of the flow meter.

Using the basic parameters, the travel time was calculated and the peaks of
the particle counters were aligned. In doing so, however, significant errors in
terms of the difference in time of the data peaks occurred. These errors lie in
the range of approximately 20 minutes to 4 hours and 40 minutes. In column 3
of table 5.4 the alignment errors of several particle counter peaks are provided.
These errors are the maximum distances between particle peaks of the sample
points (including inlet, if appropriate). One example of particle counter data with
the original parameter values is given in fig. 5.5. There one can see t he particle
counts of sample points 1, 2 and 3 aligned with the inlet counts for 8FT 6. The
alignment error in fig. 5.5 is 20 minutes. The same data are aligned perfect ly
using the final correction, see fig. 5.9.

112
 2500.----r---...,------r---~-_ _r_--

.......
]2000
en
Q)
U
.~

':::1500
....Cen
=
oto)
.£
to)
1000
'€
Cd
0..
c;
~ 500t.-_--

06:50 06:57 07:04 07:11 07:19 07:26

Time of start at inlet

Figure 5.5: Alignment of particle counts with original parameters for SFT 6 on
8.10.96. Inlet (INLET), sample point 1 (SP 1), sample point 2 (SP 2) and sample
point 3 (SP 3) are shown. On the x-axis, the time of the start of the considered
aliquot at the inlet is given. Same raw data as fig. 5.9.

Note that due to the fact the flow rate is variable over the investigated time
interval the difference between two data points (which is originally the sampling
time) might expand or contract after alignment. For instance, an aliquot of water
that has reached the point of investigation only a few minutes before the current
time, may have started from the inlet an hour earlier than the aliquot that is
currently at this point. Therefore, in this case of time extraction, the resolution
of aligned data may become very coarse and a small error in the original data of,
say, a few minutes can become several orders of magnitude larger.

5.4.3 Error Correction

Several avenues for correcting the error in the travel time were explored. The
length of the pipe and the amount of extraction flow were varied initially. but the
final correction was achieved with the help of the particle counter peaks. However.
even the best correction procedure can not eradicate all errors. They may already
be part of the original particle counter data which is used to corroborate the

113
Table 5.4: Errors on abscissa after alignment of particle counts for original
parameters, adjusted length and final correction.

Flow Date Alignment error, Alignment error. Alignment error. I

in lis original parameters adjusted length final correction
i

0.08 14.12.96 not done not done 2 hr 15 min i

0.3 30.10.96 30-40 min 0-15 min not done

I

0.45 17.12.96 2 hr 30 min 1 hr 10 min not done !

I

0.6 24.7.96 2 hr 40 min 2 hr 20 min 2 hr 20 min i

and 0 min
0.6 1.8.96 30 min o min o min I
and 0 min
0.6 6.8.96 2 hr 20 min 10 min 50 min
and 0 min
0.6 9.12.96 3 hr 10 min not done 15 min
0.6 11.12.96 not done not done 3 min
0.9 21.8.96 4 hr 40 min 1 hr 10 min 1 hr 10 min
and 5 min
0.9 27.8.96 3 hr 30 min 40 min o min
and 0 min
0.9 8.10.96 4 hr 25 min 1 hr 15 min -15 min
and 20 min and 7 min o min
0.9 14.10.96 3 hr 30 min 50 min 25 min
and 20 min and 5 min o min
0.9 24.10.96 not done not done 15 min
0.9 30.10.96 not done not done 2 min

results and as is discussed above, already a small mistake of a few minutes may
have a large impact on the alignment.

5.4.3.1 Length Correction

Initially the length of the 110 mm diameter pipes were adjusted such that the
distance between inlet and sample point 1 is reduced from 500 m to 450 m. and
the distances between sample points 1 and 2, and sample points 2 and 3 are
increased from 400 m to 450 m. Resulting alignment errors for a length corrected
pipe rig are provided in column 4 of table 5.4.
With these corrections, the maximum error of approx. 4 hours 40 minu t ('s
was reduced to 1 hour 10 minutes. The worst alignment error found wit h this

114
 12000
,.......,

~10000
Q)
"0
'ic.. 8000
'--'
"J
C
g
(,,)
6000 ... : :.
Q)
"0
'i
c,
4000
S
~

15:50 16:19 16:48 17:16 17:45 18:14

Time of start at inlet

Figure 5.6: Alignment of particle counts with final correction of parameters for
SFT 1 on 25.7.96. Sample point 1 (SP 1), sample point 2 (SP 2) and sample
point 3 (SP 3) are shown. On the x-axis, the time of the start of the considered
aliquot at the inlet is given.

procedure is 2 hours 20 minutes (from originally 2 hours 40 minutes). Frequently

the error is reduced from errors between 30 minutes to 2 hours 20 minutes to
below 15 minutes.
In summary, overall a significant improvement in alignment errors is found.
however, with some remaining errors still rather large. As mentioned above. it is
important to realise in this context, that a large time difference may not mean
that there are many data points between the maxima, thus, the source of this
error may be a small mistake in the original data (cf. section 5.4,3.3).

5.4.3.2 Adjustment of the Meter Extraction Flows

An effect similar to the one described above could be obtained by changing the me-
ter extraction flows. These meter extractions are measured manually with mea-
suring cylinders. The measured meter extractions lie within the range 0.006 lis
to 0.009 lis. Generally, it is assumed that the error on these measurements is
below 10%. This was checked and validated bv comparison wit h the flow meter

115
 1 800 .-----r----.----.----.,----,--_~

1600 J

~ 1000
c
::l

8 800
~
u
.€ 600
~
0-
e; 400 . .
~
200

16:48 18:00 19: II 20:24 21:35 22:48

Time of start at inlet

Figure 5.7: Alignment of particle counts with fina l correcti on of param t rs f or

SFT 4 on 22.8. 96. Sample point 1 (SP 1) , sam ple point 2 (SP 2) and sarnpl
poin t 3 (S P 3) are shown. On the x-axis, the tim e of th e start of th e consid r d
aliquot at th e inlet is given.

readings. It is remarked t hat t he low base lin e fl ow is 0.07 li s.

Simulations were performed with met er extraction changes by 50<rc , 100<rc and
150%. However , onl y for significant changes in meter extra ct ion flow (i.e. 100-
150% of m easured values or from 0.007 li s to 0.014 li s or 0.0 175 li s, always
per sam ple point) an alignment similarly goo d as in t he case of length correc ted
resul ts is achieved. For t hat reason it seems unlikely t hat t he met er ext raction
flows have a big impact on t he erro rs in t he t ravel t ime.

5.4.3.3 Final Correction

T he final correction was achieved by taking t he t ime distance between t h peak.

of parti cle counts at various step fl ow t rials calculati ng t he required volurn of
pip to fi t such a t ime using eq. (5.15) , and averagi ng t he re ulting velum . Th e
volume between t he pipe rig inlet and sample po int 1 i changed from \ 01 =
4.09 m 3 to Val = 3.9 m", t he volumes between ampl - poin t 1 and 2 or 2 and 3
from \ 12 = V23 = 3.22 m 3 to Vl 2 = V~3 = 3.7 m' . Thi would refl t a cha n (' r

116
 80 r--,-------,--------r-----r-=---.-----.--~-_

I
....
....
...:::: I
'"
11)
(360
.......
:':l
0.
:;'50
....
c::
:;,
o
~ 40
u
]
t
5.30
:':l
'0
f- 20

dP:11 02:24 03:35 04:48 06:00 07:11 08:24 09:35 10:48

Time of start at inlet

Figure 5.8: Align m en t of particle coun ts wi th final correction of param et I' f or

14·1 2.96 at the baselin e fl ow of 0.08 lis. In let (I N L E T ) , sam ple poin t 1 (S P 1),
sample poin t 2 (S P 2) and sample point 3 (S P 3) are sh own . On th e x-tixis ; ili
tim e of th e start of th e consi dered aliquot at th e in let is given.

length from 500 m to 410.36 m and from 400 m to 389.34 m , both for a uniform
diameter of 110 mm t hro ug ho ut t he who le rig. Note t hat t he expe rime ntally
found t ravel t imes between peaks of t he same sa m ple po ints are necessarily not
exac t ly equal (d ue to uncer t ainty in t he data), however , if t he t ra vel t imes found
t hro ug h ave raged volumes fit t he da ta well , the conside red volumes are likely to

be close to t he t r ue ones.
Co lum n 5 in table 5.4 p rov ides error resulting from a n alignment of pa rti .lc
counte r d a t a based on t his final correction . F igu res 5.6 to 5.10 how ome of t he
particle counter afte r t he final correction of t he network parameters. Fi gure 5.11

dep icts t he same alignment for total chlorine met er da ta.

T he alignm ent errors given in colum n 5 of t a ble 5.4 a re aga in t h m aximu m
distance between t he peaks of sa m ple p oin ts 1, 2 and 3 or inl et and a m pl point 1.
2 and 3. Sin ce t he calculation of t he t ravel t ime i ba ed on t he di tan ce between
p oa ks it will no t be urprising if a few of t h a lign ment erro r ill colum n 5 of

tab le 5.4 are zero . How vel', t he fact t hat eleven of 21 alig nm nt erro r. riven ill

117
 2500 r-------,-------.-- -.---- - ..--- - -.--_ ------,

E 2000
-...
Vl
V
u
t::
~

=1500
~
c::
:::l
o
U

~ 1000
u
t::
~
0-
~
o 500 . SP 1· :
E-

06:50 06:57 07:04 07: 11 07: 19 07:26

Time of stan at inlet

Figure 5.9: Alignment of particle counts with final co rrection of paramei r. for
SFT 6 on 8.10.96. Inlet (INLE T), sample point 1 (SP 1) sample point 2 (SP )
and sample point 3 (SP 3) are shown. On the x-axis, the tim e of the start of th
consi dered aliquot at the inlet is given. Sam e raw data as fig. 5.5.

t he table are 5 min or less is very st rong evide nce t hat t he con idered volum e:
are correc t .

Two of t he wor st cases in t he table a re presented in fi gures 5.6 and 5.7. T he

are peaks from SFT 1 on 24.7.96 a nd SFT 4 on 21.8.96, resp ecti vely. In bot h
cases t he alignmen t error corres ponds t o a dist ance of only one or two expa nded
sa m pling int ervals. The expa nsion of t he sa m pling t ime is d ue to t he fl ow cha ng .
Som e of t he aliquots of wa t er t hat reach t he sam ple poin ts will have ta rt d at
t he inl et before t he start of t he SFT. This resul ts in a lin early chang ing t rav I
t ime for a range of subse que nt aliquots and t he refore there are aliquot a rrivi ng
a t a sa m ple point only 5 m in after ea ch othe r t hat started at t he inlet with a t ime
dist ance of more t ha n an hour. This is exactly t he ca e in fi gs. 5.6 a nd 5.7. T im. .
an error in t he da t a of only 5 to 10 min could be re pon ible for t h ali nrn cn t
.rr or s of t hese t wo cases. Such a n erro r is likely if it i con id I' d in addi t iOI1
t hat t he slope of t he peak of t he count align well , e pecially in fi g. - . .
T h i .xpl an a ti on of t h abov eli cu ed err or i ad op tr d. How ver . t h ali 11-

11
 • I LET I
E
.........
SP 1
~
u
150 SP _
t::
:':l SP 3
Co

~
§ 100
o
U
IU
U
't
:':l
Co
::a 50
~

00:43 00:: 0 1: 11 0 1:_6 01 :40 01:: : 02:0

Time of tart at inlet

F ig u re 5 .10: A lignm nt of pariicl count lL ith final ca rr iiou of paran i i t»

for part of SFT on 11.1 .96. Inl t (I '\ LE T) . sanipl point 1 ( P 1) . stu n.pl
point (SP) and amp l point ( P) ar sho uni. On th z-axis. th tim , of
tit "tart of til. can id r d aliquot at th inl i is gi n.

ment err or of 2 hours 15 minu t es of th e dat a of 1-1.12.96 eiven in fi a. - . must

have a diff rent rea on. In this case manv sampling point ' are wit hin th e interval
of t he alianment erro r and even th e slopes do not alien: th ere was, in ad lir.i n. no
fl ow change but a const ant fl ow of 0.0 1/ . The tota l t ravel time between sa rnpl
points 1 and 2 was approximatelv 27.5 hours, which mean s t hat t he err r in the
calcula t ion of t he travel t ime is ab out .2 Cj( . This is ~ till an acceptable a curacy.

In the maj ority of cases th e ali anment is al most perfect. f r inst an ce in FT

of .10.96 provid ed in fi n. 5.9 and FT of 11.12.96 sh wn in fie. - .10. Th e
ali unm ent err rs are 0 minut e: and 3 minut es. resp ect ivelv. T he raw data th at
is t he ba i f fi a . - .9 is the a rn e a t he ra w da t a of fie. - .-. where th e oriuinal

volume lead ' t a n aliunment err I' f:..O minut es.

T he ali znm ent error of tot a l chl rin e dat a is n t include 1 in r he t ab lo. It is
liffi cult t measu re since the lat a c nt ai ns n peaks but nlv xlgcs. which a rc
sino th d ut du e t chl rine iecav. An example of total chlorine da ta is uivcn
in fiuur - .11 that lepicts t h chlorin st cp tri al of :..) \ o\'r mlwr 1 (the tl o\\'

11
 0.8 r-----,--------.-----,-----r---r-------.----~

0.7

,.......,
0.6
::::::::
eo
SO.5
Q)
c:
·§ 0.4
..c:
u
~ O.3'" ....W_....--......".~r-;
o
f-
0 .2~.....

07: 11 08:24 09:35 10:48 12:00 13:I I 14:24

Time of start at inlet

Figure 5.11: Alignment of particl e counts wit h fina l correcti on of param t rs

for th e chlorine step trial on 26.11.96. Inl et (INL E T) , sam ple point 1 (S P 1) .
sam ple point 2 (SP 2) and sam ple poin t 3 (S P 3) are shown . On th e x-axis , ili
ti me of th e start of th e considered aliquot at the inlet is given.

regime is laminar with f = 0.08 li s). It is seen from t hat gra ph t hat t he edge

agree reasonably well.

From the graphs provided in this section and t able 5.4 it is concl uded t hat t he
final correc t ion of the volume of t he pipe rig is very close to t he actua l volume
of it. An error of up to 10 % or less t han t hree sam pling t imes is considered to
be acceptable. Therefore, 110 1 = 3.9 m ' and \ 12 = \ 23 = 3.7 m ' are used for t he

ca lculation of t he travel t ime .

5.5 Implementation
All calcula t ions t ha t are di scussed in t his chapter were implemen ted in Mat lab
as part of a Wat er Quali ty Toolbox (cf. appe ndix C).
T he calculation of t he particle ra ti o of ection 5.2 i reali d in pcrat i o. The

switching of t he parti cle counters into or igina l ord er i reali ed in t he Mat la b file:

wswitch a nd preproc .

120
 The chlorine related preprocessing of section 5.3 is realised in clcalib. which
uses residuals2, reverscal, resinterp, resfilter, wdezero. wtable1. and
wxload (all written by the author). The function clcalib is called in preproc.
The values for the appraisal of different chlorine preprocessing procedures as
discussed in section 5.3.2 are calculated by testclcalib.
The calculation of the travel time as discussed above (section 5.-l) is realised
in wttime (using a conditional loop). An approximate travel time (for constant
flow) is calculated by wtraveltime.

5.6 Conclusions

In this chapter various method for the preprocessing of water quality data of the
pipe rig have been presented.
The ratios of particle counters that were switched (thus using the same water
shortly after each other) were investigated. It was shown that the particle counts
dependent on both, the respective counter sensor and the the predominat particle
type in the sampled water. Furthermore, results indicated that the water in the
whole of the pipe rig contained roughly the same particle type at most times of
counter switchings and that the type of particles in water varied significantly over
time. The counters could be switched back to their original order.
The correction of chlorine meter data was a significant part of the author's
research work. A procedure based on the error between titrations and meter data
was presented. The titrations are assumed to be more accurate, however, t hey
are also sparse. Therefore a low frequency error function based on the ratios
between these data sets was found, interpolated and filtered. This function was
used to correct the meter data. Several ways of calculating the error function are
possible, thus reasons for the choice made in this section were given.
Finally, the correct volume of the pipe rig was obtained such that the travel
time can be calculated. Volumes of' ell = 3.9 m3 between inlet and sample point 1
and Vi2 = "23 = 3.7 m3 between sample points 1 and 2, and SP 2 and 3 were
obtained.

121
 The successful preprocessing of the data allows us now to investigate the data
further for the modelling of particle counts (chapter 6) or the comparison of
mono chloramine models and the determination of decay coefficients (chapter 7).

122
Chapter 6

Modelling of Particle Counter

Data

6.1 Introduction

6.1.1 Purpose

The purpose of this chapter is to develop a mathematical model which matches

the particle counter data. This model should be able to predict such data under
similar conditions and in addition. it should give some insight into t he undorlviuu
phenomena that gave rise to it. In particular. since the step flow trials (SFTs)
were intended to provide some information about the amount of biofilm present,
it is desirable to try' to obtain information on the biofilm thickness or the amount
of material sheared off. In other words, one import ant question to be asked is
"what can these data tell us about how the distribution svst cm as a whole would
behave in terms of suspended particles and shear-off potential of the biofilm?"

6.1.2 Particle Counters in the Test System

As described earlier. the svst om under investigation contains four particle conn-
tel'S located at the inlet and three sample points along the rig. The inlet part iclp

counter was installed about half wav through the 1996 experirnent . Then-fore.

no inlet data are available for the first two SFTs (of 1996). and only sporadic
hand-logged data for SFTs 3. --l and 5. Beginning wit h SFT (1. t he inlet particle

counter was fullv connected to the data acquisition svstem and t hus the (,ollipletc

123
data are available for processing.
All particle counters measure counts of particles per ml within certain size
ranges, i.e. spheres of diameters 2-5 usn, 5-10 usn, 10-15 usn. 15-20 utu. 20-
25 usn, 25-50 iuu, 50-100 usn, and more than 100 uui. Therefore. two different
types of particle counts can be considered, total counts (e.g. all particles with
diameter > 2 j..lm etc.), and difference counts (e.g. particles with 2 uu: < diame-
ter< 5 j..lm). In the following only total particle counts will be considered, mainly
those of all particles (i.e. particles with diameter > 2 j..lm).

6.1.3 Results

The main results of this work to date consist in a model that predicts the be-
haviour of total particle counts and provides the possibility to detect a profile
of material sheared off from the biofilm over the rig and in terms of time. This
profile will be termed apparent initial biofilm shedding profile (BSP). Obviously,
after loss of biofilm material due to shedding the biofilm will not have the same
magnitude anymore. Therefore, the profile will be a snapshot at the initial time
instant.

6.2 Model Specifications

The behaviour of particle counts during step flow trials (SFTs) is described in
detail in section 4.1. The observations of section 4.1.1 form the basis for and
most important substance of what will be termed specifications in this section.
These specifications are a set of requirements which any proposed particle counts
model has to fulfil if it is to match the experimental results.
In the following, these specifications are presented.

1. The 'Slow Signal' - basic specifications.

Type of signal:

a) Particle counts Increase slowly after a positive step change in flow

('Slow Increase').

124
 b) This Slow Increase reaches a maximum value ('~1.\'. ').

c) After reaching M.V. the particle counts decrease slowly to settle down
at some value, generally higher than before step up ('Slow Decrease').

Shape of signal:

d) The shape of the Slow Increase varies considerably but on ayerage

counts increase.

e) The Slow Decrease decreases generally initially faster than later on.

Slow Increase, M.V. and Slow Decrease together form the 'Slow Signal'.

2. 'Slow Signal' - quantitatively:

The M.V. is reached after approximately one travel time (the time it takes
to reach the considered point from the inlet of the pipe rig).

3. 'Slow Signal' - extended specifications:

a) At some step flow trials, the M.V. is reached after more than one travel
time.

b) Generally, specification Ic applies, however, occasionally the Slow De-

crease degenerates to a straight line.

c) The Slow Decrease is not always monotonic. Occasionally 'humps'

appear within this part of the signal (cf. section -1.1.1).

4. There are extremely sharp and short lived peaks occurring during the Slow
Increase period (specification 1). These peaks originate mainly from the
inlet and Sample Point 1 (occasionally also from Sample Point 2 and 3)
and propagate through the pipe rig ('Peaks').

5. Sometimes there occurs a sharp increase in particle counts at Sample Point

1 with step up and a sharp decrease at step down; this signal propagate-
through the rig ('Occasional Particle Step').

6. After step down in flow the tot al particle counts decrease slowlv
 7. They settle down to a potentially different value than before 8FT.

The nature of the data made it necessary to include several effects in these
specifications that occur only occassionally (specifications 3 and 5). A basic
model of particle counts during high flow will therefore include specifications L
2, and 5, while an extended model should take specifications 3 and -l into account
as well. Specification 2 and 3a have a special position in this context. since they
already include some quantitative information. This is obtained from the dat a
and thus is potentially inaccurate.

Peaks

I----~~p

Flow Step f(t) Particle Step Total Particle Counts p(t)

Slow Signal

Figure 6.1: Qualitative Simulink model of particle counter response to a step

increase in flow

Specifications 1-5 give rise to a qualitative model of the particle counter re-
sponse to a step increase in flow as depicted in fig. 6.1. Thus, an input-output
system with flow as input and (total) particle counts as output is considered here.
Note that a different model is necessary for step down (which has to fit specifi-
cations 6 and 7). Thus, this is not a general model, it is only valid for this tvpe
of input signal (which is a nonlinear characteristic).
A quantitative model matching these specifications will be developed within
this chapter.

126
6.3 Numerical Statistical Models

In a first attempt to generate a model for the particle counts response, a bottom-
up approach is adopted. In this section, the randomised numerical implementa-
tion of several hypotheses of biofilm detachment will be explored. This model will
be entirely numerical and based on statistical considerations. The starting point
will be possible hypotheses to explain the total particle counter behaviour or the
particle counter size distributions response to a SFT and show through simu-
lation if these hypotheses match specifications 1 and 2 of total particle counts
as given in section 6.2 or the observations of particle counter size distributions
as described in section 4.2. The main aim of this section is, thus, to check the
feasibility of suggested hypotheses for particle counts behaviour.

6.3.1 Total Particle Counts

As indicated above, this work will concentrate on the development of a basic

total particle counts model covering the Slow Signal only (specifications 1 and 2
in section 6.2). Peaks are considered to be independent of the Slow Signal (cf.
section 4.1.1). In addition, the focus lies on exploring the possibilities of basic
models before the extended specifications (spec. 3) are incorporated.
To model the total particle counts the pipe rig is first simplified to a one-
dimensional system, i.e the location along the pipe rig is characterised by only
one variable: distance. This implies that only the net result of anyone pipe
circumference is considered and no variation according to the location on the
circumference can be accounted for. This is a reasonable simplification since the
current interest is into the behaviour of particles at a certain distance along t he
pipe rig and it is not attempted to distinguish according to position within a
pipe cross-area. Furthermore, this simplification can be found frequently in the
literature [115, 137, 122].
For this model the pipe rig is discretised into n cells (for high accuracy. 11

should be large) according to figure 6.2.

To each cell i amounts of attached biomass ba (i) and suspended biomass b, (i)
 Figure 6.2: Schematic showing the discretisation of the pipe rig into n c II"

are assigned . For t he modelling of a parti cle counter resp on e to a fl ow increase

it is assumed , for simplicity, that ini ti all y (t ime t = 0) b, (i ) = 0 for all i.

To sim ula t e what happens in t he pipe rig t he foll owin g algori thm wa im-

pl ement ed. It includes different types of shear-off, whi ch will be inve t igated ill
det ail b elow.

Algorithm:

• d ecide randomly at eac h t ime inst ant if t he biofilm in cell i , ba (z}, is sheared
off

• in case of she a r-off, use one of t he foll owing types of shea r-offs:

1) all a t tach ed biomass in cell i is ent ire ly sheared-off

2) a clump of pred et ermined or random size (sm alle r t ha n ba ('i)) is shea red-
off from i , (i)

3) a co nsta nt fr action (e.g. half) of t he rem ainin g biofilm ba ('i) is shea red-
off

4) a d ecreasing fr action of t he remaining biofilm ba (i) is she a red-o ff; t h

d ecrease of t he fracti on is prop orti on al to t he remaining biofilm ba (i )

• propagat e t he sus pende d bi omass bs (i ) a nd add t he sheared-o ff a mo u nt to

it

• it era t e t he a bove procedu re for all loca ti on 1 a nd for a nu mb er of t ime

in st ant s

In t he foll owing, sim ula t ion resul t of t he diff r nt ty pe of hear- off a re d is-

uss d. In parti cular biofi lm mi gh t co n i t of ; t. rea m er: . [13 . 0] whi ch co mo

off t he wall as a wh ole. In t h i ca t h increa ed hea r for ce. would t hen impact

12
 5001-~11--'---'------r-----,_ _ ,--_~_ _~
:J
u
"5450
<':l
lJ
c:
': 400
e
::l
o
E
<':l
350

]
,-'' 2- 300
o
E
lJ
250
~
lJ
0.
c: 200
c;;
' 5 150

"E
::: 100
cr
"0

~
~
..c:
'"
0,5 1 15 2 ~5 3 3.5 4
time in travel times to point of investigation

Figure 6.3: Particle counts response to an increase in flow if all atta ched bionuiss
in a cell is sheared-off. Probability of shedding is 0.01.

in such a way t ha t all t he a t tach ed biofilm ma t eri al at any poin t in t he rig would

come off a t a random t ime during t he high fl ow p eri od (i.e. only once). Alter-
natively it could be t h ick biofilm of whi ch parti cles of varying sizes are detached
continuous ly.

6.3.1.1 All attached biomass in a cell is sheared-off

T he first assu m ption investi ga t ed in t his section is a situa tio n wher e only on .c

during high flow do particles come off any given loca ti on along t he pi pe wall. mcc

a one-d imens ional model is used t his implies t hat on on e who le ircumfer me

eve ryt hing is she d ded if she ar-o ff occu rs. This mi gh t be t he case if t h biofilm i.

pa t chy a nd co ns ists m erely of strea mers.

F igs. 6.3 a nd 6.4 sho w sim ulation resul ts with a probabili ty of h ddin of 1 Yc J'

(fig. 6.3) a nd 50% (fig . 6.4). O n t he y-axis t he amount of . hear d-off mat eria l is

given d epict ed as p rcen t age of t h mat rial in eac h c -II. T he t ime giv n on t hr

x- aXIS IS normalised such t hat t h - t ravel ti me to t h - poin t a lon r the rig which

is und I' on id ra t ion i on . It i obviou t hat t he ca c of -09C pr ba bilitv of

129
 400 rT- r ri - l '- - , ----r- --,-- - , - - - - - , . - ---,
1j
u
..c
1:b 50
a.>
.=
C
~POO
E
~

;;
''::: 250
:5
....o
C200
a.>

[
.=
;;
150

'r::
a.>
c:5 100
E
:::o
I
-0 50

~
..c
'"
0.5 1 15 2 ~5 3 3.5 4
time in travel times to point of investigation

Figure 6.4: Particle counts response to an inc rease in flo w if all attached biom ass
in a cell is she ared-off. P robability of sheddin g is 0.5.

shear-off do es not fit specification 1, since simulation parti cle counts a re high
during on e t rave l t ime a nd go to zero almost im medi a tely aft er t hat. However,
also fig. 6.3 do es no t mat ch specificat ions l a and ld , since t he simulat ion parti .le
counts do no t exhibit a clear t rend . They fluctua te du ring one t rave l t ime wit h
values between zero and a few ba (i) occurring randomly t hro ugho ut t hat per iod
and slightly longer afte rwards .

6 .3.1.2 Clumps of random size are sheared-off

T he assum ption t hat is investi gat ed in t his section is a sit uation where rep t t iv
shedd ing of mat erial (pa rticles) is t aking place. In t he case conside red h r , t hr
assum ption of biofilm consist ing pred omina tely of strea me rs i still u ed , howevr r,
s veral st rea me rs are possible along a ny given circumference of t h pip (i... any
given loca ti on ).
In fi g. 6.5 resu lts of a simulation with streame r of varying izc ar glY n. T h
random variation of st re a m r size has a constant mean . A imu la t ion of mul t iple
t r amer of con tant size .qu a l to t h mea n of the rand omly ized : t rea mer.

130
 120
::i
u
~
~
~
100
.5
C
='
0
E
~
BO
]
'c
'-
0
C 60
~

...u
0
0.
.5
<; 40
'C
0
c:;
E
'-
'-
0 20
I
"1j

~
0
..c
'" O~-~:-----l.-_----IIIo.L....-_------l._ _ ---.l.-_ _.L....-_ -.J
---.L_ _
o 0 .5 1 1.5 2 2.5 3 3.5 4
time in tra vel time s to point of investigation

Figure 6.5: Particle counts response to an inc rease in flow if clumps of random
sizes are sheared-off. Probabilit y of shedding is 0.5.

gives the sa me trend. The fi gure shows (as ab ove) t he amo unt of shea red-off
mat erial on t he y-axis (given as percent age of a mo unt in each cell) and t he t ime,
normalised to on e travel time, on t he x-axis.
The results show, in principle, a lin ear incr ease of parti cle counts , a plateau
and a lin ear decr ease. The gradient of t he increase and decrease depends on t he
average size of the streamers breaking off t he wall. The sha pe of t he response
is very close to specifications La-I c, however , a pla teau is not observed in t h
data. Still , if the parameters in t he mod el a re appro priate t his plateau might b
reduced t o a single Maximum \ alue. However specifi cations l e which require a
change of derivative through the decline, is no t fulfilled given t he st raight line in
t he simulations a nd t herefore t his case ca n only be vali d on t he occ a ion wh n
specificat ion 3b (straight lin e decline) applies and does not qualify a a general

model.

131
 120 I---,------,---~-___r--_,--.__-___.--~
1j
u
~
U
<':l
~
c: 100

C
:::l
o
E
<':l
80 .
:g

O~-~:__----L--...l....-.......::::::~--........l..---...l....-------l.--~
o 0.5 1 1.5 2 2.5 3 3.5 4
time in tra vel tim es to poi nt of invest igat ion

Figure 6.6: Particle cou n ts respo ns e to an increase in flo w if 1 % of the remaining

biofilm is sh eared- off. P robabilit y of shedding is 0 .5.

6.3.1.3 A constant fraction of the remaining biofilm is sheared-off

In this case a cont inuous shear-off is ass umed. The amount of biofi lm com ing off
t he wall dep ends on t he amount t hat is present on t he wall. A consta nt fraction
of biofilm is sheared-off if shedding takes place. T his im plies t hat t he biofilm i.
'rich', i.e. cont ains not just stream ers bu t consists of a layer of cells wit h stro ng
attachment t o t he wall and t he ca pac ity to supply t he bulk water repeat ed ly with
parti cles. In t he exam ple given in fig. 6.6 I % of t he biofilm rem aining on t he wall
is shea red-off if shedd ing takes place. Shear-off occ urs in anyo ne cell with a
probabili ty of 0.5. This figure shows t he a mo unt of sheared-off ma teri al on t he
y-axis (given as p ercen t age of amo unt in each cell) and t he t ime, norm ali .cd to

one t ravel t im e, on t he x- axi s.

T he sim ulat ion resul ts presen ted 1I1 fi g. 6.6 show a beha viour which follows

a n expone ntia l fun cti on. An In crease, Max imum \ alue an d 0 -crea e i obscrv d
as required by specifica t ions la- I c. In addit ion, specifi .a t ion 2 (l\'laximum Valu e
after one t ravel ti m e) is achieved , t he cha nges in d riva t ive a required ill specifi-
at.ion I ar a ttai n d a nd t he t rend of pecification l d is pr e ent (t he va ria hilit v

132
 Biofilm
Thick-
ness

100% 50 % 37.5% 30.5% 25.8%

o 1 2 3 4 5 Tim e Step

Figure 6.7: Il lus tration for th e su bseque n t remain ing amounts of biofil m u iih (l

decreasing fract ion (double- exponential) shear-off. Always 50 % of th er maiuuuj

fraction is sheared-off.

is readily introduced by assuming a variation over len gth of a mount of biofilm

present on the wall). Thus , this model fu lfils all basic specificat ions of sect ion 6.2.

6.3.1. 4 A decreasing fraction of t he r e m a ining biofilm is sheared-off

As opposed to section 6.3 .1.3 in this case t he amount of sheared-o ff m a teri al do e.

not only d epend on how much bi ofilm was left in a cell (j ust before shear-off)

but a lso on the amount of b iofilm that was in eac h cell befor e t he pr o .ess of

sh edding d ue to high flow started. In other words, th e amount whi ch is shea red-

off is a fraction of a fraction , e.g. in t he case of fi g. 6.8, 500/( of t he percen t age

of initial biofilm which was left on the surface befor e she dd ing . I.e. if 500/( of t he

initial biofilm remains , 25 % of this remaining bi ofilm is shea red-o ff. T hi typ of

she a r-off is ill ust ra t ed in fig. 6.7.

R esults of this simulation are given in fig. 6.8. Axes a re t he ame a on t ho

previous four figures, i.e. she a re d-o ff m a t erial in t he bu lk liquid at t h poin t of

investigation is given on the y-axis and t ime on t he x-axis, In fi g. 6. it i .en

that specifica t ions 1 and 2 are fulfill ed in much t he sam e way a in t he eel sc

given in sectio n 6.3.1.3 , however, du e to t he fact t ha t a dccrea in g fracti o n f

th r maining bi ofilm is s heared -o ff, t h fun cti on involved her e i of t hp tvpe

fCt) = xp (ex p( t)) .

133
 700 1---r---.----.-----r----.-_.--_~--~
1j
u

c:
::::l
o
E5 00
~

0.5 1 1.5 2 2.5 3 3.5 4

time in travel times 10 point of investigation

Figure 6.8: Particle cou n ts respo n s e to an increas e in flo w if a decreasing fra c-

tion of th e remaining biofilm is she ared-off. P robability of shedding is O. - .

Thus, it is no t possible to say for cert ain , wh at t he exact sha pe of t he 1Il-

crease/ decrease is. It will be necessary to make sensible assump t ions to obtain a
usable model.

6.3.2 Particle Counter Size Distributions

The numeri cal st a t ist ical models of t he parti cle counte r size dist ribu tions pr -
sente d in t his section a re based on t he to t al parti cle counts simulation in whi h a
const ant fraction of t he remaining biofilm is shea red-off (section 6.3. 1.3). T hr c
size ranges are sim ula ted, t hey are labeled as 2- 5 11m, 5- 10 p m a nd 10-1- JlIll in

fi gs. 6.9 and 6.10.

P er cen t ages of total particle counts a re assigned to t hese size ran ge . Pa rt.i .lc:
originating from t he bulk liqui d and a t tached parti cle a re d i t ingui shed . Th e
sus pended parti cles occur in t he per cen t ages 85 0/( for 2-5 I JIll , 10 0/( for - - 10 I JIl l

a nd 2 0/( for 10-1 5 I',m t he a ttached parti cle hav a ize distribution of 55 Cfc- .
27 0/( a nd 6 0/( . T hus, to m a tch obse rvat ions , a ma iler fra cti on of small part iclcs

and a larger p rc -ntag - of larger par ti cle i as.. ign cd to attached pa rt iclos. Th o

134
 100 i----.-- - - - - r - - -, -- - ,- - - ,_ _, - - _ - ,_ _ ---.
~
u
"5 90
\3
c
'': 80
c
='
o
E
C':l

:£
:5 60
'-
o
C
d)
50
~
d)
0. 40
c
-a
'C
~ 30
C':l
E
::::::
o
I
-0
~ 10
~ - --,----,--'"""-- 10-1 5 11m
til

0.5 1 1.5 2 2.5 3 3.5 4

time in travel times to point of investigation

Figure 6.9: Particle coun te r size distributio ns during high flo w if on ly a cons tant
fra ction of th e remaining biofilm is sheared-off. Th ree size rang es are ituirk tl as
repres enting 2- 5 urn, 5-10 tim and 10-15 urn. Probability of sheddin g is 0.5.

ratio between material in a cell in t he bulk liquid a nd a t tached ma t -rial ill a

cell is chosen to be 1:5. Note that these numbers are not necessarily identical to
actual per centages occuring in t he investi ga ted system .

The yield of material from t he pipe wall s behaves in t he sa me way as t he tota l

particle counts of sec t ion 6.3 .1.3 , i.e. t he shear-off from t he whole of t he biofilm
is expone nt ial. Simulation resu lt s for t his case are given in fi g. 6.9. T his fi gur e
shows a very similar response t o the total particle counts , i.e. t he size ran ge. ta rt
off at t he percentages for sus pende d bacteri a and a pproach within one t ra vel t im
the attach ed bacteria percentages via an exponet ial fun cti on. Aft er on - t ravel
t ime t he percen tages in creas e or decr ease (as a ppro pria te) wit h t h upend d
bacteria percentages as assymptotes. This is clearly not wh a t i een in t h . izc
distribu tions (see section 4.2 on page 74 or a ppend ix A). It i nor. expcc tc d to
see such a pronoun ced chang e of percent ages towards su pend d ba ct eria 1('\' (' 1.'

within t he high fl ow peri od.

Ther -fore it is n cess a rv
.'
to in lud - t he fi nal val u in th num cri cal sta t ist i .a l
mod I of t h - parti cl ounte r iz di trib uti on. T h - fi nal val ue (. pccificat.ion l r )
 l 00I-----.-----,---,-------,--..---.--_~ _
1j
u
"5 90
<:l
~
C
E 80

~ 70 ~-------'
2 -5 ~m

]
:5 60
'-
o
c 50
~

~
~
c.. 40
c l
i:::
1
o
I
5-1 0 urn
-0
~ 10
~ V,,------..,...----~~~---------___J
O~-__;;_;:-----:--____:.!:_-~--......l....---.L..----L-.--J
o 0.5 1 1.5 2 2.5 3 3.5 4
time in travel times to point of investigation

F igure 6.10: Particle coun te r size distrbutions during high fl ow. Sh ear- off con-
sists of a constant fraction of th e rem aining biofilm plus a cons tan t shear-off.
Three size rang es are marked as representing 2- 5 li m , 5- 10 urn and 10-1 5 Jim .
Probability of she ddin g is 0. 5.

implies a const a nt feed of particles from t he pipe walls. T his value is negle ·ted
in sec t ion 6.3.1 , because t he re t he focus was on mimickin g t he general sha pe of
the signal for which the final value has limi t ed significance . In t his case howev r,
t he cons t a nt feed of particles is relevan t . For t he simula t ion shown in fi g. 6.10,
a const ant feed of particles from t he pipe wall was introduced. To achieve t hi:
it is necessary t o split t he biofilm into two layers one layer t hat experi nee:
exponent ia l she a r-off, and a second layer t hat is a consta nt source of part i le .
This sec ond layer has to be large com pared to t he amo unt of shea red-o ff m a t e rial ,

otherwise it will be removed com pletely. In t he case of fi g. 6.10 it is cho en to b

10 t im es larger t han t he first layer of biofilm (t he layer of expo ne t ial hr ar- off) .
The simulation resul ts d epi cted in fi g. 6. 10 ma tch pecifica ti on 1 b tt r th a n
fi g. 6.9 . Th e percen tages remain close to t he value for a ttached biofihu . To
achi eve cert a inty a bout t he underl ying pr ocesses it woul d b ncc cHV to d i scuss

mol" cases to obtain a bet t er a pprecia t ion of wha t .ff t t h par ti cle count PI'
size di stribu ti on , How v I' t he fo u of t hi work i on t he to ta l pa rt icle count s

136
and therefore only this brief discussion of size distributions is included.

6.3.3 Conclusions

The numerical simulation of statistical models for total particle counts and part i-
cle counter size distributions gives valuable insight into several possible underlying
processes. This helps in the structure determination and the interpretation of the
related parametric models.

6.4 Parametric Model of the Total Particle Re-

sponse

The models of section 6.3.1 explored various hypotheses of how biofilm detach-
ment may come about. These models are, however, difficult to fit to data. A
parametric model is better suited for that purpose. In addition, the parametric
approach allows the application of numerous mathematical methods for t he de-
velopment and usage of a model. The parametric model anticipated in this work
will express the average amount of particles shedded from the pipe walls (i.e. no
random variation as in section 6.3.1).
The standard approach to parametric modelling from data (system identifi-
cation) requires the determination of the model structure before the parameters
of the model can be fitted to any (noisy) data produced by the modelled system.
From data alone, however, it is impossible to obtain the complete model struc-
ture. But it is possible to obtain some information about it, since the structure
is reflected in the shape (e.g. exponential etc.) of the signal produced by the
investigated system.
Possible approaches to obtain models of systems with unknown or partially
known structure are on the one hand 'Black Box Models' and on t he other hand
semi-physical modelling or 'Grey Box Models'. In this work a novel flavour of
'Grey Box Models' will be used which combines information gained from data
(the specifications of section 6.2) and assumptions based on physical or hiolog-

137
ical knowledge of the system. Although there are many examples of 'Grey Box
Models' in the literature, this particular approach is a contribution of the author.
This idea was first applied to a far simpler system [95. 94]. however, In most
details it is used there in a way different from the work presented here.
The numerical statistical models of section 6.3.1 help to identify the math-
ematical structure of the parametric model (i.e. to come up with appropriate
assumptions) because they explore mainly the shape of the data.
The model developed in this section is valid only for a specific input signal.
The applicability of the structure is limited to step flow inputs and, in addition.
the parameters depend on the magnitude of this step.

6.4.1 Assumptions

Assumptions complement the specifications of section 6.2 on page 124, they help
to establish a mathematical structure of the model. As already discussed in sec-
tion 6.2 and section 4.1.1 on page 68, three different particle counter signals orig-
inating from the step increase in flow will be considered (see fig. 6.1 on page 126).
These three signals are assumed to be independent. Amongst those signals the
ones described in specifications 4 and 5 are considered to have the least informa-
tion about the state of the system. Therefore, they are termed 'face-value' signals
in this work.

6.4.1.1 Peaks

Since the Peaks (spec. 4) are very short lived and the lowest possible sampling
time is 2 min, it is impossible to fit a model with any accuracy to them (not enough
data points). In addition, they are assumed to be independent of (i.e. of different
origin than) the Slow Signal. Therefore, in general the Peaks are removed from
the particle counter response to a step increase of flow before any modelling takes
place. The process of removing peaks (given in section 6.5.1.2) can be improved
by assumptions on the shape (or structure) of these peaks. It is. thus. assumed in
some of the work on automatic peak detect and erase (section 6.5.1.:2) that peaks

138
have the same structure as the Slow Signal discussed in section 6...1.1.3. The onlv
difference are the parameter values, in particular the value of the time constant
governing the decline of the signal.

6.4.1.2 Occasional Particle Step

The Occasional Particle Step (spec. 5) is assumed to be very close to a pure step
signal occurring with flow increase at sample point 1 until the flow is decreased
again. It has only face value since a pure step does not give any time information.
This assumption implies that particles are released at a near constant rate at this
location. It cannot be acertained from the data alone if this rate is constant. i.e. if
this signal is a pure step. For, while it is straightforward to read the magnitude of
step down in particles at the end of a SFT, it is difficult to determine for certain
what magnitude the step increase at the start of the SFT has. since this signal is
combined with a peak occurring at that point in time. However, the assumption
of a pure step can be corroborated with the help of the input signal as is seen in
the discussion, section 6.4.1.7.

6.4.1.3 Basic Model of the Slow Signal

The most interesting part of the particle counter response to a flow step is the
Slow Increase, Maximum Value and Decrease (the Slow Signal, specs. 1 and 2).
It is the only part of the particle counter response to a step increase in flow that
is included in the model which will be developed below. In other words, the 'face-
value' signals will not be included in the model. This Slow Signal is believed to
contain information about the biofilm on the pipe walls.
The most important hypothesis introduced in this section is the existence
of a local particle shear-off density (i.e. in par~~les per metre) which is decreas-
ing in magnitude over time as response to a flow increase. This concept is the
cornerstone of the model developed in section 6...1.2 and, as can be sp('n from S('('-

tion 6.4.4, gives rise to a model which fulfils specifications 1 and 2. This concept
was foreshadowed in section 6.3.1 where t he pipe rig is split up into cells and t I)(~

139
shear-off behaviour is subsequently varied. This assumption was first mentioned
by the author et al. in [98]. From this the first assumption is obtained.

Assumption 1 There exists a local particle shear-off density Pd(X. t) for t > 0
un'th unz'ts particles hi h
ml-rn W ZC
i d ecreasinq
is " ui magnitude over time. The distance along
the rig is described by the variable x, time is given by t. The shear-off starts with
the flow increase, thus Pd(X, t) = 0 [or t < O.

Please note that it is possible to assume the existance of an arbitrary local

particle shear-off density (not necessarily decreasing) without loss of generality.
The only limitation is that the shear-off density gives the total shear-off of one
complete circumference, however, any type of shedding can be expressed in that
framework. Furthermore, if the data is normalised to a zero level before the step
flow trial, the assumption Pd(X, t) = 0 for t < 0 will cover also a constant shear-off
level before the biofilm.
In a next step, it is assumed that the impact of biofilm can be characterised in
a meaningful way by one variable. The magnitude of this variable will depend on
the location along the rig. It is hopefully closely related to the biofilm thickness,
however, since only information about the suspended particles is available, the
only direct information that can be derived from that is on shedding (rather than
on what remains on the wall).
To take this assumption further, it is postulated that this biofilm shedding
profile does not influence the dynamics of the shear-off, but only the start (and
the end) of the particle density function. That makes it possible to separate
time dependent and distance dependent effects. Then, the second assumption is
obtained.

Assumption 2 The scalar valued function b(x) is a meaningful parameter for

the description oj the amount o] material initially sheared-off the biofilm and it
does not influence the dynamics oj subsequent shear-off. The shear-off function
s(t) is, thus, independent oj the distance I Therefore. the local particle density

140
function is now given by

Pd(X, t) = b(x) . s(t), for t > O. (6.1)

Assuming the feasibility of a local shear-off density is still insufficient for a

full model. An additional assumption on the shape of the local shear-off density
Pd(X, t) needs to be made. As will be shown in section 6.5 and 6.6, assuming the

shape of the local time-dependent shear-off function s(t) is sufficient to obtain

a meaningful model. In the following, an exponential decay as investigated in
section 6.3.1.3 is adopted. It is of the form

s(t) = k1 (f) exp( -at) + k 2(f), for t > O. (6.2)

A final value is included via k 2 (f ) in this case. Eq. (6.2) is only one of the
possible candidates for the local shear-off function s(t). It fulfils. however, all of
specifications 1 and 2 (in particular spec. Ie), as will be seen in section 6.4.4 or
was already seen in section 6.3.1.3. and has some other arguments in its favour.
This choice of s(t) fits the specifications very well, occurs frequently in nature (e.g.
capacitor discharge, biological growth processes) and offers distinctive advantages
in mathematical handling over, e.g. the corresponding function in case 4 of the
numerical statistical models (section 6.3.1.4). Note that, although the structure
of eq. (6.2) is valid for any step flow input, the parameters k, (f), k2 (f ) depend
on the magnitude of the flow step. Thus, the last assumption is obtained.

Assumption 3 The local time-dependent shear-off function s(t) follows an ex-

ponential decay. Therefore, the local particle density Pd(X, t) is of the form

Pd(X, t) = b(x) . (k1(f) exp( -at) + k2 (f )) , for t > O. (6.3)

Here a, k, (f) and k 2 (f ) are positive parameters. t is the time variable. .r IS

distance and b(x) is the function describing the amount of material that IS uutialu,
sheared-off the biofilm, as introduced in assumption 2. The parameter k, (/) and

1-11
k 2(f) depend on the magnitude of the input flow step.

6.4.1.4 Extended Model of the Slow Signal

The assumptions of the previous section cover the basic specifications, specs. 1
and 2 of section 6.2, but do not necessarily fulfil specification 3. An extended
model is required for that purpose.
In what follows, possible extensions of the basic model are given. All the cases
given above cover all requirements of specification 3, but not all of them lead to
a usable model. Note that specification 3b, a straight line as the decreasing part
of the slow signal, is possible even within the framework of all three assumptions
of the previous section, since an exponential decay may look like a straight line
in data if the time constant of the decay is large enough (and the collected data
window covers only the beginning of the exponential signal).

1. If assumptions 2 and 3 are discarded only the general particle shear-off

density Pd(X, t) remains. This case can include all the extended specifica-
tions. However, without more information about this density this abstract
expression is to general to obtain a usable model.

2. A feasible solution would be to replace the dynamic shear-off function s(t)

by a non-exponential or higher order exponential function. In fact, this
function might take any form. The input particle data could be exploited
to obtain it. This would imply that only assumption 3 has to be modified.

If a shear-off function is used which reaches its maximum value at t > 0

(with e:' the maximum is reached at the initial time instant t = 0), the
maximum is reached after more than one travel time, similar to the 'humps'

observed in some of the data.

This approach would essentially be within the framework of the basic model.
in particular the independence of s(t) from .r is maintained.

3. If the local shear-off function depends in general on the dist anco from the
inlet. i.e. it is a function of both .1' and t: s*(x. t). this case is essPIltially

142
 the same as case 1 of this section. The distinction between b( I) and 8- (I. t)
(both depend on x) is more or less philosophical and is not observable from
the particle counter data.

This case is, however, applicable in model development if the shear-off func-
tion is uniform within each section of the pipe rig. Since there is one particle
counter after each section, this would merely imply a loss of redundancy in
the data. In this case there would be three local shear-off functions 8
sec
.d t),
8sec.2(t) and 8sec.3(t).

4. Another feasible approach is to view the 'humps' discussed in sections 4.1.1

and 6.2 as delayed shear-off. This would be seen in the data as an increase
of particles after flow increase which follows the same pattern as the Slow
Signal. The origin of this could be biofilm which resists the shear forces for
some time and only late within the step trial is actually detached from the
walls. This effect could cover both 'humps' or a maximum value after one
travel time.

The particle density changes to

(6.4)

with 81 (t) = 0 for t < 0, 82(t) = 0 for t < 0 and td > 0 an arbitrary delay
time. Obviously, several delayed shear-off events are possible.

If both shear-off events in eq. (6.4) originate from the same "type" of biofilm,
the two shear-off functions 81 (t) and 82 (t) should be the same or very similar.
In particular, the time constant of the decay should be the same. This seems
to be the case at SFT 8 (fig. 4.2 on page 69 or fig. A.8 on page 240), where
the decay of the delayed effect at sample point 3 is within the same order
of magnitude as the Slow Signal.

On the other hand, if loose deposits that are lifted into suspension substan-
tially later than the start of the SFT are investigated, t he time const ant
of 82 (t) should be close to the decay time constant of the init ial peaks.

143
 This could be the case in SFT 3 (fig. A.3 on page 235). where the origin of
the 'humps' in all sample points would be loose deposits at some point in
section 1 (between inlet and sample point 1). This case is obviously not a
true part of the Slow Signal anymore, thus the resulting 'humps' should be
treated like the other peaks.

Although a number of feasible suggestions for an extended model of the Slow

Signal were put forward in this section, none of them is easy to implement. Later
on this thesis will be restricted to modelling the basic specification. The effects
that are captured in the extended specifications do, however. generate problems
during the process of parameter identification for the basic model. Therefore,
a theory concerning the origin of these effects has to be adopted to be able to
choose a sensible data window for parameter identification.

6.4.1.5 Inlet Particle Counter Data

It is assumed that the spikes in the inlet particle counter signal described III

section 4.1.2.4 on page 73 are an artefact generated by the counter aperture, since
they do not propagate through to other sample points. The underlying (additive)
signal is therefore the part of the inlet counter data which is relevant for this work.
The inlet counter is located only a few metres downstream of the inlet valve and
pump which mark the beginning of the pipe rig, T = O. Due to the short lengt h of
pipe, the signal p(t) that is observed by this counter is very close to the pure local
particle density Pd(X, t). Further down the pipe rig the signal p(t) is convoluted
and therefore significantly different to Pd(X, t) (cf. section 6.4.2). This similarity
is very useful for validation purposes, see the discussion in section 6.4.1.7.

6.4.1.6 Model for Particle Counts After Flow Decrease

It is assumed that the local shear-off as described above ceases with decroasr- of
flow and merely the propagation of already suspended particles is responsible for
higher counts in the pipe rig (cf. section --1.1.2.5 on page 7--1). A negative slope
model based on this assumption is developed in section 6.--1.3.
6.4.1.7 Discussion of Assumptions

The distinction between basic and extended model could be due to the ..type
or quality of biofilm on the pipe walls. The first step flow trials of each regime
(SFT 1, 4, 6) fit the basic model best, whereas the last step flow trials (SFT 10.
11, 12) fit it worst. Thus, it is possible to distinguish between times of thick and
uniformly structured biofilm ("rich biofilm"), when the basic model fits well, and
times of thin and patchy biofilm ("lean biofilm"), where the extended model is
necessary. For a "lean biofilm" scenario it is expected that the material comes off
in patches, which would explain the "humps" observed in the signal. while the
shear-off function s(t) is fairly flat. This difference in biofilm ..type" may stem
from the repeated flow increases and the introduction of monochloramine.
This shear-off function might be deduced from the inlet particle counter re-
sponse. As mentioned earlier, the inlet particle counter is about 5 m downstream
of the inlet valve, which marks the beginning of the pipe rig. Because this sec-
tion is basically as small as a cell in section 6.3.1 the signal observed at the inlet
counter is almost like a pure example of the shear-off function s(t). The signal
is, obviously, noisy and due to noise in the other counters it is advisable to retain
as much redundancy as possible in the data. Therefore, the inlet particle counter
signal is used to check results and not to modify the shear-off function s(t).
The impact of the uncertainty about the exact shape of the Occasional Particle
Step can be assessed in the light of the inlet particle counter response. It will
be seen in section 6.4.4, that the shape of the shear-off function is similar to the
shape of the particle counter response after one travel time. Therefore, if the
shape of the rig counters corresponds well with the shape of the inlet counter
(which is the case) it can be assumed that the Occasional Particle Step reallv is

a pure step function.

Therefore in the following the assumptions for the basic model will be adopted
and data which belongs to effects related to what is described in the extended
specifications (i.e. the "humps") will be excluded, That implies that the assump-
tion of belated shedding for anything related to the oxtended model is adopted
(case 4 in section 6.4.1.4), since the basic model and extended model are separa-
ble.
The task in the remaining part of this chapter is then to find a model from
the assumptions in section 6.4.1.3 and to identify both the parameters of s(t) and
a numerical representation of the shape of b(x).

6.4.2 Development of the Positive Input Slope Model

In this section, a basic model of the Slow Signal (specifications 1 and 2 of sec-
tion 6.2) based on the assumptions of section 6.4.1.3 will be developed. This
model captures the response of particle counts to a flow increase (positive slope
of the input signal flow, i.e. start of a step flow trial).
The starting point is the particle count density Pd(X, t) described in sec-
tion 6.4.1.3. Its units are P%~~~es and it depends on both the point x along
the rig, and the time t. Let x = 0 at the start of the rig and x = Xo at the point
under investigation. To obtain the total particle counts p(to) at point Xo and time
to from the particle count density Pd(X, t), it is necessary to add the contributions
of the particle density at x = Xo, t = to to the contributions of the densities along
the rig at previous times (t < to), which have been propagated with the velocity
v of the water to the point x = Xo. In the example of section 6.3.1 it would be
necessary to sum up the appropriate contributions of each cell. In the case of a
true density Pd(X, t), however, it is required to take the limit of that sum and,
therefore, an integral is obtained. Thus, the total particle counts p(t) at x = Xo
are given by the integral over all particle count densities reaching x = xo:

fO 1
p(t) = lx=xo Pd(X, t - ;(xo - x)) dx, (6.5)

where v is the velocity of the water in the rig (in ,:). The limits of the integral
could be reversed, which would make it necessary to replace .r by Xo -.r. Eq. (6.5)
requires the velocity v through the pipe rig to be constant for the entire time the
water travelled from x = 0 (i.e. the inlet) to x = Xo. The argument of time,
t - ~(xo - z}, determines that thr time instants of the particle density Pd("')

146
which are considered in the integral at time t correspond to the amounts of
particles originating from the distances 0 < x < Xo that have reached Xo at
time t. If the time argument is written as given above , it is necessary. to start
integrating from the point of investigation x = xo. Note that the fact that only
one velocity v is used implies that Pd(X, t) = 0 for t < 0 (no shear-off before the
positive slope of flow occured, as assumed in section 6.4.1.3). since otherwise the
argument t - i(xo - x) would have to take into account the lower velocity before
the flow increase (cf. section 6.4.3).
If travel time T = ~v is considered instead of the distance x, it is necessary
. to
deal with the integral

p(t) =v ·l:T Pd(vr, t - T + r) d r. (6.6)

In the above equation, T = :£Q.

v represents the travel time from the inlet to the
point under investigation, x = xo.
Incorporating the second assumption of section 6.4.1.3 means separating the
dependencies on distance (or VT) and time t: Thus,

p(t) =v ·l:T b(vr) . s(t - T + r) dr, (6.7)

Here, s(·) represents the dynamic (i.e. time-dependent) function describing shear-
off from the pipe walls, while b(·) represents the initial- values of the dynamic
shear-off, thus it is the wanted biofilm thickness or the initial biofilm shedding
profile.
A substitution {) =T- T and a change of argument of the biofilm shedding
profile to b* ({)) = b(vT - v{)) = b( VT) finally yields

p(t) =v ·1:0 b'(t9) . s(t - t9) dt9. (6.8)

Eq. (6.8) is a standard convolution integral. To verify this. recall that J~oo b*(d)·
s(t - {)) d{) is the standard convolution integral; to obtain the limits as given ill
eq. (6.8) note that b*({)) = 0 for {) < o. b*({)) = 0 for {) > T and s(t) = 0 for t < O.

1-17
Excluding all values of iJ where b* ('1J) = 0 leads to the limits given above. The
fact that this is a standard convolution integral is extremely useful for obtaining
a model, since the convolution describes a linear system.
Note that this equation describes only part of the water quality system. It
considers only the response of particles to a particular input: a step in flow.
Therefore this is a linear 'subsystem' of a larger, nonlinear system. The model of
this 'subsystem' is depicted by a Simulink block diagram in fig. 6.11.

F
f
[TI
IO w~--I"·ILI Dynamic Shear-Off 0(1) ~~t------I".ceJ Total Particle Counts p(t)
f(t) Local Response Biofilm Profile
B*(s)

Figure 6.11: Linear system with input flow f (t) and output total particle counts
p(t)

Due to the commutative nature of the convolution the second half of fig. 6.11
is equivalent to fig. 6.12. This model allows us to determine the biofilm shedding
profile b* ('1J) and from there b( x) by using the inverse of the Laplace transform
of the dynamic shear-off 8(8). Based on the assumptions of section 6.4.1.3 the
equation which describes the dynamic shear-off 8(t) is known, however the ade-
quate parameters are not. Normally these parameters would be found by using
both the input and output data, however in this case the input is wanted and
therefore unknown. On the other hand, some knowledge about the biofilm shed-
ding profile is available. It has to start at the beginning of the pipe rig (the
pump), i.e. at x = 0, T = 0 or '1J = T and it has to finish (for the purpose of
investigating the point x = xo) at x = XO, T = T or '1J = 0, since it is impossible
that the biofilm downstream from the considered point has any impact on the
suspended particles. Thus, this knowledge about the biofilm shedding profile has
to be used to obtain some information about the otherwise unknown parameters
of the dynamic shear-off 8(8).

148
 Incorporating the actual system s (t)

s(t) = k 1 (f ) exp( -at) + k2 (f ), for t > 0

which was assumed in section 6.4.1.3, yields

with

I if t > 0
1()
t = (6.10)
{ o otherwise

Using the Laplace transform L:{-}, eq. (6.9) reads as

P(s) = v' B*(s) . 8(s) = vB*(s) (ks +(fa) + k s(f ) ) ,

1 2
(6.11)

where P(s) = L:{p(t)}, B*(s) = L:{b*(t)} and 8(s) = L:{s(t)}.

Biofilm 5h~t-ing-p-r-Of-i1e-----I·~[]~----T-o-ta-1p-a":~counts
b*(t) Dynamic·5hear-Off 5(s)
p(t)

(response to a step in flow)

Figure 6.12: Linear system with apparent initial shedding profile b* ('19) as input
and output total particle counts p(t), positive input slope model.

6.4.3 Development of a Negative Input Slope Model

A negative input slope model has to cover the particle counter response to the
decrease of flow at the end of a step flow trial. This section follows the assumption
of section 6.4.1.6, i.e. the local shear-off is reduced to zero with flow decrease at
t = TE . Therefore, it is Pd(X, t) = 0 for t > TE in addition to Pd(.r. t) =0 for
t < O. In contrast to the positive input slope model of section 6...l.2. different
velocities need to be considered in this case.

149
 The general fundamental equation is, in analogy to eq. (6.5).

pit) = l:xo v« ( X, t - 7(XO, v(t)) + 7(1".1'(1)) ) dz. (6.12)

Note that in previous notation T is used as travel time between r = 0 and .1' = .1'0,
thus T = T(XO). Recall that the time interval T(XO, v) - T(X, v) is the time it took
the aliquot of water that is currently at Xo to cover the distance from .1'.

If the step decrease of flow is assumed to be ideal, i.e. the time it takes to
reach the lower flow is negligible, only two different velocities need to be taken
into account, VH during high and VL during low flow. Thus, to obtain p(f) it
is necessary to find a distance XE(t) such that the aliquot of water which IS

presently at Xo travelled the distance 0 < x < XE(t) with the velocity UH and
XE(t) < x < Xo with velocity VL.

The integral of eq. (6.12) can be split to form

p(t)

dx. (6.13)

The correction T(XO' VL) -T(XE(t), VL) oft is necessary to adjust the time variable
t to the new origin XE(t).
Eq. (6.13) together with the condition Pd(X, t) = 0 for t > T E describes the
particle counts model for a negative slope input completely. The decomposition
Pd(X, t) = b(x)s(t) is still valid with the restriction of s(t) = 0 for t > T E. Xot«
that the distance between the point of observation Xo and XE(t) is the distance
covered since the end of the step flow trial TEat the low velocity rt.. thus

(6.1-1)

150
 2r----r-----.------r----, 7
~ 6
Q)
E1.5
"C E5
C
as ~

~
E 11--------------1 II)

s (])
~

.23
II) t
(]) as Travel Time
.20.5 1l..2
t
as
Il.. 1
0
20 40 60 80 0 5 10 15 20
Time Time

(a) Linear shear-off function (b) Particle counter response

Figure 6.13: Linear shear-off function and corresponding particle counter re-
sponse for a flat Biofilm Shedding Profile.

Due to the time dependence of the integral limit XE(t), it is not possible to
transform eq. (6.13) into a convolution integral. However, if all parameters of
s(·) and the biofilm shedding profile b(·) are known, it is possible to calculate an
approximation of p(t) from this equation by replacing the integrals through sums.
Since the parameters and the biofilm shedding profile can be found by means of
the positive slope model and the high flow data, this approach is feasible.

6.4.4 Examples

The aim of this section is to give some examples which illustrate the range of
possible particle counter responses p(t) for a given initial biofilm shedding profile
(BSP) b(·) and a particle shear-off function s (.) as related in the basic model

p(t) = v ·1:T b(vr) . s(t - T + r) dr, (6.15)

given in eq. (6.7) of section 6.4.2. Within the framework of eq. (6.15) also ex-
amples which differ from assumption 3 of an exponentially decaying shear-off
function are investigated.
Figs. 6.13-6.15 show the results of the variation of the shear-off function s(t)
on the particle counts p(t). For these figures a flat biofilm shedding profile b( I) =

151
 2.5.---- - - --r-- - - - -r--- - - --, 121T- - -.---,- --.------ - - - - ----.

10

E 8
'-
Q)
a.
'-
Q)
~ 6
a. 1 U
en t
Q) ctl 4 Travel Time
u a..
~0 .5
0- 2

5 10 15 5 10 15 20
Time Time

(a) Exponential shear-off funct ion (b) Particle counter resp onse

Figure 6.14: E xponentia l shear-off f un ction and correspon ding particle coun te r
res ponse f or a fiat B iofil m Sh edding P rofi le.

8 60

-
Q)
'-
Q)
E6
50
"0
c
E 40
'-
ctl Q)
a.
~4 ~30
0u
Q)
a. -e
en
Q) ezo
0-
Travel Time
o~ 2
t::
ctl 10
0-

0
50 100 150 0 5 10 15 20
Time Time

(a) Sum of two exponential shear-off (b) Part icle count er resp on e
fun cti ons (second order system)

Figure 6.15: S um of two exponen ti al she ar-off f un ct ions ( con d ord r yst m)
an d corres pon ding pa rt i cle counter response f or a fiat Bi oji lin Sh dding Profil .

152
 6 .---- ..-- - -------.-- - - -.--- - - .------,
40 1T- - --.--,-- ----.-- - -..- - ----,

....
Q)
a.
~ 20
o
t(\j
0- Travel Time
;;::
o 10
m1
O ' - - - " " ' - - - - -- - - ' - - - -....I.-- - - '----'-....J
-10 -5 0 5 5 10 15 20
Distance Time

(a) Decreasing Biofilm Shedding P rofile (b) Part icle counte r response

Figure 6.16: D ecreasin g B i ofil m Sh edding P rofile and correspon ding part icle
coun ter response f or an exponen tial sh ear-off fun ction.

6 r -- ..-- - -.-- - - ..-- - -.--.-. 35

Q)5 30
;;::
e
0- 4 E25
0> ....
c
:0 ~O
"03 . (f)
Q) Q)
.c u 15
CJ)
t
E2 (\j Travel Time
;;:: 0-10
o
m1 5
O L--"""--------'-----L----~--J

-1 0 -5 0 5 00 5 10 15 20
Distance Time

(a) Increas ing Biofilm Shedding P rofil e (b) P ar ticle counte r re pon e

Figure 6.17: In creasin g B i ofil m Sh edding P rofi le an d corresponding partie!

coun te r respons e fo r an exponen ti al she ar-off f un cti on .

153
 70
8
;;::
Q)
60
....
0
CL
6 E50
c
C> ....
:.0 ~40
"C en
~4 .~30
Q)

(J)
t
E ctl Travel Time
0-20
'52
m 10
0 00
-10 -5 0 5 5 10 15 20
Distance Time

(a) Qu adra ti c Biofilm Shedding Profile (b) Par t icle counte r resp onse

Figure 6.18: Quadrati c B iofilm Sh edd ing P rofile and correspon ding particle
coun te r res ponse fo r an exponentia l she ar-off f un ction .

1 is assumed , i.e. t he she a r-off fun cti on s(t ) is t he same as t he shea r-off density
Pd (X, t ).
Fi g. 6.13(a) provides the sim plest exam ple: afte r fl ow incr ease a co nst a nt
shear-off occurs . Thus , t he t rans fer fun cti on of t he dynam ic syste m S( s) given

in eq . (6.11) degen erat es to a pure int egrat or (k1( f) = 0). T he respon se to a

fl ow incr ease which sta rted at t he normali sed t ime t = 1 in a system wit h t ravel
t ime t = 6 is given in fig. 6.13 (b ). The start of t he a 8 FT a nd t he ela pse of
on e travel time to the point of investi gati on a re marked by vert ica l lin es in t his
and all othe r figures of particle co unter resp on ses. The parti cle counts signa l of
fi g. 6.13 (b) is a n underlying signal in all othe r parti cle counte r resp onses p(t )

given in t his section .

The case give n in fig. 6.14 is assumed in section 6.4.1.3 (ass um ption 3 on
page 141 ). There is a n ex p one ntia lly decr easing dyn a mi c she a r-off funct ion (t.) ill

fi g. 6.14( a) which leads t o a particle counter respon se p (t) as given in fi g. 6.14(b ).

T his response is monot onicly decr easing aft er t he ela pse of one t ravel t ime.

T he on ly p ossi bili ty of a m axi mum va lue la t er t ha n one t rav I t imr 1. g lv n

1Il fi g. 6.15: a sh ar-off fun cti on s(t) t ha t is itself not monot onic a nd .onta ins
55
a m aximum va lue. In t his case S( ) = 5 2 +0. 55 +0 .025 or (t) = 11.11 - 0 .05/

15
11.11e- o.5t .

Figs. 6.16-6.18 illustrate the impact of various initial biofilm shedding profiles
(BSP) on a system with exponential shear-off function as given in fig. 6.1-t(a)
(thus, assumption 3 is fulfilled). Investigated are a decreasing BSP (fig. 6.16).
an increasing BSP (fig. 6.17) and a BSP with maximum value at half distance
(fig. 6.18).
From the particle counter responses p(t) given in figs. 6.16(b). 6.1/(b) and
6.18(b), it is seen that the BSP has a significant impact on the particle counts
until one travel time is over. However, it is has no influence on the shape of
the particle counts response after the elapse of one travel time. Compare also
fig. 6.14(b), where a flat biofilm profile with the same shear-off function as in
figures 6.16-6.18 is used. From these four examples it is seen that a wide range of
shapes of particle counts within the time less than one travel time may originate
from the biofilm shedding profile, but that the shape of the particle counts after
one travel time is determined by the shear-off function alone.

6.5 Parameter Identification

The problem to find the parameters of the system 5(8) is the standard problem of
system identification. The standard solution would be to use a regressor composed
of both input and output data to find the parameters. However, the present case
is different, since there is hardly any knowledge of the input b* ('19). The only
known circumstance concerning the input is its start (at '19 = 0) and end (at
'19 = T), i.e. it is known that a physically feasible input has to be zero for '19 <0
and '19 > T.
Since the output signal p(t) extends for a longer time (the step flow trials
generally finish after 3 to 4 travel times T), this signal of p(t) for t > T can
be used to gain information about the parameters of the system. The output
response has for t > T the shape of an impulse response (as will be shown in
section 6.5.2) independent of the input signal. It is, however. not possible to
read all parameters directly from p(t) for t > T. An indirect method. which

155
involves the Laplace back-transform and a number of integrals, is employed in
section 6.5.3 to find the remaining unknown parameters.
Since the magnitude of both parameters and input are unknown, they cannot
be determined completely. There will be always an unknown factor possible that.
if it is multiplied with the input and the system is divided by it. will lean' the
output unchanged. Therefore, the input will only show the shape of the biofilm
shedding profile and not its absolute magnitude.
However, before the task of identifying the parameters of the Slow Signal p( t)
can be attempted, it is necessary to address the problem of the other signals given
in section 6.2, the 'face-value' signals. The Occasional Particle Step (spec. 5) is
readily removed by subtracting a constant value from all high flow values of the
data of sample point 1 and the data of SP 2 and 3 in the corresponding time
range. The magnitude of the particle step is determined by the step decrease of
particle counts at the end of a SFT. This procedure will not be discussed further.
The Peaks (spec. 4 in section 6.2), however, require additional analysis. They
will be considered in the following section.

6.5.1 Data Preparation (Peak Erase)

The goal of this section is to show methods for the removal of peaks from the
particle counter data. Two possible procedures for achieving a signal without
peaks were implemented: one manual, the other automated, based on classical
image processing. The procedures of this section are termed 'peak erase'. since
the peaks are supposed to be eliminated.

6.5.1.1 Peak Erase by Hand

This fairly crude approach allows the user to identify the peaks-to-be-erased by
hand (i.e. mouse click) in a graph. Only the final peak (originating from t he
inlet) is dealt with separately (cf. section 6.5.4.1).

156
6.5.1.2 Automatic Peak Detect and Erase

This method is based on an approach of classical image processing (cf. e.g.

section 7.1 in [56]). It investigates an approximation of the derivative to detect
the presence of peaks. The algorithm for peak detection consists of two major
steps:

a) Edge detect

b) Edge clustering

a) Edge detection investigates an approximation of the derivative of the total

particle counts p(t), i.e. :tP(t). In this implementation the backward difference
approximation is used. Thus,

i (t) ~ p(t) - p(t - ~t)

dtP ~t'

where ~t is the sampling time (either 2 min or 5 min), cf. [4]. A different method
(e.g. Tustin's method) could be used as well.
An edge is defined as a point were the derivative is higher than a certain,
user defined, threshold (). Depending on the sign of the derivative, positive and
negative edges are distinguished. Thus, an edge is identified via

p(t) - p(t - ~t) ()

~t > for a positive edge, (6.16)

p(t) - p(t - ~t) < _() for a negative edge. (6.17)

~t

b) Edge clustering. It is more of an issue to find peaks (compared to edges),

which on the one hand obviously consist of a positive and a negative edge and
on the other hand may include some regions with smaller derivative or opposite
derivative due to random errors.
Therefore, a procedure was included in the algorithm which clusters edges
together to form peaks by starting with a positive edge and searching for the
next negative edge. Another parameter which can be defined bv the user. the

157
look-ahead, defines how many "false" derivatives are tolerated before the edge is
discarded as not belonging to a peak.
The two design parameters, look-ahead and threshold, have to be chosen
sensibly. Especially the threshold is very critical and some reality checks are
incorporated in the Matlab implementation bioprofi to check if the threshold
is too large or too small. In particular the peaks are tested for excessive peak
width and the presence of the inlet peak is checked.

6.5.1.3 Inclusion of Peak Models

Fundamentally, there are two possible way to deal with the shape of the peaks:
to consider it completely unknown and thus only rely on the derivative (or the
magnitude of the edge) or to assume that the peaks have basically the same shape
as the Slow Signal but there are not enough samples to reconstruct it accurately
(as discussed in section 6.4.1.1). The latter implies that the declining edge of the
peak follows an exponential function. The time constant (or decay coefficient) of
this exponential will be different to that of the Slow Signal. It will be a quicker
decay (smaller time constant or higher decay coefficient). This difference in time
constant can be attributed to the fact that these particles are loose deposits in
this case which are lifted into suspension much quicker than attached material.
If this chain of thought is followed it becomes apparent that the l'v1.V. can be
dealt with in various ways. The peaks before the M.V. do not need to be consid-
ered here, since the fact that the Slow Signal is then increasing, gives a fairly clear
picture of the end of the peaks. However, for the inlet peak (which falls together
with the M.V.) it opens the possibility to fit a sum of two exponential functions
(with fast and slow decay) to the Slow Signal after the M.V. (see section 6.5"'!.1).
Thus, in the Matlab implementation of this work, bioprofi. it is possible to opt
for this way of fitting the parameters.

158
6.5.2 Time Constant Ta == !a

The knowledge of the end of the biofilm shedding profile (together with s(t) =
ofor t < 0) allows us to investigate eq. (6.9) for t > T without the step function
1(t - 'l9). This is given as

p(t) = v 1:0 b'(1J) (k1(f)e-a(H) + k2(f)) d1J

ve- 1:0
at
b'(1J)k 1(f)e
ad
d1J + v 1:0 b'(1J)k2(f) d1J (6.18)
p(t) cl(T)e- at + c2(T), for t > T, (6.19)

or

ci(T)
~
p(t) = Cl (T)e-
aT e-a(t-T) + c2(T), for t > T. (6.20)

Eq. (6.20) is the equation of an exponential function starting at t - T = O. It

is straightforward to fit this exponential function to the data p(t) for t > T in
order to obtain the system parameter a and two other parameters Cl (T), c2(T)
which depend on T and are related to the wanted parameters k 1 (f), k 2 (f ) in the
following way:

cl(T) uk,(f) 1:0 b'(1J)e

ad
d1J (6.21)

vk2(f) Lo b'(1J) d1J (6.22)

Note that, rather than obtaining Cl (T), the result of the fitting will be

(6.23)

as indicated in eq. (6.20). From this follows obviously cl(T) = c~(T)eaT.

The parameter a is the inverse of the time constant T a = ~ which is frequent lv
used as a measure of how quickly an exponential signal decavs.

159
6.5.3 Parameters k 1 (f ) and k 2 (f )

The aim of finding k 1 (f ), k 2 (f ) using eq. (6.21) and (6.22) is only partially
achieved, since it is impossible to determine k 1 (f ) and k 2 (f ) from cl(T).C2(T)
without knowledge of the two integrals over b* ('l9).
For a flat biofilm shedding profile

b'(19) ={ ~ if 0 < 'l9 < T

otherwise
(6.24)

k 1 (f) and k 2 (f ) can be found up to an unknown constant C according to the

following equations (derived from eq. (6.21), (6.22) ).

cl(T)
v~c (eaT - 1)'
c2(T)
cvT

However, it is unlikely that the biofilm shedding profile is flat. Therefore these
values are only used for initial conditions and to check if the simulation results are
theoretically correct. A model with parameters kin«, k 2fl at and a will produce the
same exponentially decaying signal for t > T as is exhibited in the total particle
counts data.
To allow for any shape of biofilm shedding profile b" ('l9), it is necessary to
determine the integrals over the biofilm shedding profile through known signals.
For this purpose the inverse of the Laplace transform 8 (s) of the shear-off function
s(t), i.e. 8- 1 (s), is used. The Laplace back-transform of 8- 1 (s) is denoted by
Sinv(t).
Thus, the problem is to find (cf. eq. (6.21), (6.22) )

cl(T)
k1 (f) (6.25)
v JOT b: ('l9)e a t? d'l9 '
c2(T)
k2(f ) (6.26)
l' JoT b: ('l9) d'l9

160
with

Iii)
b*('!J) = -
v 0
p(t)Sinv('!J - t) dt. (6.:27)

Using the inverse of the Laplace transform .c- 1

{- } . Sinl,(t) is obtained to be

(6.28)

In summary, it is necessary to find k1(f), k 2(f) from eq. (6.25). (6.26) given
eq. (6.27) and (6.28). Since eq. (6.28) includes k1(f). k 2(f). non-linear optimisa-
tion methods will be used to solve this problem.
The calculation of the inverse Laplace transform is given in appendix B. The
inverse shear-off function is obtained to be

a2k1(f)k2(f) - ak2(f) t ak1(f)

Sinv(t) = - (k1(f) + k2(f)pekl(f)+k2(f) + (k1(f) + k 2(f))2 6(t ) +

+ k,(f) ~k 2(f)
o'(t), (6.29)

where 6(t) is Dirac's delta impulse! and 6'(t) its derivative.

This result has to be incorporated in eq. (6.27), and then used in eqs. (6.25)
and (6.26) (or eqs. (6.21) and (6.22) ). Thus, with the abbreviation

IThe Dirac impulse is not a function in the normal sense. It is. however, a generalised
function or distribution, cf. [70, 79]

161
eqs. (6.21) and (6.22) can be written as

A few steps of calculations (provided in appendix B) give the following result.

By approximating the integrals numerically it is possible to obtain k 1 (f) and

k 2 (f ) from eq. (6.31) given p(.).

However, note that the given system of equations, eq. (6.31), is degenerate
with respect to k 1 (f) and k 2 (f ), i.e. it is only possible to find a solution for
a ratio k = :~~~~' and not for each parameter k 1(f), k 2 (f). To show this let
k1(f) = kk 2(f) in eq. (6.31). Thus,

(6.32)

(6.33)

This shows again that it is impossible to fully determine both input and param-
eters of the system in fig. 6.12 from the output and the system structure. In the
Matlab implementation, k 2 (f ) is set to 1, which means effectivclv that k l (f) is

162
replaced with :~~~~ and k 2 (f ) with ::~~~ in equations which feature the true val-
ues of k1 (f) and k 2 (f )· Therefore, the biofilm shedding profile is found only up
to an unknown constant, say c. The algorithm used is provided by Xlatlab (i.e.
fmin). It is a line search method based on golden section search and parabolic
interpolation.

To check on the accuracy of the results for k1(f), k 2(f) the following values d
and c" can be calculated. For this test, k1(f) and k 2(f) (obtained as given above)
are used to first calculate the BSP b" (iJ).

cl(T)
C' - k 2(f) . (6.3-1)
v JoT b* (iJ)k 1(f)e a{) diJ'
c2(T)
c" k 2(f) . (6.35)
v JoT b: (iJ)k 2(f) diJ '

Comparison of the above equations with eqs. (6.25), (6.26) shows that k1(f), k2(f)
were replaced by :~~~~ and ::~~~. In the implementation, a sum approximation is
used instead of the integral of eqs. (6.34) and (6.35).
Eqs. (6.34), (6.35) should lead to

c' ~ c" ~ c, (6.36)

as can be seen by comparison with eqs. (6.25) and (6.26). However, since data is
generally noisy, eq. (6.36) will never be fully achieved. Therefore, the difference
between c' and c" gives an indication on the "goodness" of the results.
Note that it seems possible to determine k 2 (f ) from eqs. (6.34) and (6.35),
since c ~ k 2 (f) should hold. However, it is not certain if the biofilm shedding
profile b*(iJ) is not scaled and therefore it is meaningless to determine k2 (f ) this
way. It is only correct if b* (iJ) is also correct in magnitude, and it is impossible
to know that. Despite that, the relevance (or "goodness") of the results will not
be altered as long as c' ~ C"(~ c) holds.

163
 Bio fil m Shedding Profile for Step Flow T rial o. 6
Biofilm Shedding Profile for Step Flo" Tri al . ' 0 . 6
c"" ~
=
8'" 0.8 SP 1
:l
8 0.8 SP 1
SP2 ~
.= SP 2
SP 3 t:
2.. 0.6 SP 3
§
E 0.4
o
..t:
~

e
'-= 0.2 .'
c,

,. ~ - '. . : "'"
= :.. . .
~ 0 ~~__flWj"'''''~~.~'
"""
.c
E E
~-0 . 2 ~ -o . 2
o o
:0 :0
-3000 - 2500 -2000 - 1500 - 1000 - 500 0 500 1000 - 3000 - 2500 - 2000 - I500 -1000 -500 0 500 1000
distan ce x fro m pipe inlet in m distan ce x fro m pipe inlet in m

(a) Negative edge t hreshold is t he sa me (b) Negative edg e t hres ho ld at 40% of

as t he positive ed ge t hreshold . positive edg e t hreshol d .

Figure 6.19: Effect of the reduction of the threshold f or the negative edge in th
peak detect algorithm on the apparent initial biofilm shedding profile of step ji ou
trial 6, 8 October 1996.

6.5.4 Issues in Parameter Determination

6.5.4.1 Impact of Peak Erase

Accurate and effective erasing of t he peak s is crucial for t he parameter estimation

process. In particul ar , if at any t ime t he decrease in t he signal is sharpe r t han <at

t he biofilm shedding profile (BSP) po ssibly becomes negat ive. This is espec ially
relevant during t he (er ased ) peak s, since t he signal duri ng t hese t imes is unknown
and , for want of anyt hing mor e accurate, repl aced by a linear inter polati on.
As already indica t ed in section 6.5.1. 3 on page 158, t here is a degree of freedom
in dealing with t he last peak in t he data , whi ch originat es from t he inlet (t he p ak
at t he IVI.V .). This peak is assumed to be of t he same sha pe as t he Slow Signa l,
however , with a greater decay coeffic ient in t he expo nentia l tail. To red uce t he
impact of t he p eak on t he est imation of a, t he curre nt im plement at ion fit. a
sum of two expone ntial fun cti on s and t hen discards t he higher decay coe ffic ient
(sho rte r t ime constant) . T he merit of t he high er decay coeffic ient a a number
is quest ionable, since t he sam pling rat e is not high nou gh to obtain an ac ur at e

pict ure of t he pea ks.

164
 This has, however, the unfortunate side effect that the peak as found by fitting
may be wider than the one found by peak erase based on the magnitude of edges.
Therefore, the threshold () in eq. 6.17 (i.e. for the negative edge) is modified to be
less than the threshold for positive edges (as in eq. 6.16) to widen the detected
peaks. Thus, the ratio of thresholds emerges as an additional design parameter.
The effect of this parameter is illustrated in figure 6.19. In that figure the
BSP b(x) of SFT 6 is depicted on the ordinate. On the abscissa the distance .1'

from the pipe rig inlet is shown. Particles at a negative distance (i.e. originating
from the pipe work in front of the inlet) should make no contribution to the
BSP; a zero-valued BSP for x < 0 indicates that the exponential tail of the
Slow Signal matches the data perfectly. The variation of the BSP for .1' < 0

gives therefore an indication of the error associated with the fit of the model.
However, the "drop off" towards negative values between -500 m and 0 m that
can be seen in fig. 6.19(a) indicates a quite serious mismatch between data and
model. It originates from the incomplete removal of the inlet peak when equal
thresholds for positive and negative edges are used. The effect of a reduction of
the threshold for negative edges to 40 % of the threshold for positive edges is
shown in fig. 6.19(b): the negative "drop off" is almost entirely removed. In tho
implementation a negative edge threshold reduced to 80 % of the positive edge
threshold is used as the default setting.

6.5.4.2 Normalisation of Data for Simulation

All theoretical work assumes a start of the signal at p(O) = 0 for t = O. However,
the particle counts are not zero when an arbitrary increase of flow takes place
(i.e at the start of the SFT). Therefore the data are normalised. This implies the
assumption of a constant feed of particles going into the rig from the balancing
tank, the same even after the increase of flow. The modelled particle counts
response is independent of this constant input. For simplicity. the initial value
of the particle counts at the start of the step flow trial is used as a reference for
normalisation.
However, since the final value gives some information about the steady st ate

165
biofilm shedding for the respective regime, the final value is normalised not just
with respect to the initial value but, for higher accuracy. with respect to a mean
of particle counts during a period of 4.8 hrs before the start of the SFT.

6.6 The Initial Biofilm Shedding Profile

To obtain the initial biofilm shedding profile (BSP) the system as described in
section 6.4.2 is considered. It has the following transfer function

P(s) kl (f) + k2 (f)) . v

B*(s) ( s+a s
s(k1(f) + k 2(f)) + ak2(f)
·v (6.37)
s(s+a) .

To obtain the initial biofilm shedding profile b(t) from the total particle counts
p(t) it is necessary to simulate the inverse system

B*(s) s(s+a) 1
(6.38)
P(s) s(k1(f) + k2(f)) + ak2(f) v'

which is a non-causal (or improper) system (order of numerator greater than

order of denominator). Systems of this type cannot be brought into standard
state space form and thus are not readily simulated in common tools like Matlab.
They also exhibit a strong susceptibility to noise since the derivative action reacts
strongly to edges.
To address the problem of large variations in the BSP due to high frequency
noise in the particle counts a running average filter is used to smooth the data.
Such filters are frequently used for low pass filtering in applications of digital
image processing [56]. Briefly, the data values are successively replaced by the
average of the surrounding data. The number of data points to be included in
the average is determined by an additional design parameter. the filter window
size n = 2i + 1, where i is the number of points before and. for symmetry. after
the current data instant. In this implementation. the first i points of the BSP.
the final i points of the BSP and the 2i points surrounding the Maxirnum Value

166
 Bio til m Shed ding Pro file for Step Row T rial No.2 Biofilm Shed d ing Profi le for tep Flow Tn al :'\0. 2
~ O.-l l----~ _____;======::::=:;I
8= SP 1
~ 0.3
u
.i: SP _
~ 0.2
P3
§
g 0.1
...::
..,
~ 0
C-
Ol)

:5 - 0.1
"0
"
s:
E-0 .2
0;:
o
.s
- 3000 - 2000 - 1000 0 lOoo 2000 - 3000 - 2000 - 1000 0 I000
di sta nce x fro m pipe inlet in m distance )( from pipe inlet In m

(a ) Unfilter ed biofilm sh edding profile. (b) Fil t er ed biofilm shedding profil e.

Figure 6.20: Unfiltered and filt ered apparent initia l biofil m shedding profit oj
ste p flo w trial 2, 1 August 1996. Filt ered data us es a run n ing averag e fi lt er with
window size 19.

in t he particle counts response (at x = 0 in t he BSP ) are not fil tered .

The use of t his filt er is illu stra t ed in fig. 6.20. T here t he BSP b(l;) of FT 2
is depi ct ed. In fig. 6.20(a) t he unfil t er ed signal is presen ted , whereas fi g. 6.20 (b )
shows t he BSP after filt ering with a running average fil ter of window size 11. =
19. In t his case a noi se level in t he unfil ter ed resul ts t hat is of t he sa me orde r
of magnitude as t he BSP is ob served , however , filtering reduces t he noise level
significa ntly while t he BSP signal rem ains intact.
The actual sim ulat ion of t he improp er sys te m was per formed by uSlllg an
a pp roximation of the derivati ve of p(t) as inpu t ra ther t han t he act ua l signa l.
Again (as in section 6.5.1.2) a backward differ ence approximation was used . Wi th
t he introduction of t his approximation , t he or der of t he numer a tor i red uced by
one com pa red to eq. (6.38) and t he rem aining system :;~;~ is t herefore prop er.
Fo r t he simula t ion of t he system b"('13) whi ch dep ends on () was us d , a
introduced in eq . (6.8) , however , now b(x ) is of int erest. Th e replacem nt of {)
in b*('I3 ) with vT - v'13 finall y yields t he ap parent ini ti al biofilm heckling pr ofile

b( - v'13 + vT ) = b(VT) = b(..).

167
 Table 6.1: Thresholds for peak erase and filter window sizes.

SFT Thresholds in particles/(ml·day) Filter

nQ Inlet SP 1 SP 2 SP 3 window size
1 nla 200000 150000 100000 3
2 nla 4000 4000 3000 19
3 nla 8000 10000 10000 19
4 nla 30000 30000 30000 15
5 nla 10000 10000 10000 13
6 100 30000 30000 10000 11
7 100 30000 10000 10000 11
8 100 18000 10000 10000 15
9 100 12000 10000 7000 15

6.7 Results
In previous sections, various design parameters that introduce degrees of freedom
into the identification process of the parameters of the particle counts model
were defined. In particular, there are the look-ahead and the threshold () of the
peak erase (section 6.5.1.2), the reduction of this threshold for the negative edge
as introduced in section 6.5.4.1 and the filter window size (cf. section 6.6). In
addition there is the option to include a model of the inlet peak as discussed in
section 6.5.1.3.
The thresholds used to obtain the results presented in this section are given
in columns 2-5 of table 6.1, as look-ahead either 2 data points (for 8FTs with
sampling time 5 min) or 5 data points if the sampling time is 2 min were used.
The reduction of the threshold for a negative edge is always 80% apart from
SFTs 1 and 6, where it is 50% and 40%, respectively. The filter window sizes
are also provided in table 6.1, in column 5. The peaks of SFTs 1-3 and 6 are
modelled.
Finally, due to the difference between basic and extended specifications as
discussed in section 6.4.1.4, not always all of the data at high flow of a 8FT are
consider, but the data used for identification is restricted to the range for which
the basic specifications appear to be valid. These time ranges for SFTs 1-9 are
presented in table 6.2.

168
 Table 6.2: Investigated time periods of high flow for SFTs 1-9.

SFT Date Sample Point 1 Sample Point 2 Sample Point 3

nQ Start End Start End Start End
1 25.7.96 6:55 20:20 6:55 20:20 6:55 20:20
2 1.8.96 6:40 20:10 6:40 20:10 6:40 20:10
3 7.8.96 7:30 15:35 7:25 17:30 7:25 19:10
4 22.8.96 6:55 18:40 6:55 18:40 6:55 18:40
5 28.8.96 7:10 16:45 7:05 18:10 7:05 13:50
6 8.10.96 6:50 17:40 6:50 17:40 6:50 17:40
7 15.10.96 7:05 10:50 7:05 12:00 7:05 13:10
8 24.10.96 6:40 10:30 6:45 11:45 6:40 12:50
9 30.10.96 7:15 10:45 6:45 12:10 6:45 13:15

Table 6.3: Data of model parameters a and k, (f) for step flow trials 1 to 9,
k 2 (f ) is always set to 1 due to insufficient knowledge.

SFT Ta = Ta in hours k1
nQ Inlet SP 1 SP 2 SP 3 SP 1 SP 2 SP 3
1 no data 1.69 2.00 1.87 104.48 60.20 32.31
2 no data 4.18 1.89 2.61 1.24 3.53 3.47
3 no data 2.63 2.31 2.16 2.29 5.96 11.88
4 no data 0.43 1.41 1.18 3.01 5.08 5.05
5 no data 2.84 2.73 1.50 11.05 4.59 3.17
6 0.73 2.10 2.12 2.54 13.03 7.77 4.22
7 0.36 1.48 0.84 3.90 1.33 1.57 3.60
8 0.18 1.42 0.54 0.66 2.60 2.70 1.94
9 0.33 0.29 0.64 0.62 1.86 2.18 2.60

6.7.1 Parameters

Table 6.3 gives the parameters of the particle shear-off models (as presented in
section 6.6) obtained from step flow trials 1 to 9. Note that SFTs 1-3 are flow
increases from 0.08 lis to 0.6 1/s and SFTs 4+5 from 0.08 lis to 0.9 lis. all five
with no chlorine going into the rig. SFTs 6+7 are with chlorine and step from
0.08 lis to 0.9 1/s, whereas SFTs 8+9 (again with chlorine) are flow increases

from 0.3 lis to 0.9 lis.

The time constants Ta = ~ of the exponential docav give the time it took

to decrease the amount of sheared-off material to about 37 Cj(, of its start value

169
Table 6.4: Statistical analysis of data concerning time constants Ta . All means
and standard deviations are given in hours.

SFT Sample Sample Standard Minimum Standard

nQ Size Mean Deviation Deviation a min
t; a for Normal Distribution
1 3 1.85 0.153 0.088
2 3 2.89 1.171 0.676
3 3 2.37 0.236 0.136
4 3 1.00 0.514 0.297
5 3 2.36 0.742 0.428
6 4 1.87 0.786 0.487
7 4 1.64 1.571 0.974
8 4 0.70 0.522 0.324
9 4 0.47 0.184 0.114

(both times added to the final value). Thus, a small Ta indicates quick decrease
in sheared-off material. This model is based on the assumption that Ta (or a) is
the same throughout the pipe rig. This implies that the variation in each row of
Ta-data in table 6.3 is purely random.
The smallest time constant is Ta = 0.29 h, the largest is Ta = 4.18 h. both
are found in sample point 1 data. The time constants found from inlet data are
generally lower than the values found for SP 1 to 3. All values are within the same
order of magnitude. Note that the small time constant of the data originating
from SP 1 at SFT 4 may be due to an increase of inlet particle counts during
SFT 4.
Some statistical analysis of the time constant data is provided in table 6.4.
The mean of the time constants 'I'; of the three or four sample points is given in
column 3. The sample standard deviation a is provided in column 4 of that table,
the minimum standard deviation for the data of Ta to be normally distributed
is given in column 5. These values are calculated via eq. (5.2) on page 99 using
a sample size of n=3 for SFTs 1-5 and n=4 for SFTs 6-9 (see column 2). The
used X
2
statistic is then either X~.05,2 = 6.0 or X~.05,3 = 7.8 (see [18, p.21]). These
sample sizes are too small for any strong statistically founded statements. thus.
the results have to be taken as an indication only. From these data it is seen that

170
 Table 6.5: Approximation constants d and c" during step flow trials 1 to 9.

8FT 8P 1 8P 2 8P 3
nQ c' c" c' c" c' d'
1 0.2752 0.3615 0.2757 0.2756 0.2755 0.2757
2 0.2669 0.2664 0.2693 0.2675 0.2655 0.2697
3 0.2743 0.2711 0.2727 0.2776 0.3830 0.3372
4 0.1828 0.1926 0.1735 0.1919 0.1782 0.1780
5 0.1684 0.2469 0.1732 0.1845 0.1713 0.1861
6 0.1846 0.1953 0.1843 0.1912 0.1830 0.1853
7 0.1644 0.1775 0.1657 0.1761 0.1759 0.1835
8 0.1789 0.1729 0.1564 0.1712 0.1595 0.1751
9 0.2022 0.1890 0.1664 0.1915 0.1682 0.1823

especially for 8FTs 2 and 7 a high standard deviation would be needed for this
data to be normally distributed, all other 8FTs may be normally distributed with
standard deviations of less than 0.5 hours. This indicates that the time constants
generally agree, including the results from inlet data.
The other coefficients given in the previously mentioned table 6.3 are the k 1 (f)
of section 6.6. It is impossible under in the given circumstances to determine all
three k 1 (f), k2 (f ) and the biofilm shedding profile b(x) completely, but it is
possible to obtain them up to a common factor c. Therefore, k 2 (f ) is chosen to
be 1 for all step flow trials. The values of k 1 (f) may differ from sample point to
sample point reflecting the different sensitivities of the particle counters, however.
if all particle counters were counting the same numbers for the same sample of
water, all k, (f) would have been the same.
Table 6.5 gives the coefficients c' and e" of section 6.5.3. These coefficients
should be equal, i.e. e' = e" = c should hold. Thus, these values give an indication
on the goodness of the identification procedure and the appropriateness of the
model for a particular 8FT. From the table it is seen that the values agree well.
Finally, table 6.6 gives the final values of total particle counts at high flow
(during the step flow trial) together with information on the inlet flow and inlet
total chlorine regime. The values presented in columns 4-6 of table 6.6 are the
difference between the final value of the particle counts response (to an increase in

171
Table 6.6: Final values of total particle counts at high flow during step flow
trials 1 to 9.

SFT Flow Total Chlorine Final Value of p(t)

nQ in lis in mg/l SP 1 SP 2 SP 3
1 0.6 none 49.9 59.3 24.0
2 0.6 none 39.1 19.0 17.6
3 0.6 none 12.3 20.7 37.2
4 0.9 none 42.8 19.0 21.0
5 0.9 none 11.8 15.8 14.0
6 0.9 0.4 14.5 2.7 2.2
7 0.9 0.4 5.5 0.4 7.9
8 0.9 0.4 9.6 9.9 6.4
9 0.9 0.4 8.9 4.5 6.1

flow) and the average of the pre-SFT particle counts after any occasional particle
step has been removed. These values are quite important, because they give an
indication of the combined net effect of the shear-off function s (t) and the biofilm
shedding profile b(x).
The results presented in table 6.6 indicate that the trend in final values is
generally downwards; later SFTs have frequently lower final values. In particular,
large final values are seen at the first SFT, values of more that 10 particles per
ml at SFTs 2 to 5 and final values predominately below 10 particles at SFTs 6-
9. There does not seem to be a relationship with flow rate, however, with the
introduction of total chlorine as disinfectant into the rig, the final values become
generally smaller. However, these variations are not very pronounced.

6.7.2 Biofilm Shedding Profile

The apparent initial biofilm shedding profiles (BSP) of step flow trials 1 to 9
are presented in figs. 6.21 to 6.29. If appropriate, the data was filtered wit h a
running average filter of the window size given in the captions. All sctt ings of
the design parameters are provided earlier in this section (see tables 6.1 and 6.2).
Each graph shows the BSP as found from particle counts measured at sample
point 1 (SP 1), sample point 2 (SP 2) and sample point 3 (SP 3). These curves

172
 Biofilm Shedding Profile for Step Flow Trial o. 1
~ 0.25 I'-------r-------,,-------r---.---...-------~____,
c::
::3 -.
oo --
'.
Q) SP 1 --
~ 0.2 SP2 --
.~ --
--
--
0.. SP 3
-.... 0 15
~
O. . .:-
--
--
·---
.... f-
.
: ; .

--:
:- -- ---
. -: :
·-
8o : ::: ·-
-· --
tl:: : -: ~ . -- --
--
-o 0.11- ':' :- ',: ' .
t+:: : - ·- .-
o . : 'w ': : ~i
~ ~ : : _' I - 10_ :-
0.. I ' - !' t .:

;r~ :
r\1 .. .. ',f i -.. .~ .. . Ij--~
gfO 05 : · -
~. ' : "' ~ ..,..~ " ' '' '' '' '' ' .. .. .. -

Q) . ' _ • , :~ t, :
: . :: -- .'

..c: 'I : -- IJ ~ h " " "~ I :' :: :: ·-- ...

~
. r' I ,.... : -. , t· ~ - ,-
V'J
· . -- I " ., ."" ' . : /- ,
. .- -,-
_8 o .u ..~ ..........:. \ 'j ' " . ' '., ~.'J -: '=' ,'
"
.-' .
!..§ J : i i -.,..':
....... , ~ -'
.D
o 200 400 600 800 1000 1200
distance x from pipe inlet in m

Figure 6.21: Apparent initi al biofilm shedding profile obtained from total particle
coun ts of step flo w trial 1, 25 July 1996. Filter window size is three negative edge
threshold is 50% of positive edge threshold.

should id eally overlap. Negative di stances as shown in fi gures 6. 19 or 6.20 were

excl uded from t he figures of t his section , sim ilarly negati ve BSP values (be low

-0 .03 ) were crop ped as well. Negative va lues of t he BSP are phys ica lly im possible

a nd , therefore, indicate data t hat do es not fi t t he m odel. This m ay be d ue t o

noise in t he particle counts or b ecause of an in complet e peak erase . T he la t t er

will in troduce negative or positi ve spikes in t he vicinities of x = 0 m x :::::: 450 m ,

x :::::: 900 m and at t he end of t he BSP.

T he va lues of t he BSP give n on t he ord inate have to be sca led to ca rr y a ny

direct physical meaning. T his unknown scaling fact or wa a ll' ad y introduc d at

t he end of sectio n 6. 5.3. Thus t he BSP m agni tudes are mea ningfu l in com pa ri-

so n wi th other va lu s of t h sa me BSP, b ut it i not po ibk to r ad ily o m pa ro

num ri al va lu s t hat originat from diffr I' nt SFT . Note th at t he biofilm . hod-

173
 Biofilm Shedding Profile for Step Flow Trial o. 2
....c::
~

::;j 0.16 I-
o
~
u
0.14 .... .
SP 1
.-....
o
l-<
z:
-
--
--- SP 2
[ 0.12 .-..
..
1:-: .
SP 3
...... .---
ro
o.... 0.1 -. . ~

·-
'.

~
··
8 · . 11
o 0.08 ~.
,.

~ -
~ t

-
I
· -.
......
~
(!)
. 0.06
I:
r· .:.....~
I
I
., ' . ·st:
ol-< I r-- .
· I I
. :
-I
I-
i ·· .I
~
ft. ·. ···· ,!
~

0.. ....'.. . .
0.04 i -' .

.-e
0.0

""'0 0.02 .
·
..·
· I
··- I
,. ~ .
I · · ··
,: "". i :
i ·· ·· : ;\:
- ~: · . . :
I
I

,·-· i , ,
I
I·
- .:
.
- .-
.
- . .-
'. .
- .
""'0
(!)
..c:
1..
•
t=. \=1 ··., \-.!.,
J ., -- , '\

., .·· / ' !t. ,I,

\ , I
p
...\ I ·
. ·
I .' ..
,

..
. ,
:
.. .. --.-..
-.-
,
'
\1\. o,l " " '1" t ,
:\ "" ." ·
or1. .~
00
~
'- . .. .. -

- ..' I
8
...... \
I
.
t;- -0.02 .' .
I

t
' " .-

..D I ;
"
•
I

o 200 400 600 800 1000 1200

distance x from pipe inlet in m

Figure 6.22: A pparent initial biofilm shedding profile obtain ed from total particle
counts of step flow trial 2, 1 August 1996. Filter win dow size is 19.

ding profi le is not a biofilm t hickn ess profile. It is t he amount of biofilm t hat is
ini ti all y sheared-off from t he pip e wa ll (up to an unknown factor ) and as such
shows t he origins of high particle counts or t he po tential for shear-off over t he
distance of t he rig.
T he BSP as found from sample point 3 during SFT 3 (fig. 6.23 ) is not provided
since particle counter 3 did not wor k properly d uring t he first hours of t hat ste p
flow trial. The high BSP val ues between 400 m and 750 m of SF T 4 in fi g. 6.24 are
not com pletely reliable since t hey m ight be partially due to t he unu su al iner a. c

of in let particle counts during t hat SFT.

In t he first t hree SFTs (figs. 6.21 to 6.23 ) t he BSP is ini ti all y vr rv high a nd
decr eases with a la rge grad ient within t he ini ti al 400 m . In t he. t hre FT a
high fl ow of 0.6 li s was u ed a nd t here was no chlorine pr e. ent . FT s -1 anel =-

174
 Biofilm Shedding Profile for Step Flow Trial o. 3
~ 0.35 r:-r-----.----.-------.----r---r--~___.
c::
:::3
o
U
o 0.3 ". . SP 1
......
.-.....
u SP 2
'"'"
[0.25
......
~
.....
o
8 0.2
o
~
~ 0.15
c.;::
o
'"'"
0..
eo 0.1
.-c::
"'0
"'0
I

]0.05 I .1- 1
I II
- .. II
~ I~
VJ

8
..
V - I I I ,II

',.~
I I I 'I I • '1
......
c.;:: 0 • iJ .. - - - II _
-- J'
-- "I""
I
I. "
.-..n
0 ~I , ,I
~ ;
0 200 400 600 800 1000 1200
distance x from pipe inlet in m

Figure 6.23: Apparent initial biofilm shedding profile obtain ed from total particle
counts of st ep flow trial 3, 7 August 1996. Filter win dow size is 19.

are t he last SFTs with no chlorine, however , t he higher flow rate was 0.9 li s.
From figs. 6.24 and 6.25 it is seen that t he BSP s of t hese SFTs ex hibit rela ti vely
la rge magnitudes a lso for larger distances from the pip e rig inlet (x > 500 m ).
However , as m entioned above, t he large counts of fi g. 6.24 (8 FT 4) migh t b

spu rious . Note t hat t his in crease in BSP for larger distances does not neces a rily

imply that more particles originate from t hese loca t ions com pared to 8 FT 1-3 ,
but it indicates that relative to t he biofilm shedding in t he first 400 m of -a h

BSP more particles are sh eared-off in the la t er sections of t he pi pe rig.

St ep flow t rial 6 (fig. 6.26 ) is t he first step fl ow t rial after chlorin 1. 1' -

introd uced in t o t he pip e rig. The high fl ow ra t e was 0.9 li s. T hi FT exhibit. ' a
close to co nsta nt BSP for t he first 300 -350 III a nd t hen a harp declin e t o sma ll

val ues.
 Biofilm Shedding Profile for Step Flow Trial o. 4
.....
v:l

=
g 3
o SP 1
......
Q)

.-~2.5
o
.. . ... . . .. . " ·. " .
: .
SP2
0.. SP 3
......
~
.....
.9 2
E
o
et:

..
. .. , ' " I
• • I I • • I
...
".

• • • ~ -: : " .... • .. •• ~ '. '• • ' . ;. , ~ . ' , I ; J • • " : •

1 . .. - - ,
..' ,''""
:: " I I

·
· : \ I'
\ .. "
-
.

'\ r
~ '- " _ , .'. , ',' .

a 200 400 600 800 1000 1200

distance x from pipe inlet in ill

Figure 6.24: Apparent initial biofilm shedding profile obtain ed from total particle
counts of step flo w trial 4, 22 A ugust 1996. Filter window size is 15.

SFTs 7-9 (figs. 6.27-6 .29) all had a high flow rate of 0.9 li s wit h total chlorin
pr esent in t he pip e rig. T he BSPs shown in figs. 6.27- 6.29 are distri bu ted ov I'

t he whole pipe rig with simila r magnit ud e, however t hey ar e very pat chy. Not
t hat t hese SFTs exhibit many t raits of t he extended model (cf. sectio n 6.4.1.4)
and t herefore t he application of t he basic model, as done in t his work, i. more
crit ical t han in other SFTs . T he dat a ran ge was restrict ed to a t ime span were
t he basic model seems appropriat e (cf. table 6.2), which hould all via te t hat

pr ob lem.
 Biofilm Shedding Profile for Step Flo w Tri al o. 5
~ 0.351'"-------r-------,~--.__--r___--.__--..,_____,
e
.- .
.- .
:::3
o
o
~
...... 0.3 SP 1
.-.....
o SP 2
[0.25
......
SP 3
ro
.....
o
E 0.2
o : I ~
J::= : I a:
~0.15 ..: .I"t ·. . ..' .....
1.C : I {
o
l-o
0..
; I
: Ii
I ::
:: :: : :; 11:
0.1
T:\f\!\i\...· -.'
.. : .. 1 I : .

-
b1)
s:::
."'0
"'0
]0.05 ..
"
.. I . .

.--
.

.
..
': :": f
... ... .
: -.- 1 I :

'"" ' 0',; "

I
"
~ ~~. r.
I .. :
..
- I I : .
.' -.
C/J .
" , v >:
E
ao 0_
.- - "' - , , " . ,, \ ~

.-
~
",I

o 200 400 600 800 1000 1200

distance x from pipe inlet in m

Figure 6.25: Apparent initial biofil m shedding profile obtained from totol ptirticle
counts of step fl ow trial 5, 28 August 1996. Filter wi n dow size is 13.

6.8 Discussion

T he ini ti al biofilm she d d ing profi le gives some ind icat ion ab ou t t he state of t he

sys t em b efo re t he step flow trial. It is a sna ps hot of she a red-o ff m a t erial at t he
start of t he 8FT. Thus , loca ti on s of t he d etachment of large numb ers of part i le

a re id entifi ed . T he rela ti on ship to biofi lm t hickn ess or cons iste ncy is in principl e

unknown . Possibilities include a linea r rela t ionshi p to biofilm t h ickne , a lin ea r

rela ti on ship to t he t hickness of one layer of biofilm (perha ps relat ed to t h impact

of t he t urb ulent flow) or a d epend en ce on t he struct ure of biofilm . e.g. a low r

she a r-off for biofi lm with many filam en t s.

A cha nge in biofilm t hickness would be expected with t he in trodu ct ion of

hl orin e. \1\ ith out any di sinfectan t in t he y t m it i likely that th cl plct ion
of nu tri nt is th e ovc rriding rea on for a reducti on of hi film t hickn ~ ver t IH'

177
 Biofilm Shedding Profile for Step Flow Tri al o. 6
en
......
§ 0.4
oo : I
,
..... I . . . . ..

\.
..

SP 1
~ 0.35
.-'"'"
.. _. . . .

......
~
\
SP 2
..
~
.
SP 3
- 0.3 .. . . .. . . . . . ~ . .~ .
0..
~
...... : "
""
20.25 .. . . ... . I .

6 :\ I I
o
t.l:: 0.2
:. I •
. .:'1 .. I . . . .' . , . ..
.~ I
.,

.-80.15
-
Q) :.
•
•
•
.. 11 .' ..
n u , I • •
.
4-4
•. . .,1.1.'1
0.. I ~' "
01)
. .
~ 0.1
.... .. . .

""0
Q)
..
~0.05
.. -, - .. -
" .
.-. .... -
-.-
-:
6 '1 ' . "
.
0' ,.,. :~ .
"-I ' "''''.;',
'
*' •
" . , . ., . .""-;'"'.
• : . . • " ' . ' ~
~ ' .~ . " ," I~.· ':': ,. ~.. ..-~ .." _ ~~ " ', , - ': " . ~
c.;:: 0 . . . . .. . . .. . . . '.,.. .... . . ,.
o
..D
o 200 400 600 800 1000 1200
distance x from pipe inlet in m

Figure 6.26: Apparent ini tial biofilm shedding profile obtain ed from total particle
counts of ste p fl ow tria l 6, 8 October 1996. Filter win dow size is 11, negative edge
threshold is 40% of positiv e edge threshold.

len gth of t he rig, however, wi th t he in troduction of monochl or a mi ne it is expected

to see some d estruction of bi ofilm in t he part of t he pip e rig t hat is reached by t he

di sinfect ant . SFT 6 holds a special p osi ti on in t his contex t, since it was performed

soon afte r t he re-introduction of m onochlorami ne and , t herefore, it may sho w a

p ipe ri g co nd itio n in between t hese two cases .

These changes of bi ofilm t h ickness sho uld effect t he BSP. As presen tsd 1Il

t he previous section , a var iation of t he shape of t he b iofil m she d d ing profil 1.

se n for different SFTs. T he first t h ree SFTs provide a good m a t ch with t he

expect a t ions of a d ecreasing amou nt of b iofilm wit h di st ance. FT 4 a nd 5 show

t h sam b ehaviou r , however t he grad ient i mailer . A m nti on ed ab ove. FT

hol ds a spe ial p osi ti on in c it is likelv t hat t he bi ofilm had not .yet rea ched a
~

st ady s t a t aft er t h in t roducti on of monochl oramin e. Th e init ial plateau th at

17
 Biofilm Shedding Profile for Step Flow Trial o. 7
....t::
tI:l

:::3
oo 1
SP 1
-
.~
Q)

o SP 2
SP 3
-........""
0..0.8
o
E
00.6
e.t
-
!.C
(])

o
I
I

aOA :, ...
I
I

.-
"'0
"'0
OJ)
t::
:,
I
.,
'\
,
,
I
I
I
., , I
]0.2 • • f •
:' ,," """ . , I
tI:l
~

..9-
i I
E I t I
~ a:: .', !:. :."'''. " ,_,~', . ' .,
.n 0 ~'-----------'L.L..&J.-----IILL-...LLL.L.a-.L.
' _' •••• '
-- -
---L-
-
I
;.;, '.:;, ,. . ' '# .
-.. I . "' ". -~. ". --
...l....-.L-_ _L-:_ _----.l.._---.J

o 200 400 600 800 1000 1200

distance x from pipe inlet in ill

Figure 6.27: Apparent initial biofilm shedding profile obtain ed from total particle
counts of step flow trial 7, 15 October 1996. Filter win dow size is 11.

is apparent in fi g. 6.26 matches t hat int er pr et ati on . However, t he following BSPs

(SFTs 7-9) should start wit h smaller values (where monochloramine disinfected
t he rig), reach a m aximum and t hen decli ne wit h declining availabili ty of nu tri ents
(like SFTs 1-5) . Bu t t hey exhibit fairl y large counts at severa l places t hrough th e
pipe rig and do not show any clear pat t ern , neither decreasing nor with maximum
value. T he reason for t his is likely to be t hat t he amo unt of biofilm t hat i 1 ft
on t he pipe surface is significantly less t han during t he first SFTs cl ue to t h
combin d effect of t he previous SFTs: t he introd uct ion of monochlor ami ne and
t h reduction of te m pe rat ure .
T he fina l values of p(t ) (t able 6.6) suppo rt t his concl usion. The. e valu "
whi ch r presen t a net resul t of sheddi ng, be orne mu h mall et toward lat or
SFT s. It is important to note t hat t h num rical value obt ained from pa rti -

179
 Biofilm Shedding Profile for Step Flow Tri al o. 8
~

§0.6
oo
(l)
SP 1
]0.5 SP2
~ SP 3
0..
......
~0.4
....
S
o
tb 0.3
I
(l)

tao , ~
~.
I-. - ~.
0..0.2 _: ... ·Ii·
co -- i-
.-
"0
t:
. ,- -~
11
i-
"0 ..
.... - .
]0.1 =.'~',' ~- . . ,, . . -.: 'I' : .:-- :
V'J .
s
...... . :
, : .. : ,.~ -,_:
' :.i···..-, ., .. ,....·~. ~ , 1 \
I "
'I : . _:
I I "" '.: ' .;:
- -, , . ,
.
, . ; ..: .... . ,.
.

. ........

.-
",... , "" #
1.C ~ ~I'
0 0 .. .
I'
.D
0 200 400 600 800 1000 1200
distance x from pipe inlet in m

Figure 6.28: Apparent ini tial biofilm shedding profile obtain ed from total par-tid e
counts of step fl ow trial 8, 24 October 1996. Filter window size is 15.

cle counts are su bject to measurem ent errors arising from variations in parti cle
shapes, as discu ssed in section 5.2. T he t ime const ant is not effected by t hi:
problem. This paramet er describes t he t ransient t ime behaviour of particles aft r
a fl ow increase wh er eas t he final value descri bes t he steady st ate . The t ran. ient
behaviour describes t he stre ngth of t he attachme nt of t he biofilm. Both , teady
state and t rans ient paramet ers provid e informati on on t he effect of t he cond i-
t ions (or regime) in t he pipe rig on t he amo unt of sheared-off ma t rial. 1Il t he
investi gation discussed here was dominated by relati vely sho rt 8FT (3- 4 t rav I
t imes a pproximately 14-48 hou rs), it i possible that t he final valu de cribes a
sho rt-term steady state and t hat it i modified in t he lon g-t rm .
T hes lon g- ter m dynam ics ar , howey 1', not part of t h parti cl ount III d 1
presen ted in t his chapter. T hi model i pr ed ominat ly onc rn d with th e t ran -

1 0
 Biofilm Shedding Profile for Step Flow Trial o. 9
~ 0. 6
c
::3
o
U SP 1
.-~ 0.5
t::
-- ---

,
I
--- SP ? .-

C'd
0.. SP 3
.. \
30.4 -- ~
....o :: I.
: :11
E -- - ~I
o -- -- I _.
~0.3 ."
--
--
~: Il • -
: I _I

.-oQ)
.........
4-<
I-t
• --
•- -
-
I
I :1

-I l
:J

0..0.2 ~; -I ~ -
- --- I
.- ,
OJ) [
c - --
I ----- t
""d . , .- I
-
-. -. II
t: ,
-- -
-~ I~
""d t
---
- r.: ..
]0.1 -\
rJ)
,:,:
-'
-- -
--- I -,
I
-
, :.
~ . - :-
- - - '. .
::: . -'--..
-- ".-- .
-- .-.-,. ~
. ;,-
-
,
t ... ...,. (' ( -, ' " ; ~ - -.,
.-bE
~

" ' ': --

......... _J •
" ~
...'- :: I ---.- \,\~ ~ ,'"
- I

~j 1-
... , "

.-
.D
0 -u-
I~
- :' : : I -
:, ..- ---
I
~ "

o 200 400 600 800 1000 1200

distance x fro m pipe inlet in m

Figure 6.29: Apparent initial biofilm shedding profile obtain ed from total particle
counts of step flow trial 9, 30 October 1996. Filter window size is 15.

sient response of particles to a step in flow. The dynamics involved in biofi lm

detachmen t are gene rally reduced to a lin ear t ime dep endency in t he litera-
t ure , as is discussed in the literat ure review in section 2.4.3 on page 17, cf.
[137, 115, 122 , 28, 61, 6] . However , in t he case conside red in t his t hesis t he
t ra nsient shear-off d id have an important impact on t he biofilm in t he pipe rig.
Not on ly does it dominate t he 10-20 hours of t he parti cle counts re pon . , but
it a lso reduces th e amount of biofi lm in t he pipe rig noti cea bly. T he expo nent ial
dynamics as used in this model are an impor tant part of t he new app roa ch of

t his t hes is.

Thus, t he model presen t ed in t his cha pte r was shown to be a u: eful tool for
und rstanding t he underl ying proce ses of biofilm detachm en t. a nd t he dvn.u nir s
of sus pende d parti le in t h pi pe rig. It paramet er a nd t he init ial biolilm

1 1
shedding profile give information about the condition of the system
,
and the likely
,

locations of the origin of high particles incidents. The measurement of biofilm

thickness in situ is currently still a major challenge (cf. section 2...1.5), with this
model, step flow trials and the initial biofilm shedding profile provide together an
indirect (and destructive) measure of the biofilm shedding potential which can
give valuable information about the condition of the distribution system.

6.9 Validation and Limitations

The validity of the parametric particle counts model developed in this chapter
cannot be proven, since its development is based on data, however, evidence of
the validity of the model can be shown.

• The inlet particle counter response matches the assumed shear-off function
s(t) (cf. section 6.4.1.7).

• The resulting Biofilm Shedding Profile has a meaningful interpretation .

• The errors between the exponential tail data and the fitted exponential
function are small.

A possible further validation of the model would be to install in-situ biofilm

sensors to monitor the thickness and, thus, the shear-off from biofilms. However,
this technology is not currently available. Another option would be to freeze and
cut out the pipes and analyse the amount of biofilm left on the surface. It would
be possible to obtain only sparse results using this method and although feasible
it is not very practical. The most practical way for improved validation are more
experiments of the kind reported in this thesis.
It is also important to be understand clearly the limitations of this model.

• The model is valid for a step flow input only.

• The parameters k 1 (I) and k 2 ( / ) depend on the magnitude of the flow step.

• It may be necessary to extend the basic model of the Slow Signal to match
data (see 'Extended Model of the Slow Signal.' section 6.-1.1.4).

182
 • Absolute values of different particle counters do not agree well (~l'(' ~t'('­
tion 5.2) and therefore it is necessary to fit k1(f) and k2(f) for each counter.

6.10 Implementation

:\11 calculations presented in this chapter are included in the Water Qualitv Tool-
box for Matlab (cf. appendix C).

The numerical statistical simulations of section 6.3.1 are done in the Xlarlab
scripts st rtpc I (figures 6.3 and 6.4), st rtpczb (figure 6.5), st_tpc3a (figure 6.6)
and st_tpc4 (figure 6.8). The calculations for the particle counter sizp distribu-
tions (section 6.3.2) are done in s t.isdpc I and st..adpcz.
The analysis of the total particle counts as discussed in sections 6.--1 to 6.~

is implemented in bioprof i. This function calls several ot her functions during

execution, i.e peakdetect is used for the detection of peaks (cf. section 6.5.1.2).
wf l tfun is the error function for the identification of the parameters Cl (T), C2 (T)

and a of the exponential tail (see section 6.5.2) and bioapprox is used for t.ho
determination of the parameters k 1 (f) and k 2 (f ) as discussed in srxtion 6.5.3. Tho
function bspinvs im simulates the inverse system to obt ain the biofilm shedding
profile as introduced in section 6.6.
A data file shdata. mat containing the step heights of the occasional particle
steps was generated before using bioprofi. This is done by dostepheight. This
script uses the function findstepheight, which is interactive. for that purpose.
This function is also called automatically from bioprofi if the file shdata. mat
does not exist in the search path.
The script bioscript, that employs b i oprof i , was used to generate the re-
sults presented in section 6./. The script biogi veout produces t he h()d~' of the
tables given in that section for inclusion in this thesis.
6.11 Conclusions

This chapter discussed the particle counts response to a step increase in flow. For
that both non-parametric (i.e. numerical and statistical) and parametric mod-
els of total particle counts we~e presented and a brief discussion of a numerical
statistical model of particle counter size distributions was included.
All models are based on observed data. Thus, the difficult issue of the deter-
mination of the model structure was solved via specifications derived from data
and assumptions validated by model fitting.
The basic parametric model covers the 'Slow Signal' of the particle counter re-
sponse. Specifications that have to be fulfilled in this case for the model to match
the data are provided in section 6.2, specifications for an extended model are pre-
sented as well in that section. The assumptions of sections 6.4.1.3 and 6.4.1.4 set
the foundation for the development of the model. They introduce a local shear-off
density and the separation of the effects of biofilm over distance and time dynam-
ics. In addition the dynamics are assumed to be exponential. The mathematical
model for a step increase of flow (see section 6.4.2) becomes a linear model (a con-
volution) which allows for parameter identification despite incomplete knowledge
of the input signal.
Results include parameters that describe the net effect of the sheared-off
biofilm over the length of the rig and the dynamics associated with the shear-off
process. The initial biofilm shedding profile (BSP) is another important result. it
gives a profile of the locations of sheared-off material along the rig and therefore
provides a tool for the assessment of the condition of the pipe rig.
Particles and biofilm are an important indicator of water quality. The model
presented in this section gives some more insight into the processes associated
with flow increases and particle shear-off from biofilms. Another frequent lv used
indicator of water quality, the disinfectant, will be the object of the next chapter.

184
Chapter 7

Comparison of Models of
Monochloramine Decay

7.1 Introduction

The aim of this chapter is to investigate two commonly used chlorine decay models

in detail, i.e. a single decay coefficient model and the combined bulk and wall

demand model used in Epanet [123]. These models are adapted for the use of

monochloramine as disinfectant and compared to each other. In addition, data

from bottled samples are presented and compared to results obt ained from the

pipe ng.
An introduction to chlorine or monochloramine as disinfect ant and an ove-rview

of the relevant literature was given in section 2.6.2 on page :3--1. Section .).:3 Oil

page 101 provides an account of the preprocessing procedure that was done prior

to the calculations given in this chapter.

The publication by the author together with R. S. Powell and C. A. \\'ood-

ward of 1997 [97] gives an account of earlier results, using the same or verv similar
met hods. Additional analvses are included in this chapter. \' ot p that. however.

a different correction procedure (i.e. reverse-calibration with no further window-

ing) was used for t he preprocessing of the data considered in this cha pt or (s('<'

section 5.3.2).
7.2 Review of Chlorine Kinetics
Most work on disinfection kinetics is concerned with free chlorine (C1 2 or HOCI-).
It is frequently assumed that free chlorine decays according to first order reaction
kinetics [25, 60]. A chlorine concentration c(x, t) that is decaying with time in
a flow system is described by the general partial differential equation given in
eq. (2.7) on page 35 [126]

ac(x, t) LBci», t) _ _ ( )
at + A ax - K ex, t . (7.1 )

This equation is valid for a constant velocity v = ~, where f is the volumetric

flow rate and A is the area of a pipe cross-section. The variables x and t represent
distance along the rig and time, respectively, and K is the decay coefficient of
chlorine. If K is constant this type of decay is referred to as first order.
For the simple one-dimensional system that is considered in this thesis, eq. (7.1)
is solved by

(7.2)

with

(7.3)

where t., t 2 represent time, Xl, X2 represent distance along the pipe and v is the
constant velocity of water in the pipe. Therefore, the decay of free chlorine is
characterised by just one (constant) decay coefficient K. The travel time between
points Xl and X2 is T = ~ = t2 - tl' For discrete time, eq. (7.2) can be rewritten

in the following form

c(.r + v . !1t, t + !1t) = c(x, t) exp( -1\ ~t). (7.4)

This equation is equivalent to eq. (7.2). however it allows us to deal with varving
parameters v and K since they need to be constant only within t he hydraulic

186
time step ~t, which in the case considered here is chosen to be equivalent to the
sampling time.
In particular, a possible variation of the decay coefficient K may be required.
since in recent years more evidence has been published that shows that simple
first order kinetics do not fully match the behaviour of free chlorine in a distri-
bution system. Lungwitz et al. [92] investigated booster (impulse) chlorination
and found that they had to postulate several decay coefficients depending on the
chlorine gradient. Schneider et al. [128] concluded that the decay coefficient (and
possibly other parameters) are flow dependent. Woodward et al. [160] found ex-
perimentally that the mono chloramine decay coefficient changes with flow rate.
Heraud et al. [67] used different decay coefficients depending on pipe materials
and diameter to fit their model to the data.
Simulation packages reflect these findings to some extent. In Piccolo [127]
it is possible to use a chlorine decay coefficient, K H 20 , which depends on total
organic carbon and temperature or to use diameter dependent decay coefficients
within the distribution system. Epanet [123, 124] includes a wall demand and
a flow dependent mass transfer coefficient. Thus, it replaces the constant decay
coefficient K with

(7.5)

where k f is a flow dependent mass transfer coefficient, k b , k w are constants rep-

resenting bulk and wall demand respectively and R H is the hydraulic radius.
The mass-transfer coefficient kf is calculated as

kf = Sh D
d
(7.6)

Sh= 0.023Re o. 83 SCO. 333 for Re > 2300 (7.7)

0.0668jReSc
Sh= 3.65 + ) 0.67
for Re < 2300 (7.8)
1+0.04 ( rReSc
d

187
 where kf mass transfer coefficient [rri/s],
Sh Sherwood number,
(or Nusselt number for mass transfer),
Re=~ Reynolds number.
Se= ~ Schmidt number,
D

d pipe diameter [m],

L pipe length [m],
f flow rate [m3/s],
A cross-sectional flow area of pipe [m2],
D molecular diffusivity of substance in fluid [m2/sL
u kinematic viscosity of fluid [m2/s].
Compare p. 17 in [124], pp. 128 and 142 in [50] or p. 176 in [68] to validate these
equations. The kinematic viscosity u of water is 10- 6 ";2, the Schmidt number
of diluted chlorine in water at 68°F is Be = 824. These numbers and the above
units are given for k w in ra]« and k b , K in l/s. In the implementation a factor of
3600 seconds/hour converts the term that includes k w to the unit l/h such that
k b and K can be used in l/h, while k w remains in tu]«.

7.3 Total Chlorine Kinetics

In the present study monochloramine is investigated since it is used in the London
distribution system, and it also finds application in other water networks world-
wide, e.g. [141]. Monochloramine (NH 2CI) is generally believed to behave in the
same fashion as free chlorine, only that it reacts more slowly than free chlorine
[156], also it was suggested that monochloramine is more efficient against biofilms
than free chlorine [85, 151].
To date, it is impossible to measure monochloramine directly online. Thus,
total chlorine, which includes free chlorine, monochloramine and organic chlo-
ramines, is measured using online meters. When ammonia is added in t he correct
ratio to free chlorine residuals, there will be no free chlorine left. However, organic
chloramines, in addition to monochloramine, are generally still present in t he wa-

188
 0.6

0.5 C(X,t) = (C(O,to)-CJ e- Kt + Coo

t.tl
z0.4
0::
o
~
U
.....l 0.3
g
E-- 0.2

C
0.1 - - - - - - - - • - - - - - - - - • - - •• _ •• _ •• _ •

oo;---------:----~------'-------L------.J
2 4 6 8 10
DISTANCE OR TRAVEL TIME

Figure 7.1: Simple first order decay of total chlorine.

ter. These organic chloramines are relatively stable and have a very low bacterial
action [52] and thus are considered as a constant fraction of total chlorine in this
work.
Thus, first order kinetics as in fig. 7.1 are obtained, which can be described
as

(7.9)

Note that c(x, t) is the total chlorine concentration at x, t. Monochloramine is

the reactive fraction of total chlorine, or c(x, t) - Coo, while Coo is the non-reactive
fraction of total chlorine, i.e. the organic chloramines.
To simplify notation in the following only the pipe rig inlet will be considerd
as reference point Xl = O. To signify this, to is written instead of t l and the

index 2 is dropped from the left hand side of eq. (7.9). As before, the travel time
T = t2 - t l = t - to can be used in the argument of the exponential function.
With these changes eq. (7.9) becomes

c(x. t) = (c(O, to) - coo) . e- K r + ex. (7.10)

189
 As elaborated earlier, it may be more realistic to replace the constant coeffi-
cient K in eq. (7.10) by a function depending on flow, pipe materials. and possibly
other variables. In sections 7.4 and 7.6, the dependencies of K on flow and inlet
chlorine concentration are investigated.

7.4 Determination of Decay Coefficients

Two possible descriptions of mono chloramine decay as in eq. (7.10) were studied.

1) K = const., and

To calibrate both models the unknown parameters K and Coo (case 1) or k b •

k w and Coo (case 2) have to be found such that the data fits a model according to
eq. (7.10).

7.4.1 Parameter Estimation for the Single Decay Coeffi-

cient Model

If only free chlorine and a single decay coefficient as in case 1 are considered.
it is possible to reduce this problem to a linear regression by taking the natu-
ral logarithm on both sides of eq. (7.2) and estimating In C(Xl' t 1 ) or In c(O, to)
and K. A weighted least-squares algorithm has to be used to remove the bias
introduced by taking the logarithm. This approach may also be suitable for the
mono chloramine problem eq. (7.10) if Coo is known or otherwise estimated, how-
ever , data values that are smaller than the final value Coo can not be included in
the estimation since they would lead to a logarithm of a negative value, which is
a complex number.
To apply this approach, first the natural logarithm of eq. (7.10) has to be
taken to obtain

C(l:, t) = c(O, to) - K T. (7.11)

190
with

C(X, t) = In (cfz, t) - coo) , (7.12)

C(O, to) = In (c(O, to) - coo) . (7.13)

A standard weighted least-squares error function J 1 has to be minimised over

K and c(O, to) with

nl n2

J1 = LL (w(i,j) [cc(Xj, t i ) + KT(i,j) - c(O. t o,i)])2 (7.14)

j=1 i=1

The index i of t i , to,i indicates that chlorine readings at n2 time instants are
considered, the index j of Xj implies that n1 locations along the rig are used
to obtain a best estimate of K. The upper limits n1 and n2 define the size of
the sample used for estimating K, in general a larger sample will give a better
estimate. In the case of the pipe rig n1 = 3, since there are three sample points
in the pipe rig. The weights w( i, j) and the travel time T(i, j) depend on distance
and point in time. The total chlorine reading cc(Xj, t i ) is the corrected meter
reading as opposed to calculated values which are not indexed.
The weight w (i, j) for an unbiased estimate is found by considering an ap-
proximation of the derivative of the logarithm. It is

d In x 1 In (x + ~x) - In x ~ In x
(7.15)
--=-~ =.
dx x ~x ~x

The error function J 1 contains for w(i,j) = 1 a difference of logarithms of

c(x, t) as in the above equation. An unbiased estimate, however, has to include a
difference of c(x, t). This difference of c(x, t) relates to the difference of logarithms
according to eq. (7.15) as ~c(x, t) = c(x, t) . ~ In c(x, t), and thus

(7.16)

191
 Due to the fact that data values smaller than Coo have to be excluded. this
approach does not necessarily perform well, instead, solving a nonlinear optimi-
sation problem is preferable.

Then, the problem is reformulated to find the minimum over 1\" and c'x of

nl n2

J2 = L L (cc(Xj, t i) - 91(j, i; K. cx ))2. (7.17)

j=1 i=1

with

(7.18)

where cc(Xj, t i ) is the corrected chlorine meter reading at time t, and distance Xj,
and r( i, j) is the travel time between Xj and x = 0 determined for the aliquot of
water that is at time t i at Xj' The measurement cc(O. ta,i) is the corrected meter
reading at the inlet for the time when the aliquot of water that reaches .fj at
time t i started its travel. The indices i and j indicate again that several time and
distance values are used for estimation. Instead of using the data value cc(O, ta,i)
this value could be estimated as well.
To minimise the cost function, J2 , a Nelder-Mead type simplex algorithm for
nonlinear systems as implemented in MATLAB is employed [107]. It is among
the most popular and successful direct search methods which do not require the
gradient. Roughly speaking the minimum is found by repetitive 'casting of a net'.
In more detail, a simplex of order n + 1 is constructed among the initial values
of the n parameters to be estimated. The initial value is iteratively improved
by contracting or expanding the simplex and shifting the corner with the highest
cost function toward the minimum. for instance by reflecting the worst corner
through an appropriately chosen centroid.

192
7.4.2 Parameter Estimation for the Model With Two De-
cay Coefficients

In case 2, the decay coefficient K is defined by eq. (7.5)~ thus it includes the
flow and diameter dependent parameters kf and R H . In this work only the flow
dependence is considered (flow is varied in steps during SFTs). Lsing a decay
coefficient as in eq. (7.5) in eq. (7.10) is valid only if for the whole period of
investigation T = t - to the flow does not change. Since the distance between the
investigated points x j is in the considered case relatively large. this condition is
not fulfilled. For this reason it is necessary to break up the time interval T into
small time steps ~T ~t in which the flow is constant. Thus, eq. (7.10) becomes

(7.19)

where n3 is defined by the travel time T(i, j) ,

n3

T(i, j) = L ~t(m). (7.20)

m=O

The travel time T ( i, j) is calculated through the knowledge of distance (x]) and
velocity v (or flow j) following eqs. (5.15) and (5.16) on page 110.
Note that in this consideration of the pipe rig, the portions of it with smaller
diameter are neglected. Therefore the hydraulic radius does not change, RH(m) =
R H = const.
The resulting optimisation problem is to find a minimum over all k b • klL' and
Coo of

nl n2

J3 = LL [cc(Xj, f i ) - g2(j, i; k b. k w . cx )]2. (7.21)

j=1 i=1

193
with

92(j, i; k b , k w , coo) =C oo + (cc(O, to,d - coo) x

-
x exp (-kbT(i, j) - f:
m=O
kf(m)k w

RH(kf(m) + kw )
~t(m)) . ( 1. .).))
__

and

na
T(i,j) = L ~t(m), (7.23)
m=O

where cc(Xj, t i ) , Xj, t i , and cc(O, td are defined as above.

A NeIder-Mead type simplex algorithm is used to minimise J 3 as in the case
of a single decay coefficient.

7.4.3 Issues in Estimation

Minimising these cost functions concerning Coo and k w poses some generic prob-
lems. Estimation of Coo is very difficult if data points are only available in the
initial portion of the exponential decay of fig. 7.1. The sensitivity of the estimate
of Coo on errors in these data points is very high. This leads in the case of the data
considered in this work to spurious results, which include physically impossible
solutions like large negative values of Coo' Generally, it appeared that it is futile
to attempt to estimate Coo unless the chlorine reading for the largest travel time
is close to coo'

Hence the problem of obtaining Coo has to be investigated from a different

source. It is possible to get a fairly reliable estimate of the fraction of organic
chloramines from a full chlorine speciation. In particular before ammonia dos-
ing, there is generally assumed to be only free chlorine present besides organic
chloramines. A value based on these titrations is used in the following as Coo'

Estimation of k w poses a different problem. It is possible only if

(7.:24)

194
where C is an arbitrary constant. In other words, either kf(m) or RH(m) needs
to change within the travel time r( i, j). Otherwise, there remains the problem of
estimating

nl n2

J4 = LL (cc(Xj, t i ) - (cc(O, td - coo) exp[-kbr(i, j) - c-i.. j)] + Lx )2.

j=l i=l

(;.25)

It is not possible to obtain an estimate for both kb and C from J 4 • only the sum
of both can be identified.
Therefore, in the following it will be attempted to estimate the Epanet decay
coefficient (i.e. case 2) only around flow changes, when k f varies.

7.5 Bottled Samples

During September, October and November 1996 bottled samples of disinfected
water were taken to evaluate the bulk water chlorine demand independent of the
wall demand. The method of sampling is provided in section 3.2.7 on page 58,
the schedule of these experiments is given in section 3.3.2 on page 62. Results are
discussed quantitatively in section 4.3.2 on page 79. In this section the decay coef-
ficients and other chlorine model parameters of free chlorine and monochloramine
as found from these bottled samples are provided.
Figs. 7.3 and 7.2 show the chlorine bottles data as discussed earlier together
with the fitted exponential functions. Monochloramine data is presented in
fig. 7.2, results of free chlorine bottles are given in fig. 7.3. Three parameters
K, Co and Coo were estimated by fitting the data c(t) to

(7.26)

The NeIder-Mead type simplex and the Levenberg-Marquardt gradient method.

both as implemented in Matlab,' were used to find the model parameters. Bot h

1 Functions fmins and leastsq of Matlab 5.

195
 0.45

..
0.4

:::,0 .35
eo
E
~ 0.3
Z 0 .25 •
02
0
....:l
::I: 0.2
u •
....:l
<r:0.15
E-<
• •
•
..
0
E-< 0.1

0.05

0
0 2 4 6 8 10 12 14
TIME [days]

F igure 7 .2: Monochloram in e in glass bottles and an exponen tial fun ction is fitt ed
to all data.

methods yield t he same resul ts within an accuracy of at least six digit s. T h

est imated paramet ers K; Co and Coo are pr ovid ed in table 7. 1, values for t he decay
coeffic ient K are given in l/hours and l/days, and t he t ime const ant Tl( = l~

is also provided in hours and days. Note t hat in t his case t here is no generic
probl em in estimat ing Coo (as t here is in t he pipe rig dat a ), beca use t he fi na l data
values are close to Coo . W hereas fig. 7.2 shows a goo d overa ll fi t to t he data ,
fig. 7.3 has a poorer fi t wh ich might affect t he estimation of the fi na l value.

Table 7.1 : D ecay coefficients and oth er fitt ed param et ers of monochlora m in e and
f ree chlorine bottled samp les.

l\1onochloramine
Decay Coeffic ient K Time Constant T J( Ini ti al Value Co F ina l Value Coo

l /hours l/days hours days mg/! mg/]

0.0236 0.566 42.4 1.77 0.337 0.1 55
Free Chlorine
Decay Coeffic ient K Time Constant Tl( Ini ti al Value Co Final Value Coo

l /hours l /days hours days mg/ l mg/l

0.125 3 8 0.333 0.343 0.0972

Fro m table 7.1 it is seen t hat frec chlorin decay mu ch more rapidly t hall

196
 0.45 r-----.-----,--------,-----,---r---~-~

0.4

::::,0.35
eo
E
~ 0.3
Z
C2 0.25
o-J
::r: 0.2
u •

..
-J
<C 0.15
t-
O •
t- 0.1 • .II

• • •
• •
0.05

2 4 6 8 10 12 14
TIME [days]

Figure 7.3: Free chlorine in glass bottles and an exponen ti al fun ctio n is fitt ed to
all dat a.

monochloramine (as expected) . If t his table is com pared to table 7.2 on page 198
it is see n , in add ition, t hat t he mon ochloramine decay in bottles is generally
smaller than the decay in t he pipe rig. Sin ce t he same wat er as in t he pip e rig at
rou ghl y t he same tem pe rat ure was used t he addit ional decay is most likely du
to t he influen ce of t he pipe system , not ably t he biofilm.

7.6 Analysis of Pipe Rig Data

7.6.1 Single Decay Coefficient

Only step fl ow t rials t o 0.6 li s or less were considered for the calculation of t he
decay coe fficient, since at higher fl ow (i.e. 0.9 li s) all chlorine read ings t hroughout

t he rig are a pproximately t he same.

Sing le decay coe ffic ients K , as in case 1 were ca lculated within 1- per iod.
of consecut ive d at a with vary ing length at di fferent fl ow and chlorin - cond it ion. .
avo id ing outliers in t he dat a wh ere po sible. These peri od wer furt h -r divid cl
int o 40 samples of approx imately equal lengt h (be tw en 39 and 0 cl cay c r f-

197
Table 7.2: M onochloramine decay coefficient data, sorted into regimes.

Period Mean Standard Sample Flow Inlet Tempera-

Number of K Deviation Size Chlorine ture
h- 1 h- 1
1 0.0496
lis mgll °c ,

0.0058 57 0.081 0.39 1--1..,1 I

i
2 0.0522 0.0028 57 0.081 0.39 13.9
3 0.0517 0.0043 57 0.081 0.38 13.7
4 0.0513 0.0034 57 0.081 0.36 1.,1.0
5 0.0503 0.0031 60 0.082 0.37 1.,1 A
6 0.0363 0.0014 50 0.083 0.37 13.3
7 0.0338 0.0021 51 0.083 0.39 13.0
8 0.0378 0.0024 50 0.079 0.39 6..,1
9 0.0332 0.0053 50 0.079 0.40 6.1 I
10 0.0315 0.0011 52 0.080 0.40 6.7
11 0.0285 0.0020 39 0.080 0.37 6.9
12 0.0280 0.0114 58 0.079 0.36 6.5
13 0.0616 0.0078 58 0.080 0.36 6.8
14 0.0310 0.0021 56 0.085 0.36 6.2
15 0.0329 0.0023 56 0.086 0.34 6.5
16 0.0578 0.0079 56 0.086 0.35 6.6
17 0.0512 0.0170 56 0.085 0.36 6.6
18 0.0480 0.0136 56 0.085 0.37 6.6
19 0.0375 0.0039 56 0.084 0.38 6.6
20 0.0333 0.0040 57 0.087 0.35 6.6
21 0.0207 0.0033 63 0.079 0.67 6.8
22 0.0173 0.0019 64 0.079 0.61 6.5
23 0.0349 0.0045 59 0.308 0.36 13.2
24 0.0274 0.0045 59 0.306 0.34 13.6
25 0.0393 0.0084 59 0.306 0.35 14.0
26 0.0422 0.0040 59 0.306 0.32 1.,1.0
27 0.0521 0.0033 61 0.306 0.32 14.0
28 0.0434 0.0046 62 0.308 0.31 14.2
29 0.0427 0.0062 62 0.309 0.31 14.0
30 0.0584 0.0094 62 0.308 0.32 13.6
31 0.0813 0.0060 62 0.308 0.34 13.5
32 0.0504 0.0075 63 0.302 0.66 7.6
33 0.0369 0.0052 64 0.298 0.66 6.9
34 0.0506 0.0101 72 0.452 0.35 6.8
35 0.0609 0.0081 72 0.448 0.31 7.0
36 0.0418 0.0056 70 0.449 0.61 7.5
37 0.0375 0.0034 70 0..,1.,18 0.59 7.5
38 0.0757 0.0320 80 0.604 0.34 7.0
39 0.0972 0.0228 40 0.606 0.35 6.8
40 0.0694 0.0138 5.,1 0.609 0.61 7.1

198
 0.09

-~ 0.08
I-
z
W
U 0 .07
u::
u.
w
8 0 .06
>
<t
~ 0.05 . ..
o
w
z
~ 0.04
<t
a:
9
J: 0.03
o
oz
~ 0.02

0.0 1

14:24 16:48 19 :11 21:35 23-Oct 02:24 04:48 07:11 09 :35 12:00
TIME

Figure 7.4: D ecay Coefficients in period 12 (22/23 Oct ober 1996) as exam ple oj
estim ated single monochloramine decay coeffic ien ts 1< oj data with low stan dard
deviation (0.0033 tc ") . Inlet fl ow is 0.3 li s, inlet chlorine is 0.32 mg/ l and
tem perature is 14°C.

ficient est im ates per sample) , to allow for extensive comparison of period s with
similar condit ions . The flow rates investigated were 0.08 li s, 0.3 li s, 0.45 li s,
and 0.6 li s, t he inl et chlorine conce nt rations were approximately 0.4 mgll and
0.65 mgll of total chlorine , t he te mpe rat ure was eit her 13-1 40 C or 6-7 0 C . In every
case, the chlorine readings were preprocessed as descri bed in section 5.3. Eac h
est im ate of the decay coeffic ient is based on five t ime instants or 15 dat a points.
The final value Coo is assumed to be 0.08 mg/I. Two examples of monochloramine
decay coefficients are depi cted in fig. 7.4 and 7.5, all 40 avera ge decay coeffi ci nt..
togethe r with other relevant data are given in table 7.2.
The presentation of the data in t a ble 7.2 follows regim es (separat d by hori-
zontal lin es). Thus, sections of t a ble 7.2 t hat are not separated by a horizontal
lin e have ro ug hly t he same flow inl et chlorine concentrat ion and t mpera t ure.
T he given values for mean decay coeffic i nt , flow inl t chlorin and temperat lire
a rt averages over each am ple. From t he t able it i er n that th hi rhes t (1\' -

.rag decay coeffic ient was found in peri od 39 (I{ = 0.0972 h- 1L at How o. 1/.'.

199
 0.1
1 ---.-- - ---.--- - , -- -----,- - ---,-- - -.-_ _,----_--..,

0.09

-:s 0.08
f-
Z
W
U 0.07
u:
u.
w
8 0 .06
>
«
~ 0.05
Cl
w
Z
~ 0.04
-c
a::
o...J
J: 0.03
o
oz
o
::2 0.02

0.01

04 :48 06 :00 07 :1 1 08 :24 09 :35 10:48 12:00 13:11

TIME

Figure 7.5: D ecay Coefficients in period 30 (8 Decemb er 1996) , as exam ple of

estim ated single monochloramine decay coefficie nts 1< of dat a with high standard
deviation (0.0170 h- 1 ) . Inlet flo w is 0.085 li s, inlet chlorin e 0.36 mgll and
temperature is about 1J C.

inl et chlorine 0.61 mg/l and t emperature 7oC, t he lowest decay coe ffic ient is
1< =0.0173 17, -1 (period 22, 0.08 lis , 0.61 mg/! and 6. 50C). Note, for com pa rison,
that the monochloramine decay coefficient in bottles with no chlorine dem and
was found to be 0.0236 h- 1 in the previous section. Som e of t he investi ga ted
periods have high standard deviations, in particular, during peri od s 38- 40 (a ll at
flow 0.6 lis) the inl et chlorine meter reading was oscillating whi ch implies t hat
these data have only limited validity. If periods 38-39 are disregard t he high est
decay coe fficient is 0.0813 h: ' (pe riod 31 , flow 0.3 li s, inl et chlorine 0.34 mg/I ,

temperature 13.50C).
From a first investigation for relationships between fl ow and monochloraminr
decay coe fficients, ther e is no clear pat t ern eme rging. Althou gh t he high t d cay
coe fficients wer e achi eved a t high er fl ows, even a t baselin e fl ow (0.0 11 ) t he coef-
fi cien t may rise to 0.0 578 h- 1 (pe riod 16) and also high er fl ow rate ustain deem'
coeffi cient s almost as low as t he on foun d in bottle te t ( .g. !\' = 0.027·1 h - 1 ill

p riod 24, with a fl ow of 0.3 li s).

200
 Investigations of temperature and inlet chlorine lead to similar results. Both
at high (14°C) and low (7°C) temperature both relatively large and quite small
decay coefficients are observed, and although the lowest decay coefficients are
found at an inlet chlorine residual of approximately 0.65 mg/l (and flow 0.08 lis),
very high decay coefficients of 0.0694 h- 1 (period 40) and 0.0418 h- 1 (period 36)
are also found at that level of inlet chlorine.
Figures 7.4 and 7.5 illustrate this further. Examples of data with low standard
deviation (period 12, fig. 7.4) and high standard deviation (period 30, fig. 1.5) are
provided. Both figures have the same y-axis limits for better comparison. The
main reason for the large drop in decay coefficient in fig. 7.5 is the (continuous)
decrease of the chlorine meter reading at sample point 1 during period 30, the
smaller fluctuations are probably due a slight oscillatory behaviour, chiefly of the
inlet chlorine meter.
A more rigorous investigation of relationships between decay coefficients is
the application of the t-test. It can be used to test if the means of two normally
distributed samples with unknown, but equal, variance are equal [78, 139].
T-tests on all 39~40 = 780 possible pairs of the 40 periods given in table 7.2 were
performed. If a significance level of 0.05 is applied, then 709 pairs of the 780 pairs
show a significant difference in mean, i.e. only 9.1 % or 71 time period pairs
have average monochloramine decay coefficient that are not signifcantly different
amongst the 40 investigated periods. The period numbers, flows, temperatures
and inlet chlorine residuals of these 71 period pairs are given in table 7.3 at the
end of this section.
Of the 780 possible pairs, 140 compare two average decay coefficients with each
one being in the same regime (in terms of flow, inlet chlorine and temperature).
Of these 140, only 17 period pairs (12.1 %) show no significant difference in mean
(see table 7.3). If only flow regimes are considered, only 31 of 295 pairs with the
same flow (or 10.5 %) show no significant difference in mean (and are therefore
included in table 7.3). Most of the 71 pairs of time periods given in table 7.3
are pairs of period with different flow, inlet chlorine or temperature conditions
(regimes), and most of the pairs of same conditions han' a significant l~' different

201
mean of the decay coefficient.
Assuming that the requirements for the t-test are fulfilled, the conclution
from these results is that there is no clear relationship between monochloramine
decay coefficient and flow, inlet chlorine or temperature. Although the t-r est

did discover sample pairs which exhibit means of only insignificant difference (as
given in table 7.3) the results imply that these relationships occur at random. i.e.
without a pattern related to either flow, inlet chlorine or temperature.

Table 7.3: Pairs of time periods with average decay co-

efficients that are not significantly different (according to
a t-test with significance 0.05). Period numbers are the
same as in table 7.2, the data is organised in terms of
pairs of flows (separated by horizontal lines).

Period Flows Temperatures Inlet Chlorine

Numbers lis DC mg/I
3 2 0.08 0.08 13.7 13.9 0.38 0.39
4 1 0.08 0.08 14.0 14.4 0.36 0.39
4 2 0.08 0.08 14.0 13.9 0.36 0.39
4 3 0.08 0.08 14.0 13.7 0.36 0.38
5 1 0.08 0.08 14.4 14.4 0.37 0.39
5 4 0.08 0.08 14.4 14.0 0.37 0.36
9 7 0.08 0.08 6.1 13.0 0.40 0.39
12 11 0.08 0.08 6.5 6.9 0.36 0.37
14 10 0.09 0.08 6.2 6.7 0.36 0.40
14 12 0.09 0.08 6.2 6.5 0.36 0.36
15 9 0.09 0.08 6.5 6.1 0.34 0.40
17 1 0.09 0.08 6.6 14.4 0.36 0.39
17 2 0.09 0.08 6.6 13.9 0.36 0.39
17 3 0.09 0.08 6.6 13.7 0.36 0.38
17 4 0.09 0.08 6.6 14.0 0.36 0.36
17 5 0.09 0.08 6.6 14.4 0.36 0.37
18 1 0.09 0.08 6.6 14.4 0.37 0.39
18 3 0.09 0.08 6.6 13.7 0.37 0.38
18 4 0.09 0.08 6.6 14.0 0.37 0.36
18 5 0.09 0.08 6.6 14.4 0.37 0.37
18 17 0.09 0.09 6.6 6.6 0.37 0.36
19 8 0.08 0.08 6.6 6.4 0.38 0.39
20 7 0.09 0.08 6.6 13.0 0.35 0.39
20 9 0.09 0.08 6.6 6.1 0.35 0.40
20 15 0.09 0.09 6.6 6.5 0.35 0.34
23 7 0.31 0.08 13.2 13.0 0.36 0.39
continued on next page

202
continued from previous page
Period Flows Temperatures Inlet Chlorine
Numbers lis DC mg/I
27 2 0.31 0.08 14.0 13.9 0.32 0.39
27 3 0.31 0.08 14.0 13.7 0.32 0.38
27 4 0.31 0.08 14.0 14.0 0.32 0.36
8 25 0.08 0.31 6.4 14.0 0.39 0.35
I
!,

9 23 0.08 0.31 6.1 13.2 0.40 0.36

11 24 0.08 0.31 6.9 13.6 0.37 0.34
12 24 0.08 0.31 6.5 13.6 0.36 0.34
32 1 0.30 0.08 7.6 14.4 0.66 0.39
32 2 0.30 0.08 7.6 13.9 0.66 0.39
32 3 0.30 0.08 7.6 13.7 0.66 0.38
32 4 0.30 0.08 7.6 14.0 0.66 0.36
32 5 0.30 0.08 7.6 14.4 0.66 0.37
33 6 0.30 0.08 6.9 13.3 0.66 0.37
33 8 0.30 0.08 6.9 6.4 0.66 0.39
16 30 0.09 0.31 6.6 13.6 0.35 0.32
17 27 0.09 0.31 6.6 14.0 0.36 0.32
17 32 0.09 0.30 6.6 7.6 0.36 0.66
18 32 0.09 0.30 6.6 7.6 0.37 0.66
19 25 0.08 0.31 6.6 14.0 0.38 0.35
19 33 0.08 0.30 6.6 6.9 0.38 0.66
28 26 0.31 0.31 14.2 14.0 0.31 0.32
29 26 0.31 0.31 14.0 14.0 0.31 0.32
29 28 0.31 0.31 14.0 14.2 0.31 0.31
32 27 0.30 0.31 7.6 14.0 0.66 0.32
33 25 0.30 0.31 6.9 14.0 0.66 0.35
34 1 0.45 0.08 6.8 14.4 0.35 0.39
34 2 0.45 0.08 6.8 13.9 0.35 0.39
34 3 0.45 0.08 6.8 13.7 0.35 0.38
34 4 0.45 0.08 6.8 14.0 0.35 0.36
34 5 0.45 0.08 6.8 14.4 0.35 0.37
34 17 0.45 0.09 6.8 6.6 0.35 0.36
34 18 0.45 0.09 6.8 6.6 0.35 0.37
35 13 0.45 0.08 7.0 6.8 0.31 0.36
37 8 0.45 0.08 7.5 6.4 0.59 0.39
37 19 0.45 0.08 7.5 6.6 0.59 0.38
34 27 0.45 0.31 6.8 14.0 0.35 0.32
34 32 0.45 0.30 6.8 7.6 0.35 0.66
35 30 0.45 0.31 7.0 13.6 0.31 0.32
36 26 0.45 0.31 7.5 14.0 0.61 0.32
36 28 0.45 0.31 7.5 14.2 0.61 0.31
36 29 0.45 0.31 7.5 14.0 0.61 0.31
37 25 0.45 0.31 7.5 14.0 0.59 0.35
continued on next page

203
 0.03 15 r-------r--.---.----.--------r--,------.----,-----,--~~

0.03 1
K
0.0305

Figure 7 .6: Th e com bined decay coeffic ient 1< as us ed in Epan et v rsus tli II all
demand k w (h ere k b = 0.03 l/h, R H = 0.0275 m, k f = 10- 7 m/ s) .

con t i n ued from previou s p age

Period Flow s Temperatures Inlet Chlorine
Numbe rs l/ s °c mg/ !
37 33 0.45 0.30 7.5 6.9 0.59 0.66
38 31 0.60 0.31 7.0 13.5 0.34 0.34
40 38 0.61 0.60 7.1 7.0 0.61 0.34

7.6.2 Combined Decay Coefficients

The decay coefficients of case 2 were also investi gated. If t he variat ion of t he decay
coeffic ient followed t his model, t he fi t of t hat mod el would be bet ter t ha n t hat of
a single decay coefficient mod el. For four flow chan ges bo t h single and combined
decay coefficients were calculated for comparison, again assum ing c = 0.08 mg/l.
The investi gat ed fl ow changes were from 0.08 l/ s to 0.3 l/ s (period A) from
0.3 l/s t o 0.08 l/s, from 0.08 l/s t o 0.45 l/ s (pe riod C) and from 0.08 l/ s to 0.9 l/ s
(period B). Three optimisation procedures were used for t he calculatio n of t he
decay coeffi cients: a elder-Mead type simplex, a Levenberg-Marqu ard t gracli nt
method and a Gau ss- [ewt on gradient based algor it hm all as impl emented in
Matla b" . A maximum of 8000 it erati ons were done in each case. \'\ it hin t hi limi t
of iter ati ons, t he Nelder-Mead method always produced t he smallest residual co t

fun cti on (or root mean square erro r) .

To clarify t h expected order of magn itude of value , con ider t he Epa nct

2F\mc t ions fmins for Nelder- Mea d sim plex and leastsq for t he ot h r twa III t ha d. i ll Ma t-
lab 5.

204
decay coefficient as given in equation (7.5):

( -, . _')-)
I

The factor 3600 is introduced for conversion of the term with ku . from l/seconds
to l/hours. If reasonable approximate values of kb ~ 0.03 l/h. R H = 0.0275 m
and k f ~ 10- 7 tu]« are used, a plot of K versus k w as given in fig. 7.6 can be
obtained.
From this figure it is obvious that the sensitivity of K towards changes of klL'
is negligible for k w > 10- 5 and k w < 10- 11 . This range shifts with varving k f .
Note that Rossman [124, p.89] uses 1 foot/day=3.53 . 10- 6 metre/second in an
example in the manual of Epanet.
From the investigated time periods, in one case k w converged to a negati Y('
(i.e. infeasible) value. In three of four cases values of k w that are between 10- 8
and 108 are obtained. Only time periods with positive k w are investigated further,
they are given in table 7.4. These periods are 18.10.96 00:00 to 19.10.96 12:00
(period A), 14.10.96 12:00 to 15.10.96 12:00 (period B) and 16.12.96 12:00-18:30
(period C).
The results were compared by taking the root mean square error between the
N data points Yj and the corresponding model output f) for each chlorine decay
models with decay coefficients K according to case 1 and 2, respectively. The
error is defined as

RMS Error = ~L (Yj - f))2. (7.28)

j=l

It is seen from table 7.4 that the decay coefficients of case 2 perform bettor
than a single decay coefficient as in case 1, however, the improvement in period A
is neglibile. In periods Band C, however, an improvement of 22.8 % and 17.-1 %
over the values of case 1 is observed. The results from these periods WPf(' not
included in [97]. It is important to note again in this context that the scnsit ivitv
of the determination of k w is verv small. i.e. in period B, a root mean square error

205
Table 7.4: Results of the estimation of combined decay coefficients (as in
Epanet). All decay coefficients are found through a Nelder-Mead type simplex
as implemented in M atlab.

Time Period A B C
Data Points 1728 1560 2184
Flow Change [lis] 0.08--+0.3 0.08--+0.9 0.08--+0.45
K, Case 1 [h 1] 0.0348 0.0566 0.0309
k b , Case 2 [h 1] 0.0340 0.0490 0.0242
k w , Case 2 [h- 1] 1.95.10- 7
4.11.10 8 3.90.107
RMS Error, Case 1 [mg/l] 0.012538 0.022498 0.021032
RMS Error, Case 2 [mg/I] 0.012534 0.017364 0.017364

that is only 2.70.10- 9 mg/I larger than the one given in column 3 of table 7.-1
leads to k w = 5.93 ui]«, a reduction by a factor of 10- 8 . A wall demand of
kw = 10-2 ui]« in period C (column 4) leads to an increase of the root mean
square error of only 5.19.10- 7 mg/I, a wall demand of k w = 10- 6 m/s would,
however, increase to error by 3.85.10- 3 mg/I.
In summary, an improvement of model error due to the use of a combined
decay coefficient with flow dependence as currently used in Epanet is observed.
However, reliability of wall demand kw estimation results can be very low, i.e. in-
accuracies involved in estimating kw are several orders of magnitude for periods B

and C.

7. 7 Discussion
Although a relatively large variability of the single decay coefficientis is observed,
this does not appear to be related to flow, chlorine changes or temperature.
However, the results of columns 3 and 4 in table 7.4 give strong evidence that the
model with combined bulk and wall decay gives better results than a single decay
coefficient. The flow changes in these periods are higher than in period A. This
is a likely reason for the improved performance of the Epanet model compared

to period A.
But, as pointed out earlier, the sensir vitv of the model and cost function to

206
changes in kw may be very small. This is the case in periods Band C. where
the actual value of k w is of little importance. As long as it is much larger than
k f (e.g. 1000 times), the flow dependent parameter k f will determine the second
term of the Epanet model.

Although from a discussion of initial results it was concluded that a single

decay coefficient describes the monochloramine decay adequately [97]. the addi-
tional data of periods Band C in table 7.4 implies that a model as used in Epanet
(with flow dependence) performs better than a decay coefficient as in case 1.
From the detailed analysis of the single decay coefficient data it emerged.
however, that the prime obstacle for the validation of chlorine models are the
inaccuracies in the data. Even after performing a preprocessing procedure based
on an independent set of measurements (i.e. the titrations). estimated decay
coefficients vary significantly even within the same regime. It is noted in this
context that the investigated step flow trials were undertaken after a series of
prior trials. Thus, if an increase of decay coefficient is due to an increase in
entrained particulate matter or erosion of biofilm. the previous step flow trials
will have decreased that effect.

7.8 Implementation
The decay coefficients in bottled samples presented in section 7.5 were found with
wdecay. This Matlab function can calculate decay coefficient in 18 different wavs.
They differ in the type of decay coefficient (single or Epanet), the type of optimi-
sation procedure (least-squares fit with logarithmic approximation, Nelder-Mead
type simplex, Levenberg-Marquardt or Gauss-Newton gradient methods) and in
the amount of data taken into account (either one estimation per time instant or
a cumulative estimation of all in all one coefficient for a whole period). These cal-
culations are either done by calling separate functions (wdclse. wdcepa for fitting
with a logarithmic approximation) or they are done in wdecay with t he help of a
separate error function (errfundclse. errfundclsel, errfundclse2. fundclse.
fundclsel, errfundcepa, errfundcepal. errfundcepa2. fundcepa, fundcepa2;

-
')0-(
the numbers indicate either (no number) estimation of decay coefficient and final
value, (one) additional estimation of initial value, or (two) estimation of decay
coefficient only).

The analysis of pipe rig data of section 7.6.1 and section 7.6.2 were done with
calcdecay, which uses wdecay internally. This function calculates a series of
subsequent cumulative decay coefficients and saves the result to a xlatlab data
file. The methods 17 and 18 in wdecay or calcdecay were used for the calculation
of single or Epanet coefficients, respectively (see help wdecay).
Matlab scripts were written to calculate the presented results. The script
calcmax uses calcdecay to compute decay coefficients, chlmeans2 generates all
the data provided in tables 7.2 and 7.3 and chlgiveout produces ASCII files of
that data for the inclusion in this thesis. The function testepa uses some of the
outputs of wdecay to test the performance of different optimisation procedures.
It was used in section 7.6.2 together with rmsechl, which calculates root mean
square errors of chlorine data.
All of the above discussed functions and scripts are part of the Water Quality
Toolbox that was written by the author (see appendix C).

7.9 Conclusions

The important task of chlorine model calibration was addressed in this chapter.
It appeared that the issue of data accuracy is paramount, since the sensitivity of
decay coefficient estimates to data inaccuracies is very large.
It was shown that the final value Coo cannot normally be estimated. Furt her-
more, it can only be attempted to estimate the wall demand k w of the Epanet
model if a flow or diameter change takes place, but even then it may be impossible
to determine the exact value due to a possibly very shallow cost function.
Finally, no relationships between single decay coefficients and flow. inlet chlo-
rine or temperature could be found, however, some evidence for an improved
performance of the Epanet model over single decay coefficients for flow changes
by more than 0.3 lis was presented.

208
 It is concluded that there is a variation of monochloramine decay coefficent
due to flow changes, however, because of the frequently high uncertainty asso-
ciated with chlorine data, a simpler model with only a single decay coefficient
may be a good approximation of mono chloramine decay in a system like the one
investigated in this thesis.

209
 Part IV

Conclusion

:ZIO
Chapter 8

Discussion and Conclusions

This thesis covered experimental work and analyses of data aimed at the devel-
opment of water quality models for water distribution networks. A total particle
counts model was constructed for a step flow input and two monochloramino de-
cay models were discussed and compared. In this chapter, some of t he broader
issues that are associated with water quality modelling will be discussed and a
summary will conclude this work.

8.1 Merits of this Approach to Modelling

In this thesis, techniques of phvsical modelling were combined with svst cm iden-
tification to determine a model of particle counts. The 'hybrid' method that is
presented in this work is somewhat different from earlier proposed approaches
such as empirical modelling or semi-physical modelling. The system is first split
int 0 suitable sub-svst.ems, derived specifications for the model from t lIP dat a for
those sub-systems and then proposed models that are validated through their
interpretability and by fitting them to data. Thus, a model ,,"as found t hat lends
itself to phvsical or biological interpretation. The other criterion for a good model
is t hat it should be usable, e.g. for the determination of the biofilm shedding pro-

file.
The approach taken in this thesis (as all met hods of S\"st ern idcnt inca t ion) i:--
dat a oriented, i.e. net e£frcts are capt urcd rather than t II(' fine dot ails of uudorlvi Ilg

:211
processes. This has the advantage that the effects that are of the most interest
are modelled. This approach is limited by the assumptions. which need to be
fulfilled for the model to be valid and therefore might have to change in a trial
and error process. Nevertheless, a model structure was derived directly from data.
Although as much physical (or biological) knowledge of the underlying processes
as possible was incorporated into finding the model structure to avoid a lengthy
period of trial and error, this knowledge is only then absolutely required when
the meaning of the model is being examined.

8.2 Generalisation

Naturally it is desirable to extend the results of this work (which were derived for
the "TORUS" pipe rig) to more general water distribution systems. more general
flow regimes (i.e. not just step changes) or long-term effects.
One major problem is the impact of different pipe materials. From the results
of the pipe rig, it is very difficult to extrapolate information about systems that
are not predominately MDPE pipes. The pipe rig has less than 2 % (15 m) of
lined or unlined iron pipes. Despite that iron particles are found throughout the
system (see [46]), which implies that even these 2 % may influence the particle
counts.
Another Issue is that flow regimes that do not consist of steps alone will
probably make it impossible to find a linear model. This is demonstrated in
section 6.4.3, where the negative input slope model is developed. Because the
contribution of particles before flow change cannot be neglected (in this case
flow decrease), the obtained model is not a convolution anymore and. therefore.
nonlinear. This is likely to be even more the case if the flow varies arbitrarily.
For the possibility of long-term predictions, a full model has to include long-
term effects as well. Those were studied only briefly in this work. because t hev
lie outside the scope of this thesis. A model of long-term effects will either
require several years of experiments or be based on published results and their
extrapolation.

212
 The particle counts model of chapter 6 may also be a part of a full model. It
would, however, cover short-term effects only. In particular, it would cover shear
stress related erosion of biofilm.

8.3 Water Industry Perspective

Some of the results of this thesis have implications on the operation of water
distribution systems.
The relationships between particle counts and plate counts or epifluorescence
microscopy counts as discussed in section 4.5.3 implies that particle counters
may give valuable information on the bacteriological condition of the network.
More research will be necessary into these findings. From the results presented in
this work it is concluded that particle counts can be used effectively to measure
suspended bacteria in a high shear stress environment where 'clumps' of biomass
dominate.
Apart from the insight in the processes of biofilm detachment and the dynam-
ics of suspended particles, the main use of the particle counts model of chapter 6
is the computation of the biofilm shedding profile. This profile gives informa-
tion about the state of the system and, in particular, potentially critical areas in
terms of high shear-off. With this method, a step flow trial could be used to test
a portion of the distribution system to locate the sources of increased shear-off.
The discussions of the peaks (see section 4.1.2.1) are also very relevant. The
results give evidence to the effect that high particle incidents originate from valves.
Thus, this implies that the maintenance of valves is important for the compliance
with water quality standards.
The preprocessing of chlorine meter data of section 5.3 highlighted that it is
necessary to take careful account of inaccuracies in the readings of these meters
and , therefore , more accurate meters will be necessary for effective calibration of
chlorine models. The preprocessing procedure to improve data accuracy with the
help of titrations can be used for real distribution systems.
In the comparison between different chlorine decay models of chapter 7. addi-

213
tional insight is given that is of use if chlorine decay models are to be calibrated
for distribution systems; in particular, the difficulties associated with fitting the
parameters are discussed in some detail (sections 7.4.3 and 7.6.2). This has impli-
cations on the design of experiments for chlorine model calibration: if it is desired
to find the wall demand in the Epanet model it will be necessary to consider a
time span with a flow change or, alternatively, a network area with change of
diameter. In addition, it was seen that even if these constraints are satisfied it
may be impossible to find the wall demand due to the poor sensitivity of the
model and therefore it might be necessary to obtain it from an alternative source
[153] or just assume some reasonable value.
The Reynolds numbers used in this work (see section 3.3.1) are relatively low
compared to operational conditions in strategic mains, however, they generally
match the conditions in pipes of similar sizes as the pipe rig (i.e. 110 mm diame-
ter). Thus, the results of this thesis are particularly valid for the pipe work near
to the customer's tap.

8.4 Future Perspective

As already discussed when the extended model was introduced in section 6.4.1.4, a
continuation of this work may include a more general particle counts model. This
model would cover the suggested extensions to the model of biofilm detachment
as measured by suspended particles together with an indication of the biofilm on
the pipe walls. In addition, diurnal variations, e.g. input flows that are not steps,
or long-term aspects may feature in the model.
A piecewise linear decomposition of a more general input signal could be used
to make the current particle counts model applicable to more general inputs.
however, it will be required to distinguish between positive and negative slopes
in the input and there is likely to be a threshold for the positive slope below which
the input signal will not result in a response. In addition, it will be necessary
to take into account the dependence of the parameters on the magnitude of the
input, e.g. bv the generation of a look-up table of parameter values. However,

214
such a look-up table will require many experiments under similar conditions.
These problems will have to be addressed before the particle counts model can
be integrated into a software package. For this integration it may be a natural
choice to use the numerical models discussed in section 6.3.1. however. they
, . have
similar shortcomings to the parametric model. After reformulation in terms of a
numerical model, the parametric model could be adopted as well.
More work on monochloramine will require more and better data. Other
model structures should be considered and it is desirable to find a chlorine decay
model where the wall demand exhibits a better sensitivity than in the Epa net
model.
Other water quality variables, like temperature, free chlorine. bacterial counts
and nutrient content (Biodegradable Dissolved Organic Carbon. BDOC) could
be investigated in more detail. This will require further literature work and
experiments on possibly both the pipe rig and a distribution system (with meters).
These experiments could feature longer periods of high flow than in the present
work and it would be beneficial to be able to control water temperature in the
pipe rig. Simply heating the water may serve that purpose because most of the
system is buried underground.
Fuzzy modelling (cf. section 2.10) is probably a natural choice to combine
several linear models obtained by these further investigations into one large non-
linear model.
Alternatively, a model derived only from already published knowledge (as
e.g. in [49]) could be attempted. It is likely that parts of such a model are not
described in literature to date, these aspects would have to be investigated sepa-
rately by similar methods as used in this thesis. Results of modelling from dat a
concerning, for instance, biofilm detachment, the relationship of particle counts
and bacterial counts or monochloramine/free chlorine decay may be included

symbiotically.

215
8.5 Conclusions

This thesis focused on two aspects of water quality in distribution systems: parti-
cle counts and disinfectant. Additional observations of heterotrphic plate counts
and epifluorescence microscopy counts are reported.
The response of the particle counts to a step increase of flow was modelled
from experimental data. This model introduces the concepts of a local particles
shear-off function and an initial biofilm shedding profile. They gin> insight into
the dynamics and the condition of the network. The parameters of this model
describe the net effect and the speed of biofilm shear-off. It is observed that
the net effect decreases towards the end of the experiment which is likely to
be due to repeated shear-off, decrease of temperature and the introduction of a
disinfectant. The biofilm shedding profile and the identification of peaks allow to
pinpoint potentially critical sections of the system (in terms of high shear-off).
The approach taken in this work uses specifications derived from data and
assumptions to obtain a meaningful model structure. In particular, the degree
to which it is interpretable, and the fit of the parameters and biofilm shedding
profile are used for validation.
Monochloramine is used as disinfectant in the experimental system considered
in this work. The chlorine meter readings were first preprocessed with the help
of titrations to remove meter drift, and then two different monochloramine decay
models are calibrated and compared. The calibration (or parameter identifica-
tion) highlighted several generic problems that come about due to the mathe-
matical structure of the models, i.e. it is impossible to estimate both the bulk
decay coefficient and the wall demand unless there is a flow or diameter change
in the considered time or distance interval. The decay coefficients are compared
and tested for flow, temperature or inlet chlorine residual dependence, however,
only a weak dependence on flow could be found.
It is anticipated that this work will facilitate the assessment of potential dan-
ger from biofilm (in terms of high shear-off). Furthermore. the work has shed
light on biofilm detachment dynamics and chlorine decay model calibration. A

216
generalisation of this work will facilitate the incorporation of these findings into
a water quality monitoring and control system.

')1-'
- I
Bibliography

[1] O. Adam and Y. Kott. Evaluation of water quality. as measured by

. bacterial
maximum growth rate. Water Research, 23(11):1--107-1--112, 1989.

[2] S. M. Alexander, N. L. Glenn, and D. W. Bird. Advanced techniques in

the mathematical modeling of water distribution systems. Journal of the
American Water Works Association, pages 343 -346. July 1975.

[3] J. H. Andersen and R. S. Powell. Simulation of water networks containing

controlling elements. Submitted for publication to the Journal of ~rater
Resources Planning and Management, 1998.

[4] K. J. Astrom and B. Wittenmark. Computer-Controlled Systems. Prentice

Hall International, London, 3rd edition, 1997. Chapter 8.

[5] R. Babuska and H. B. Verbruggen. An overview of fuzzy modeling for

control. Control Engineering Practice, 4(11): 1593-1606, 1996.

[6] R. Bakke, W. G. Characklis, M. H. Turakhia, and A.-I. Yeh. Modelling

a monopopulation biofilm system: Pseudomonas Aeruginosa. In \Y. G.
Characklis and K. C. Marshall, editors, Biofilms, pages 487-520. J. wn«,
& Sons, New York, 1990.

[7] R. Bakke and P. Q. Olsson. Biofilm thickness measurements by light mi-

croscopy. J. Microbial. Meth., 5:93, 1986.

[8] G. Billen. Protein degradation in aquatic environments. In R. Chrost , edi-

tor, Microbial Enzyme in Aquatic Environments, pages 3--18-389. Springer-
Verlag, Berlin, 1991.

[9] S. A. Billings. Identification of nonlinear systems-a survey. lEE Proceed-

ings, Part D, 127(6):272-285, 1980.

[10] P. Biswas, C. Lu, and R. 1\1. Clark. Particle and chemical transport in
drinking water systems. In Water Quality Modeling in Distribution Sys-
tems, pages 323-361, Cincinnati. Ohio, Feb. 1991. American \Yater \Yorks
Association.

[11] P. Biswas, C. Lu, and R. i\1. Clark. .-\ model for chlorine concentrat ion
docav in pipes. Water Research, 27(12):1715 172--1. 1993.

218
[12] J. C. Block, F. Bois, D. J. Reasoner, M. Dutang. L. Xlathieu. J. L. Paquin.
and J. Maillard. Disinfection of drinking water distribution svstems. lFater
Supply, 13(2):1-11,1995. '

[13] J. C. Block, M. Dutang, J. Maillard, and D. Reasoner. Growth of attached

bacteria in water distribution systems. Water Supply, 12(1/2):SSl-8 - SSl-
12, 1994. Special subject: Biological activity in distribution systems.

[14] P. F. Boulos, T. Altman, R. W. Bowcock, A. K. Dhingra, and F. Collevati.

An explicit algorithm for modeling distribution system water quality with
applications. In Proceedings of the 2nd International Conference on IVater
Pipeline Systems, pages 405-423, Edinburgh, 1994.

[15] P. F. Boulos, T. Altman, P.-A. Jarrige, and F. Collevati. An event-driven

method for modelling contaminant propagation in water networks. Applied
Mathematical Modelling, 18:84-92, Feb. 1994.

[16] M. A. Brdys, H. Puta, E. Arnold, K. Chen, and S. Hopfgarten. Operational

control of integrated quality and quantity in water systems. In P. D. Roberts
and J. E. Ellis, editors, IFAC Symposium on Large Scale Systems: Theory
and Applications, volume 2, pages 715-721, London, 1995. Pergamon.

[17] M. A. Brdys and B. Ulanicki. Operational Control of Water Systems: Struc-

tures, Algorithms and Applications. Prentice Hall, London, 1994.

[18] I. N. Bronstein and K. A. Semendjajew. Taschenbuch der Mathematik. B.G.

Teubner Verlagsgesellschaft, Stuttgart, 1991.

[19] M. R. W. Brown, R. J. Courcol, and G. P. Testing sensitivity of biofilms to

antimicrobial agents. In J. W. T. Wimpenny, W. Nichols, D. Stickler, and
H. Lappin-Scott, editors, Bacterial Biofilms and Their Control in Medicine
and Industry, pages 135-140. BioLine, Cardiff, 1994.

[20] J. D. Bryers. Biologically active surfaces: Processes governing the formation

and persistence of biofilms. Biotechnology Progress, 3(2):57-68, June 1987.

[21] P. Burgess, P. Clayton, and D. Vitanage. Dynamic water quality modelling

in the sydney water board. In Australian Institute of Engineers WATER-
COMP, pages 363-372, Melbourne, Australia, 30th March - 1st April 1993.

[22] P. L. Burgess, D. C. Vitanage, A. H. Morton, and P. J. Bliss. Determination

of kinetic decay factors to model chlorine in water supply distribution sys-
tems. In Hydrotop'94: Colloque 'Mieux gerer l'Eau', Vol. 2, pages 143-151.
Marseille, France, 1994.

[23] V. K. Chambers. Long term variability of chlorine decay rate in distribu-

tion. Technical Report UC 2295, WRc Swindon. Frankland Road, Blagrovr.
Swindon, Wilts, Apr. 1994.

219
[24] V. K. Chambers, J. D. Creasey, and J. S. Joy. Xlodelling free and total
chlorine decay in potable water distribution systems. Journal Water Supply
Research and Technology - Aqua, 44(2) :60-69. 1995.

[25] V. K. Chambers and J. S. Joy. Validity of modelling chlorine residuals in

distribution. Technical Report DC 1942, WRc Swindon, Frankland Road.
Blagrove, Swindon, Wilts, Apr. 1993.

[26] W. G. Characklis. Fouling biofilm development: A process analysis.

Biotechnology and Bioengineering, 23:1923-1960, 1981. .

[27] W. G. Characklis and K. C. Marshall. Biofilms. J. Wilev S: Sons. \'ew

York, 1990.

[28] W. G. Characklis, M. H. Turakhia, and N. Zelver. Transport and interfacial

transfer phenomena. In W. G. Characklis and K. C. Marshall, editors,
Biofilms, pages 265-340. J. Wiley & Sons, New York, 1990.

[29] K. Chen and M. A. Brdys. Set membership estimation of state and pa-
rameters of large scale water supply and distribution systems - integrated
quantity and quality. In P. D. Roberts and J. E. Ellis, editors, IFAC Sym-
posium on Large Scale Systems: Theory and Applications, pages 5-!3-548,
London, 1995. Pergamon.

[30] R. M. Clark. Water quality modeling in distribution systems. Journal of

Environmental Science and Health, Part A, 27(5):1329-1366, 1992.

[31] R. M. Clark, D. J. Ehreth, and J. J. Convery. Water legislation in the U.S.:

an overview of the Safe Drinking Water Act. Toxicology and Industrial
Health: An International Journal, 7(516):43-52, 1991.

[32] R. M. Clark and J. A. Goodrich. Modelling human exposure to contami-

nants from drinking water. Journal Water Supply Research and Technology
- Aqua, 41 (4) :224-230, 1992.

[33] R. M. Clark, J. A. Goodrich, and L. J. Wymer. Effect of the distribu-

tion system on drinking-water quality. Journal Water Supply Research and
Technology - Aqua, 42(1):30-38, 1993.

[34] R. M. Clark, W. M. Grayman, J. A. Goodrich, R. A. Deininger, and

K. Skov. Measuring and modeling chlorine propagation in water distri-
bution systems. Journal of Water Resources Planning and Management.
120(6):871-887, November/December 1994.

[35] R. M. Clark, W. M. Grayman, R. M. Males, and J. D. Coyle. Predicting wa-

ter quality in distribution systems. Technical Report EPA/600/D-86/230.
Drinking Water Research Division, U.S. Environmental Protection Agency.
Cincinnati, Ohio, Sept. 1986.

220
[36] R. M. Clark, B. W. Lykins, J. C. Block, L. J. Wymer. and D. J. Reasoner.
Water quality changes in a simulated distribution system. Journal It'ater
Supply Research and Technology - Aqua, 43(6):263-277. 1994.

[37] R. M. Clark, G. Smalley, J. A. Goodrich. R. Tull. L. A. Rossman. J. J.

Vasconcelos, and P. F. Boulos. Managing water quality in distribution
systems: Simulating TTHM and chlorine residual propagation. Journal
Water Supply Research and Technology - Aqua, 43(4):182-191, 199-1.

[38] J. Cohen. The development of a dynamic calculation model for drinking-

water networks. Journal Water Supply Research and Technology - Aqua.
39(3):172-187,1990.

[39] J. Cohen. Further development of a dynamic calculation and simulation

model for water distribution. In Water Quality Modeling in Distribution
Systems, pages 155-181, Cincinnati, Ohio, Feb. 1991. American Water
Works Association.

[40] F. Colin and G. Grapin. Design of a hydraulic and water quality model
of a meshed distribution network: The case of chemical species in a singel
kinetics reaction. In Water Quality Modeling in Distribution Systems. pages
141-153, Cincinnati, Ohio, Feb. 1991. American Water Works Association.

[41] F. Colin, G. Grapin, J. Cheron, Y. Levi, E. Pozzoli, J. Miazza, and O. Pas-

cal. Etude de l'evolution de la qualite d'eau potable dans les reseaux de
distribution, une approche et des moyens nouveaux. Techn. Sci. Mimic.
Eau, 12:565-574, 1987.

[42] B. Coulbeck and C. H. Orr. A network analysis and simulation program for
water distribution systems. Civil Engineering Systems, 1(3):139-1-14, 1984.

[43] B. Coulbeck, B. Ulanicki, and C. H. Orr. Optimal scheduling of water

supply taking account of quality and transmission costs. In Water Quality
Modeling in Distribution Systems, pages 131-139, Cincinnati. Ohio, Feb.
1991. American Water Works Association.

[44] A. Delanoue and D. M. Holt. "TORUS" pipe test rig design and con-
struction. Technical Report Projects R9311 & R9374. Thames Water Plc,
Group Research & Development, Spencer House, Manor Farm Road. Read-
ing, RG2 OJN, England, Feb. 1995. Company Confidential.

[45] A. Delanoue, D. M. Holt, H. A. Anderson. S. 1\1. 1\IcMath, S. E. Smith.

C. A. Woodward, A. R. Fraser, and 1\1. Roe. Analysis of material (biofilm)
present on the internal surfaces of a drinking water distribution syst cm. In
C. Keevil, editor, Proceedings of the International Conference on Biofilms iu
Aquatic Systems, Coventry. April 13-16 1997. Royal Societv of Chemist rv.

[46] A. Delanoue, D. 1\1. Holt, S. 1\1. ~Ic~Iath. and S. Smith. "TORCS" pipe
test rig strip down 1996. Technical Report Project R19610, Thames \\'atpr

221
 Pic, Group Research & Development, Spencer House. Xlanor Farm Road.
Reading, RG2 OJN, England, Dec. 1997. Company Confidential.

[47] J. Dojlido and G. A. Best. Chemistry of Water and Water Pollution. Ellis
Horwood, Ltd., London, 1993.

[48] R. M. Donlan and W. O. Pipes. Selected drinking water characteristics and

attached microbial population density. Journal American Water lrork...
Association, 80(11):70-76, Nov. 1988.

[49] S. Dukan, Y. Levi, P. Piriou, F. Guyon, and P. Villon. Dynamic modelling of

bacterial growth in drinking water networks. Water Research, 30(9):1991-
2002, 1996.

[50] D. K. Edwards, V. E. Denny, and A. F. Mills. Transfer Processes. ~IcGraw­

Hill, New York, 1976.

[51] A. Elton, L. F. Brammer, and N. S. Tansley. Water quality modeling in

distribution networks. Journal of the American Water Works Association,
87(7):44-52, July 1995.

[52] T. H. Feng. Behaviour of organic chloramines in disinfection. J. Water

Pollut. Control Fed., 38(4):614-627, 1966.

[53] J. W. Forrester. Principles of Systems. Wright-Allen Press, Cambridge,

Massachusetts, 1968.

[54] G. Fransolet, G. Villers, and W. J. Masschelein. Influence of temperature

on bacterial development in waters. Ozone Sci. Engng., 7:205-227, 1985.

[55] H. J. Gold. Mathematical Modeling of Biological Systems-an Introductory

Guidebook. John Wiley & Sons, New York, 1977.

[56] R. C. Gonzalez and R. E. Woods. Digital Image Processing. Addison-

Wesley, New York, 1992.

[57] M. A. Goshko, H. A. Minnigh, W. O. Pipes, and R. R. Christian. Re-

lationships between standard plate counts and other parameters in water
distribution systems. Journal American Water Works Association, pages
568-571, Nov. 1983.
[58] K. Gotoh. Residual chlorine concentration decreasing rate coefficients for
various pipe materials. Water Supply, 1(2/3):SS 21-17 - SS 21-18, 1989.

[59] W. M. Grayman and R. M. Clark. An overview of water quality modeling in

a distribution system. In The 1990 Annual Conference Proceedings, pages
81-92, Cincinnati, Ohio, 1990. American Water Works Association.

[60] W. M. Grayman, R. M. Clark, and R. ~I. Xlales. Xlodeling distribution-

system water quality: Dynamic approach. Journal of lFatl'r Resource»
Planning and Management, 11--1(3):295-312, ~Ia~' 1988.

222
[61] W. Gujer and O. Wanner. A multispecies biofilm model. Biotechnology and
Bioengineering, 28:314-328, 1986.

[62] W. Gujer and O. Wanner. Modelling mixed population biofilms. In \Y. G.

Characklis and K. C. Marshall, editors, Biofilms. pages 397-443. J. Wilev
& Sons, New York, 1990. .

[63] C. N. Haas, M. A. Meyer, and M. S. Paller. Microbial alterations in water

distribution systems and their relationship to physical-chemical character-
istics. Journal American Water Works Association, pages 475-481, Sept.
1983.

[64] R. Haber and H. Unbehauen. Structure identification of nonlinear dynamic

systems-a survey on input/output approaches. Automatica, 26(--1):651-
677, 1990.

[65] E. E. Hargesheimer and C. M. Lewis. A practical guide to on-line particle

counting. Technical report, American Water Works Association Research
Foundation, Denver, Colorado, 1995. ISBN 0-89867-785-8.

[66] S. Hatukai, Y. Ben-Tzur, and M. Rebhun. Particle counts and size dis-
tribution in sytem design for removal of turbidity by granular deep bed
filtration. Water Science and Technology: A Journal of the International
Association of Water Polution Research, 36(4):225-230, 1997.

[67] J. Heraud, L. Kiene, M. Detay, and Y. Levi. Optimised modelling of chlorine

residuals in a drinking water distribution system with a combination of
on-line sensors. Journal Water Supply Research and Technology - Aqua,
46(2) :59-70, 1997.

[68] A. L. Hines and R. N. Maddox. Mass Transfer - Fundamentals and Appli-

cations. Prentice-Hall, Englewood Cliffs, NJ, 1985.

[69] D. M. Holt, C. A. Woodward, T. Rachwal, A. Delanaoue, and J. S. Col-

bourne. A pipe based experimental system to study factors influencing bac-
terial growth in London's water mains. In Proceedings of the Water Qual-
ity Technical Conference, San Francisco, 6-10th November 1994. American
Water Works Association, Denver.

[70] R. F. Hoskins. Generalised Functions. Ellis Horwood Ltd. Chichester. 1979.

[71] P. M. Huck. Measurement of biodegradable organic matter and bacterial

growth potential in drinking water. Journal of the American Wt'ater ~l'orks
Association, 82:78-86, July 1990.

[72] W. A. Hunt and J. R. Kroon. Model calibration for chlorine residuals in

distribution systems. In Water Quality Modeling in Distribution Systems.
pages 237-261, Cincinnati, Ohio. Feb. 1991. American Water Works Asso-
ciation.

223
[73] IHD-WHO Working Group on the Quality of Water. ~Fater Quality Sur-
veys. UNESCO/WHO, Paris/Geneva, 1978.

[74] Internatioal Water Supply Association. The practice of chlorination: Appli-

cation, efficacy, problems and alternatives. Journal ~rater Supply Research
and Technology - Aqua, 46( 4):inserted as The Blue Pages, Aug. 1997.

[75] J. Jass, J. W. Costerton, and H. M. Lappin-Scott. Assessment of a

chemostat-coupled modified Robbins device to study biofilms. Journal of
Industrial Microbiology, 15:283-289, 1995.

[76] T. A. Johansen and B. A. Foss. Empirical modeling of a heat transfer pro-

cess using local models and interpolation. In Proceedings of the American
Control Conference, pages 3654-3658, Seattle, Washington, 1995.

[77] R. Johansson. System Modeling and Identification. Prentice-Hall, Engle-

wood Cliffs, NJ, 1993.

[78] R. A. Johnson. Miller and Freund's Probability and Statistics for Engineers.
Prentice-Hall International, London, 1994.

[79] R. P. Kanwal. Generalised Functions: Theory and Technique. Academic

Press, New York, 1983.

[80] C. W. Keevil. Methods for assesing the activity of biofilm microorganisms in

situ. In J. W. T. Wimpenny, W. Nichols, D. Stickler, and H. Lappin-Scott
editors, Bacterial Biofilms and Their Control in Medicine and Industry,
pages 45-47. BioLine, Cardiff, 1994.

[81] M. S. Kennedy, S. Sarikelle, and K. Suravallop. Travel time calibration in

water distribution systems. In Water Quality Modeling in Distribution Sys-
tems, pages 219-233, Cincinnati, Ohio, Feb. 1991. American Water Works
Association.

[82] E. Kreyszig. Advanced Engineering Mathematics. John Wiley & Sons, New
York, 7th edition, 1993.

[83] M. W. LeChevallier. Microbial processes within the distribution system. In

Water Quality Modeling in Distribution Systems, pages 365-380, Cincin-
nati, Ohio, Feb. 1991. American Water Works Association.

[84] M. W. LeChevallier, T. M. Babcock, and R. G. Lee. Examination and

characterization of distribution system biofilms. Applied and Environmental
Microbiology, 53(12):2714-2724, Dec. 1987.

[85] M. W. LeChevallier, C. D. Cawthon, and R. G. Lee. Inactivation of biofilm

bacteria. Applied and Environmental Microbiology, 5--1(10):2492-2499, Oct.
1988.

22--1
[86] M. W. LeChevallier, C. D. Lowry, and R. G. Lee. Disinfection of biofilms in
a model distribution system. In M. \\T. LeChevallier, B. H. Olson. and G. A.
McFeters, editors, Assessing and Controlling Bacterial Regrowth in Distri-
bution Systems, chapter 1, pages 1-47. American Water Works Association
Research Foundation, 1990.

[87] P. Lindskog and L. Ljung. Tools for semi-physical modelling. In

T. Soderstrom and M. Blanke, editors, Proceedings of the IFAC Sympo-
sium on System Identification, volume 3, pages 1199-1204, Copenhagen,
Denmark, 1995. Pergamon.

[88] C. P. Liou and J. R. Kroon. Modeling the propagation of waterborne

substances in distribution networks. Journal of the American Water Work...
Association, pages 54-58, Nov. 1987.

[89] L. Ljung. System Identification: Theory for the User. Prentice-Hall. En-
glewood Cliffs, NJ, 1987.
[90] L. Ljung and T. Glad. Modeling of Dynamic Systems. Prentice-Hall, En-
glewood Cliffs, NJ, 1994.
[91] C. Lu, P. Biswas, and R. M. Clark. Simultaneous transport of sub-
strates, disinfectants and microorganisms in water pipes. Water Research,
29(3) :881-894, 1995.

[92] M. Lungwitz, S. Hopfgarten, K. Chen, M. Brdys, and H. Puta. Modelling

and simulation of chlorine concentration in water supply systems for control
purposes. In Proceedings of the 39th International Scientific Colloquium,
Ilmenau, Germany, Sept. 1994.
[93] D. Maier, M. Maier, and J. Ulmer. Automatische WassergiitemeBstationen;
Teil I: Stand und Entwicklung von Meflgeraten zur kontinuierlichen
Wassergiiteiiberwachung. Wasser - Abwasser, 136(5):242-249, 1995.

[94] S. H. Maier, S. Kaattari, and M. E. Magana. Analysis of limiting dilution

assays conducted under non-saturating conditions. Journal of Theoretical
Biology, 179(2):173-177, 1996.

[95] S. H. Maier, M. E. Magana, R. R. Mohler, and S. Kaattari. Modified limit-

ing dilution assay for cell systems. Applied Mathematics and Computation,
78:187-195, 1996.

[96] S. H. Maier and R. S. Powell. A systems approach to modelling chlorine

deacy in water distribution systems. In Proceedings of the Doctoral Research
Conference 1996, Uxbridge, 19-20 June 1996. Department of Manufact ur-
ing and Engineering Systems, Brunel University.

[97] S. H. Maier, R. S. Powell, and C. A. Woodward. Behaviour of monochlo-

ramine in a pipe test rig. In R. S. Powell and K. S. Hindi, editors, Pro-
ceedings of the International Conference on Computing and Control [or ilu:
Water Industry, Uxbridge, September -1-5 1997. Research Studies Pross.

225
[98] S. H. Maier, C. A. Woodward, A. Delanoue, D. ~I. Holt. S. ~I. ~Ic~lath,
and R. S. Powell. Relationships between total chlorine, particle counts
and biofilm detachment in a drinking water distribution system. In C. \Y.
Keevil, editor, Proceedings of the International Conference on Biofilms in
Aquatic Systems, Coventry, April 13-16 1997. Royal Society. of Chemistrv,
.
[99] R. M. Males. Graph-theoretic approaches to hydraulic network issues. In
Water Quality Modeling in Distribution Systems, pages 95-109. Cincinnati,
Ohio, Feb. 1991. American Water Works Association.

[100] K. Marshall, editor. Microbial Adhesion and Aggregation, Dahlem K on-

ferenzen, Berlin, 1984. Springer-Verlag.

[101] R. E. Martin, M. Y. Ramirez, and V. P. Olivieri. Attachment of bacteria to

surfaces in drinking water distribution systems. In ASM Meetinq'S'). 1987.

[102] B. Massey. Mechanics of Fluids. Chapman & HalL London, 1995.

[103] S. M. McMath, A. Delanoue, D. M. Holt, A. H. L. Chamberlain, and B. J.

Lloyd. The optimization of an epifluorescence microscopy technique for the
direct counting of clumped bacteria. Journal of the Chartered Institution
of Water and Environmental Management, 12(2):113, April 1998.

[104] S. M. McMath, A. Delanoue, D. M. Holt, S. H. Maier, and C. A. Woodward.

Importance of "clumps" shed from biofilms in a water distribution system.
In C. W. Keevil, editor, Proceedings of the International Conference on
Biofilms in Aquatic Systems, Coventry, April 13-16 1997. Royal Society of
Chemistry.

[105] R. Murga, P. S. Stewart, and D. Daly. Quantitative analysis of biofilm

thickness variability. Biotechnology and Bioengineering, 45:503-510, 1995.

[106] L. A. Nagy and B. H. Olson. Occurence and significance of bacteria, fungi

and yeasts associated with distribution pipe surfaces. In Proceedings of
the Water Quality Technical Conference, Houston, Texas, 1985. American
Water Works Association, Denver.

[107] J. A. Nelder and R. Mead. A simplex method for function minimization.

Comput. J., 7:308-313, 1965.

[108] O. Nelles, O. Hecker, and R. Isermann. Automatic model selection in 10-

cal linear model trees (LOLIMOT) for nonlinear system identification of a
transport delay process. In Y. Sawaragi and S. Sagara, editors, Prepriuts
of the 11th IFA C Symposium on System Identification, volume 2, pages
727-732, Fukuoka, Japan, 1997. Pergamon.

[109] C. D. Norton and M. W. LeChevallier. Chloramination: its effect 011 dis-

tribution system water quality. Journal of the American lrater lr01'k..;
Association, 89(7):66-77, July 1997.

226
[110] P. T. O'Shaughnessy, M. G. Barsotti, J. \Y. Fay. and S. \Y. Tighe. E\'al-
uating particle counters. Journal American Water Works Association,
89(12):60-70, Dec. 1997.

[111] A. Ostfeld and U. Shamir. Optimal operation of multiqualitv distribution

systems: Steady state conditions. In Water Quality Modeling in Distribu-
tion Systems, pages 111-129, Cincinnati, Ohio, Feb. 1991. American Water
Works Association.

[112] P. J. Packer. Water distribution modelling effect on advanced water treat-

ment. Final Report Project R92/C2, Physiology & Biofilms Section, Le-
gionella & Aerobiology, Pathology Division, PHLS-CA~IR (Centre for Ap-
plied Microbiology and Research), Porton Down, Salisbury, Wilts, SP-l OJT.
Mar. 1994.

[113] B. M. Peyton and W. G. Characklis. Kinetics of biofilm detachment. Poster

presented at IAWPRC Conference - Water Quality International'92, Wash-
ington D.C., May 24-30 1992.

[114] B. M. Peyton and W. G. Characklis. Kinetics of biofilm detachment. IVater

Science and Technology, 26(9-11): 1995-1998, 1992.

[115] B. M. Peyton and W. G. Characklis. A statistical analysis of the effect of

substrate utilization and shear stress on the kinetics of biofilm detachment.
Biotechnology and Bioengineering, 41:728-735, 1993.

[116] P. Piriou. A new tool for the study of the evolution of water quality in
distribution systems: design of a network pilot. In Proceedings of the Water
Quality Technical Conference, New York, 19-23rd June 1994. American
Water Works Association, Denver.

[117] P. Piriou, S. Dukan, and Y. Levi. Piccobio: Modelling bacterial regrowth

in drinking water distribution systems. In R. S. Powell and K. S. Hindi,
editors, Computing and Control for the Water Industry. Research Studies
Press Ltd., Taunton, England, 1998.

[118] P. Piriou, Y. Levi, J. Heraud, and L. Kiene. New tools and applications
in modelling and monitoring water quality in drinking water distribution
systems. Water Supply, 15(2):119-135, 1997.

[119] R. S. Powell. WASMACS functional description. Technical Report 3.4.

WASMACS Ltd., Control Engineering Centre, Brunel Universitv. CB8 3PH.
U.K., Sept. 1994.

[120] N. M. Ram and J. P. Malley Jr. Chlorine residual monitoring in the presence
of N-organic compounds. Journal of the American Water IForks Associa-
tion, pages 74-81, Sept. 1984.

')2-
... I
[121] M. Reike, K. H. Fasol, G. Sinai. and D. Pessen. Hierarchical control of
multi-quality water distribution systems. Computers and Electronics in
Agriculture, 2:1-13, 1987.

[122] B. E. Rittman. Detachment from biofilms. In \Y. G. Characklis and P. A.

Wilderer, editors, Structure and Function of Biofilms. pages 49-57. J. Wilev
& Sons, New York, 1989. .

[123] L. A. Rossman. The EPANET water quality model. In B. Coulbeck, editor.

Integrated Computer Applications in Water Supply. Volume 2 - Applica-
tions and Implementations for Systems Operation and Management, pages
79-93. Research Studies Press Ltd., Taunton, England, 1993.

[124] L. A. Rossman. EPANET User's Manual, EPA-600/R-94/057. Risk R~­

duction Engineering Laboratory, U.S. Environmental Protection Agency.
Cincinnati, Ohio, Jan. 1994. Version 1.1.

[125] L. A. Rossman and P. F. Boulos. Numerical methods for modelling water

quality in distribution systems: A comparison. Journal of Water Resources
Planning and Management, 122(2):137-146, March/April 1996.

[126] L. A. Rossman, P. F. Boulos, and T. Altman. Discrete volume-element

method for water-quality models. Journal of Water Resources Planning
and Management, 119(5):505-517. September/October 1993.

[127] SAFEGE, Consulting Engineers. Piccolo 3.10 Reference Manual, Novem-

ber/December 1992.

[128] S. Schneider, M. A. Brdys, H. Puta, and S. Hopfgarten. Modelling of

chlorine concentration in water systems directed to integrated operational
control. In B. Coulbeck, editor, Integrated Computer Applications in Water
Supply, pages 143-159, 1993.

[129] P. Servais, P. Laurent, and G. Randon. Comparison of the bacterial dynam-

ics in various French distribution systems. Journal Water Supply Research
and Technology - Aqua, 44(1):10-17, 1995.

[130] U. Shamir and C. D. D. Howard. Water distribution system analysis. Jour-

nal of the Hydraulics Division, 94(HY1):219-234, Jan. 1968.

[131] U. Shamir and C. D. D. Howard. Topics in modelling water quality in distri-

bution systems. In Water Quality Modeling in Distribution Systems, pages
183-192, Cincinnati, Ohio, Feb. 1991. American Water Works Association.

[132] W. W. Sharp, J. Pfeffer, and M. Morgan. Insitu chlorine decay rate' tpst-
ing. In Water Quality Modeling in Distribution Systems, pages 311 -321.
Cincinnati, Ohio, Feb. 1991. American \\Tater \Yorks Association.

[133] R. Sinha, P. Gupta, and P. K. Jain. Water quality modelling of a citv water
distribution system. Indian J. Environ. Hlth .. 36(4) :258 -262. 1994.

228
[134] J. Sjoberg, Q. Zhang, L. Ljung, A. Benveniste. B. Deylon, P.-Y. Glorennec.
H. Hjalmarsson, and A. Juditsky. Xon-linear black-box modelling in system
identification: a unified overview. Automatica, 31(12):1691-172-1. 1995.

[135] T. Soderstrom and P. Stoica. System Identification. Prentice-Hall. Engle-

wood Cliffs, N J, 1989.

[136] M. Somville and G. Billen. A method for determining exoproteolytic activ-

ity in natural waters. Limnol. Oceanogr., 28:190-193. 1983.

[137] P. S. Stewart. A model of biofilm detachment. Biotechnology and Bioeruii-

neering, 41:111-117,1993.

[138] P. Stoodley, A. Cunningham, Z. Lewandowski, J. Boyle. and H. Lappin-

Scott. Biofilm structure and influence on biofouling under laminar and
turbulent flows. In C. Keevil, editor, Proceedings of the International Con-
ference on Biofilms in Aquatic Systems, Coventry, April 13-16 1997. Royal
Society of Chemistry.

[139] P. T. Strait. Probability and Statistics with Applications. Harcourt Brace

Jovanovich, London, 2nd edition, 1989.

[140] T. Ta. Analysis of the particle counts and the chlorine decay in PIPE RIG
during the step flow trials. Technical report, Thames Water Utilities Ltd.,
Group R&D, Spencer Hse, Manor Farm Rd, Reading, Berks RG2 OJN, July
1995.

[141] P. M. Thomas. Chloramination of water supplies. Progress report no.l,

Urban Water Research Association, State Water Laboratory, Engineering
and Water Supply Department, S.A., Australia, Jan. 1988.

[142] R. Todd. Reaction of nitrite with chlorine species. Experimental report.

Operational Science, Thames Water Utilities Ltd., Wraysbury, Aug. 1995.

[143] R. E. Thain. Quality Criteria for Water. Castle House Pub., London, 1979.

[144] M. G. Trulear. Dynamics of biofilm processes in an annular reactor. xlas-

ter's thesis, Rice University, Houston, Texas, 1980.

[145] M. G. Trulear and W. G. Characklis. Dynamics ofbiofilm processes. Journal

WPCF, 54(9):1288-1301, Sept. 1982.

[146] H. Unbehauen. Regelungstechnik III. Vieweg Verlag, Braunschweig. 5th

edition, 1995.

[147] H. Unbehauen. Some new trends in identification and modeling of nonlinear

dynamical systems. Applied Mathematics and Computation. 78:279 -297.
1996.
[148] H. Unbehauen and G. P. Rao. Identification of Continuous Systems. Xort h
Holland, Amsterdam. 1987.

229
[149] D. van der Kooij and H. R. Veenendaal. Assessment of the biofilm formation
characteristics of drinking water. In Proceedings of the American lrater
Works Association Water Quality Conference, pages 1099-1110, Toronto.
Canada, November 15-19 1992.

[150] D. van der Kooij and H. R. Veenendaal. Biofilm development on surfaces in

drinking water distribution systems. Water Supply, 12(1/2):SSl-1 - SSl-7.
1994. Special subject: Biological activity in distribution systems.

[151] E. van der Wende and W. G. Characklis. Biofilms in potable water dis-
tribution systems. In A. McFeters, editor, Drinking Water Microbiology,
chapter 12, pages 249-268. Springer-Verlag, 1990.

[152] E. van der Wende, W. G. Characklis, and D. B. Smith. Biofilms and bac-
terial drinking water quality. Water Research, 23: 1313-1320, 1989.

[153] J. J. Vasconcelos, L. A. Rossman, W. M. Grayman, P. F. Boulos. and

R. M. Clark. Kinetics of chlorine decay. Journal American Water Works
Association, 89(7):54-65, July 1997.

[154] O. Wable, N. Dumoutier, J. P. Duguet, P. A. Jarrige, G. Gelas, and J. F.

Depierre. Modelling chlorine concentrations in a network and applications
to Paris distribution system. In Water Quality Modeling in Distribution
Systems, pages 77-87, Cincinnati, Ohio, Feb. 1991. American Water Works
Association.

[155] E.Walter and L. Pronzato. Identification of Parametric Models. Springer-

Verlag, London, 1997.

[156] G. C. White. Handbook of Chlorination and Alternative Disinfectants. John

Wiley & Sons Inc., Taunton, England, 1998.

[157] J. W. T. Wimpenny. The spatial organisation of biofilm. In J. \\'. T.

Wimpenny, W. Nichols, D. Stickler, and H. Lappin-Scott, editors, Bacterial
Biofilms and Their Control in Medicine and Industry, pages 1-6. Biol.ine,
Cardiff, 1994.
[158] S. Wolfram. The Mathematica Book. Wolfram Media and Cambridge Cni-
versity Press, Cambridge, Mass., 3rd edition, 1996.

[159] C. A. Woodward, C. T. Ta, J. Colbourne, and D. ~1. Holt. Behaviour of

particles in a large scale experimental pipe system. In Proceedings of the
Water Quality Technology Conference, New Orleans, 12-16th :\ow'mbpr
1995. American Water Works Association.

[160] C. A. Woodward, C. T. Ta, D. 1\1. Holt, and J. Colbourne. Relationship be-

tween observed monochloramine decay rates and other physical and chemi-
cal parameters. In Proceedings of the Water Quality Technology Conference,
New Orleans, 12-16th November 1995. American Water Works Association.

230
Appendix

231
Appendix A

Model-Relevant Data

This appendix contains graphs of the data most relevant for the development of
a water quality model. In particular it covers the particle counter data for the
twelve step flow trials of the 1996 experiment. Both total particle counts and size
distributions are included. This appendix is meant as a reference to allow the
visual verification of the characterisation of data as given in chapter 4, which is
the basis of the models obtained in chapter 6.
All graphs given in this appendix are generated with the Matlab function
plotallpc that is part of the Water Quality Toolbox.

A.I Total Particle Counts

Observations from the total particle counter data presented in this section are
given in section 4.1 on page 67.
Figs. A.l to A.12 present total particle counts, i.e. the number of all particles
> 2J..Lm, for inlet shed (if appropriate), sample points L 2 and 3 together with

flow rate (for reference) versus time. All graphs are split into four plots according
to sample point. Note that some of the peaks are not visible in these graphs as
the graphs have been scaled to show the slow pattern better. All t irnes are in

days since 1 Januarv 1995.

Note that after step down at of 8FT 1 (fig. A.l) there was a treatment plant
failure. Therefore the flow rate is interrupted (i.e. zero) for several hours.

232
 Sample Point 1
1200 ,------,----n-~-----~-----,
E
~ 1000
...
Q)
/I

800
Qi
E
ltl
'6 600
'0
l/l
Q) 400
U
t
ltl
n,
200
~'--- - - - - ~==-=="""""""'"'"":--I.....r--~
O'-----~--~--~...Joo..1.:...----.J
572 ,2 572 .4 572 .6 572 .8 573
Time

Sample Point 2 Sample Point 3

1000 .--------rr--r-r.----~--~--~ 1000
E E
:l
:l
N 800 N 800
...
/I
Q)
... /I
Q)

Qi 600 Qi 600
E E
ltl ltl
'6 '6
'0 400
IN ' '0 400 "

~
l/l

u
~
l/l

o
w
---»~-
t 200 " '/ " :'
t
ltl
200
ltl
n, o,
_ _ _ , _ _ _ _ , '=-:::::-:::-~~,.-----l - - - ' -
o ' ' o T

572.2 572.4 572 .6 572 .8 573 572 .2 572 .4 572 ,6 572.8 573
Time Time

Figure A .I : Step fl ow tria l 1, 25 July 1996: total particle counts data during
high fl ow period; tim e is in days since 1 January 1995 fl ow (dashed lin ) is
multip lied by 100 and included for reference. Inlet particle counter data wa. not
yet installed.

233
 Sample Point 1

E 80
::J
C\J
1\
Qj 60
Qj
E
<tl
'6 40
'0
C/)
Q)

~ 20
<tl
a..
OL.----'------'----~ _ ___.J
579 .2 579 .4 579 .6 579 .8 580
Time

Sample Point 2 Sample Point 3

80 .------n-"""T"""T"rr---~--__,_--____. 80 r---......".~_ur____~--~--__,
E E
::J ::J
C\J
~ C\J ~",,-.
1\ 60 . . .. """"" . . _. - :- . - - '- ' JI ' , ' 1\ 60 " ;/. -

-
....
Q)
Q)
E
.!!! 40
I
I
I
....
Q)
Qj
E
.!!! 40
I
I
I
"0 I "0 1
'0 1 '0 I
C/) C/)
Q) I . . . I Q) I
:2 20 . . . . ..... . . . . . . . . . . . . . . . . .. ·. · 1··· · · ·· · :2 20
t::: I t:::
<tl <tl
a.. a..
0'-- --'--------'-------'--------' OL.-----'----~--~-------.J
579.2 579.4 579.6 579 .8 580 579 .2 579 .4 579.6 579 .8 580
Time Time

Figure A.2: St ep flo w trial 2, 1 August 1996: total particle counts data during
high flo w period; time is in days since 1 January 1995, flow (dashed lin ) i.
multiplied by 100 and included f or reference. Inlet particle counter dauiu a not
yet installed.

23
 Inlet Particle Counts, D>2um Sample Point 1
35 ,------,-~----.---~---,----, 70 r---r,--".---~--~------,
E . .
~ 30
., '" ..
§ 60
C\J
"
0 . 25
lJ)
~ 50
Q) :
I
I
§o 20 ~ell 40
o
Q)
13
15 ~o 30
.€ lJ)
10 ~
o 20
ell
n,
'e
]i 5 ~ 10 _ J L _ _
c:
OL------'--------'------'---..!.....------.J O'------'--------'---~------J
585.2 585.4 585.6 585 .8 586 585 .2 585 .4 585 .6 585.8 586
Time Time

Sample Point 2 Sample Point 3

70 ,----rr--...,,-----.---~-------, 150 r---l~-~..-___,r~--~----,

§ 60 . . . .. .1.. . . . E
1- :::J
C\J C\J
. I
~ 50 " .
Q)
J.
I
: I "
(i) 100
a; .: .. .1. . . ..
a;
E 40 " · 1· E
ell I ell
I :.c
~o 30 '"

'0
lJ) I I ~ 50 . ' 1'
~ 20....,....,....'- .. . . .I' .
13 ,
o
tell 10 " " "I " " .. . . . I . tell
. . . . . . . . L. . _ _
a.. _ J c,
_ J
O l - - - - - ' - - - - - - - - ' - - - - - - ' - - -----l O '------'--------'---~----J
585.2 585.4 585.6 585.8 586 585 .2 585 .4 585.6 585.8 586
Time Time

Figure A.3: St ep flow trial 3, 7 August 1996: t ot al part icle coun ts dat a during
high flow period; tim e is in days since 1 Ja n ua ry 1995, fl ow (dashed lin ) is
multiplied by 100 and in clu ded for referen ce. In let parti cle coun ter data was not
yet com pute r logged (hand-logg ed data is sh own) .

235
 Inlet Particle Counts , D>2um Sample Point 1
80 r-----.---~--~--~,..---~ 500 .----IT~--~--____._,r---~
E E
:::l
C\J :::l
C\J 400
~ 60
...
1\
Q)

Qi 300
E
Ctl
'6
'0200
Vl
Q)
U
t 100 -- - - - - - - - - ,
Ctl
o,
. L _
OL.----=---'------'-------'--------.J
600 .2 600.4 600 .6 600 .8 601 600. 4 600 .6 600.8 601
Time Time

Sample Point 2 Sample Point 3

1000 r---rr-'""T"r"--~--~--~

E E 1200
:::l :::l
C\J 800 ~ 1000
...
1\
Q)
...
Q)

Qi 600 Qi 800
E E
Ctl Ctl
'6 '6 600
'0 400 '0
(/) Vl
~ ~ 400
o o
t 200 t
Ctl
o,
Ctl
o, 200
- :- - - - :- - - - :'
O==---'------'-------'-.h....=~...-=.J
600.2 600.4 600.6 600 .8 601 600.4 600 .6 600.8 601
Time Time

Figure A.4: St ep fl ow tria l 4, 22 August 1996: total particle counts data during
high fl ow period; tim e is in days since 1 January 1995, fl ow (dashed lin ) is
multiplied by 100 and included for reference. Inlet particle counter data U a not.
yet computer logged (hand-logged data is shown) .

:..36
 Inlet Particle Counts, D>2um Sample Point 1
150 r---,----rT",---~----"..---~

E
:l
N
1\
en CD 100
§1O
o
Q)
E
ctl
o '5
<1l
U '0
~ 5 . ... lB 50

-
a..
C
<1l
u
t
ctl
a..
\... - --
o'--------'------'-------'--------.J o ' - - - - - - ' - - - - - - ' - - - - - - - ' - - ------.J
606.2 606.4 606.6 606.8 607 606 .2 606.4 606 .6 606 .8 607
Time Time

Sample Point 2 Sample Point 3

200 .----Ir--rr------.----~-------, 300 .----......----...---~--~--~

E E
:l
N
~ 250
1\ 150 . . .
... 1\

-<1l
<1l
E
.!!! 100
CD
Q)
E
~ 150
200

-
"0
o
en
<1l
'0
lB 100
I '
U 50 . . U
'e
ctl ~ 50 I
a.. a.. I·
~---
oc......:=-----'------'-------=:.....=......:=--=::.......::J
606.4 606 .6 606 .8 607 606 .2 606.4 606 .6 606 .8 607
Time Time

Figure A.5: St ep fl ow trial 5, 28 August 1996: total particle counts data during
high flow period; time is in days since 1 January 1995, flow (dashed lin ) is
multiplied by 100 and included f or ref erence. Inlet particle counter data 11 a, not
yet co mputer logged (han d-logged data is shown) .

-
') 3-{
 Inlet Sample Point 1
200 ,-----rT---.----~----.------, 500
E E
:J :J
N
1\ 150 .... . . .. N 400
...
Q)
...
1\
Q)
Qj Qj 300
E . .: . E
.!!1 100 . ctl

-
". '

u _ . t, ..:.. _ _ - ,
'5
o '0 200
rJ) rJ)
Q) Q)
U 50 . . .. U
t t 100
ctl ctl
a.. a..
0'---'------'-------'---.....0....------1 0
647 .2 647.4 647 .6 647.8 648 647.2 647.4 647.6 647.8 648
Time Time

Sample Point 2 Sampl e Point 3

500 .-----.---,.,-----.---~------, 150
E E
:J
:J
N 400 . N

...
1\
Q)
1\
Q3 100 .. .. .. . ..
Qj
Qj 300 E ,V\_ - -,,--,:,---,
E
ctl
'5 ~ I
'0200 '0 I
rJ)
Q)
~ 50 . ·· 1
U U I
t 100 .
ctl ~ I
a.. a.. I
O=--;;;;,.",::;_--'--_ _----'-_ _-=....;:::L...:= --=-=.J o t:::=:z:::::!...._ - - ' - -_ _ ~ ~ _ _ ...J

647.2 647.4 647 .6 647.8 648 647.2 647.4 647.6 647.8 648
Time Time

Figure A.6: St ep flo w trial 6, 8 October 1996: total particle counts data duruiq
high flow period; time is in days since 1 January 1995, flow (da hed lin ) is
multiplied by 100 and included fo r ref erence.

23
 Inlet Sample Point 1
80 ,------rr-~~-~-.___.___~--___,
1 :
E E 250
:J 1 :J
N N
.. .. .. ~ : .
... 60
. . . . .
~ 200
1\
1
.$ Q)
Q)
1
Q)
E
.~ 40 ~ 150
"0 '6
'0
en
Q)
~
Q)
100
:2 20 U
t
nl ~ 50
a.. \.- :- -- - - a..
O'----'------'-------'---............- - - - - . J - - _= -"'"-_I
654.2 654.4 654.6 654 .8 655 654.4 654 .6 654.8 655
Time Time

Sample Point 2 Sample Point 3

1 00 .-----.----rr--.-----~----.---------, 350 ,----,.,--rr----~--~----,

E I.... , A"' .. \. ... J~ ....... \._

-.II . I § 300
~ 80 .. .. I . N
1
I
...
1\
Q) 1
I ~ 250
Q)

~ 60 " '1
.. . j . . .: . ..
~nl 200
~
1

~o
1

'0 40 . ... 1. 1 150

en I en
Q)
~ 100 - .;. _ r - - -e- - - --,
U 1 o
.~ 20 .. T t
a.. g: 50
o ""--.....L._-'--_ _- - " -"--_ _---I
\... - -- - -

654.2 654.4 654.6 654 .8 655 654.4 654 .6 654 .8 655

Time Time

Figure A.7: St ep fl ow tria l 7, 15 October 1996: total particle co unts data duriuq
high fl ow period; tim e is in days since 1 January 1995, flow (dashed lin ) L'
multip lied by 100 and included f or ref erence.

239
 Inlet Sample Point 1
80 150
E I
:::::l
E
:::::l
C\I
I
C\I
.. . .........
... 60
1\
Ql
i
I
1\

2 100
Q) Ql
E I
E
.~ 40 ell
"6
,_"__... _.-.....
"0
'0 '0
1Il 1Il 50 .. .
Ql Ql
:§ 20 .U-e
t:: ./

ell ell
a.. a..
0 0
663 .2 663.4 663.6 663.8 664 663.2 663.4 663.6 663.8 664
Time Time

Sample Point 2 Sample Point 3

... . . . . ..
E 150 E 200
:::::l :::::l
C\I C\I

...
1\
Ql
1\
Q; 150
Q) 100 Q)
E E
ell ell
"6 "6 100
'0 '0
1Il 1Il
Ql 50 Ql
U U
.-e 50
t
ell
a..
ell
a.. -- - --
0 0
663 .2 663.4 663.6 663.8 664 663.2 663.4 663.6 663.8 664
Time Time

Figure A.8: St ep flow trial 8, 24 October 1996: total particle counts data during
high flow period; time is in days since 1 January 1995, flow (dash d linc) is
multiplied by 100 and included f or ref erence.

240
 Inlet Samp le Point 1
50 r--.--rr-------.--.--------,----..---------, 150 ,---,----.-~r--_____r_.---____r___...._--~

E
::J
E
::J
N 40 N

....1\ 1\
Q) Qi 100
Q) 30 Q)
E E
C'Il C'Il
'5 '5
'020 '0
C/)
Q) ~ 50
u U
:eC'Il 10 :eC'Il
a.. a..
O'-----~--~------L~------.J
669.4 669 .6 669.2 669.4 669 .6 669 .8 670
Time Time

Sample Point 2 Samp le Point 3

150 .-----.--~-----,.---~--
--, 200 .-------c----~---~-------,

E E
::J
::J
N N
1\ 1\ 150
....Q)
-Qi 100
Q)
E
C'Il
I
I
IL Q)
E
~ 100
'5
'0 I '0
~ 50 .. . I. C/)
Q)
U 1 U 50
:e -'- .€
C'Il
C'Il
a.. a..
OL-_ _--'-_ _----' .L.-_ _---.J

669.2 669.4 669 .6 669.8 670 669.4 669 .6

Time Time

Figure A.9: St ep fl ow tria l 9, 30 October 1996: total particle counts data during
high fl ow period; tim e is in days since 1 January 1995, flow (dashed lin ) is
multiplied by 100 and in cluded f or ref erence.

2-11
 Inlet Sample Point 1
12 ,---~-----or----~-------,
40
E E
=:l =:l
C\l C\l
10 .. .
...A
Q)
~
Q)
30 . \. ,l.. ,, _ ~

Qj Qj
E E
co
~ 8 i5 20 ..
'0 's
(f) (f)
Q) Q)
13 6 :§ 10
'e t
co co
a.. a..
4L..11...-.L...----"""----------'-------.J o'--.L...- -'-- ~ _.J

704.4 704 .6 704 .8 705 704.4 704 .6 704 .8 705

Time Time

Sample Point 2 Samp le Point 3

50
E 60 . E
=:l =:l

~ 40
C\l
A 50 ...
Q)

~co 30
i5
020
(f)
Q)
13
'e
co 10
a..
OL..L..-"-----"""----------'--------J
o LL-"-- -'-- ~ _l

704.4 704.6 704.8 705 704.4 704.6 704 .8 705

Time Time

Figure A.IO: Step flo w trial 10, 4- 5 December 1996: total particle count.' data
during high fl ow period; tim e is in days since 1 January 1995, flow (da li d lin )
is multiplied by 100 and included f or ref erence.

2-12
 Inlet Sample Point 1
30 30 rnr------,.--rr----~-~-------.
E E
~ 25 ~ 25
A A
Q; 20 .. ., .. Q; 20
a> a>
E E
.~ 15 .s 15
-
"0
o
.. ..
"0
'0
~
l/)

o
10

~ ll'~ - I.. L.L : . Il',lU ... I ,I ~ U

~ 10

'eC'll
~
0-
5 .- ' ~ :"' . :
5 ' 1 ..
I: 0-
r~
1
o 0 l......J.--~--~----"-----.::..l
710.5 711 711.5 712 712 .5 710.5 711 711 .5 712 712.5
Time Time

Sample Point 2 Samp le Point 3

30 r--1rrT""-----.----,r-r-----.----.-----.--~-., 40 ,---,rrr--~-"'T'T""---~----.-.

E E
~ 25 ::J
C\J
A
Q; 20
... 30
A
Q)

a> a>
E E
.~ 15 .~ 20
"0 "0
'0 '0
l/) 10 " l/)
Q)
~
o ~ 10
tC'll 5 t::
C'll
0- I 0-
r
o L...l....o _ _------'_ _ ------' _ _------' _ _-----"---l
O O"""--~--~----"---~
710.5 711 711 .5 7 12 7 12.5 710 .5 711 71 1.5 712 712 .5
T ime Time

Figure A .II: St ep flow trial 11, 10-1 2 December 1996: total particle count, '
data during high flow period; tim e is in days since 1 January 1995, flow [da It d
line) is multiplied by 10 and included f or reference.

2-13
 Inlet Sample Point 1
10 r-r~"TT""----.-...----.---,-,-----.---.....,..----r-----,
E E 15
:::J :::J
C\J 8 C\J

...
1\
Q.l
1\
Q;
Qj
E
6 ..
~ 10
C1l C1l
:0 ,.~ _ ~ _ .:.. _ -_.....• - ~. _ «- I. :0
'0 4 · 1.. ' " j '0
(J) (J)
I 1
~
o
2 I ... . I
~ 5 ,- ----------- - --- .- -- -- ---- - -.
~ ~
t 'e I
C1l I I C1l I .
a. a. I
I - I
r
717.5 718 718.5 719 719.5 720 717 .5 718 71 8.5 719 719 .5 720
Time Time

Samp le Point 2 Sample Point 3

. '. ' .. . . . ; .. ..
E 12 E 15
:::J :::J

~ 10 . .. . ' ., ' .: . . C\J

...
Q.l
1\

Qj 8
E
C1l
:0 6
'0
(J)
Q.l 4 .1
U
'e 2 . I . ~ : : .
C1l
a. I :
O L....L..-'--_---'-_ _............_----'-_ _--'------=..L- - l o '-'--'---~--~-~--~~-'

717.5 7 18 718.5 719 7 19 .5 72 0 717 .5 718 718 .5 719 719 .5 720

Time Time

Figure A.12: Step flow trial 12, 17- 19 Decem ber 1996: total particl count.
data during high flo w period; tim e is in days since 1 January 1995, flow (do. h d
line) is multiplied by 10 and included f or reference.

244
A.2 Size Distributions of Particle Counts
Observations from the particle counter size distributions presented in this section
are given in section 4.2 on page 74.
The figures A.13 to A.24 show particle counter size distributions, one subplot
per sample point, with percentages on the y-axis and time in days since 1.1.1995
on the x-axis. Flow is given for reference as a dashed line, its value in litre
per second is multiplied by 100 to fit on the same scale as the percentages. In
addition, a vertical dashed-dotted line marks the elapse of one travel time since
the start of the step flow trial to the respective sample point. The lines depicted
on the graphs show (from larger to smaller values) particles in the range 2-5Jllll.
5-l0{lm, 10-15{lm and l5-20{lm (cf. fig. 4.3 on page 75).

245
 Sample Point 1
100
(/)
Q)
OJ
c:: 80
ltl
....
Q)
u I
'f : 60 1..1 - I - ~ - - - - ... - _
ltl
a. 1 I
(;
Q) 40 ... . ....... 1 . . . 1
OJ 1 I I
ltl
C
Q)
1 I
~
20
Q)
a.
- - --
0
572 572.2 572.4 572.6 572.8 573
time

Sample Point 2 Sample Point 3

100 100
(/) (/)
Q) Q)
OJ OJ
c:: 80 c:: 80
ltl ltl
....
....
Q) Q)
u U
t
III
a.
60 ~ r - .:. . :- -".. _::.- 'f :
ltl
a.
60 - ~'l - :.. - . . - -
I I
(; I (;
. ... . ... . . . .. .. J. 40 I
Q) 40 ","
Q)
OJ I I OJ I
III III
C I C
Q)
o
....
Q)
20 . . .. . .. ...
' 1'
Q)
~
Q)
20 :j
a. a.
----
0 0
572 572 .2 572.4 572.6 572.8 573 572 572.2 572.4 572.6 572.8 573
time time

Figure A.13: St ep flo w trial 1, 25 July 1996: particle counter size distribution. ;
time is in days sin ce 1 January 1995, flow (dashed line) is mult iplied by 100.
the elapse oj one travel tim e is indicated by a vertical dashed-dotted lin both ar
included Jar reference. Inlet particle counter size distribution data i not a' ailoble
Jar this step fl ow tria l.

246
 Sample Point 1
100
l/)
Q)
Ol
c: 80
~
Q) I '
u \I '
'e 60 1" ,- '- :- -_ r
Cll I
a. I I
'0
Q) 40
I,! ,. ' .'
I
'I'
Ol I I
Cll I
C I I
Q) I
...o
Q)
20
I
a.
0
579 579.5 580 580.5 581
time

Sample Point 2 Sample Point 3

100 I 100
l/)
Q)
1
,. l/)
Q)
Ol
Ol 1.
!1':' , c:
~ 80 "'1: Cll 80
...
Q) I Q)
U
t 60
I _L:_ -
I" - '~
U
t 60
\ I
- 1" - '- .- - -' "
Cll I Cll
a. I I a. I I I
'0 I
I I '0 I I I
Q) 40 ' ,'
I' ' ,'
Q) 40 ' I . . -. I '"

-Ol
Cll
c:
~ 20
I I
I I
f ' L -: '
I
I
Ol
Cll
C
Q)
...
o 20
I
,, '
I
I
Q)
a.
'T
- -
I
I

- ~ '

- -
r
... .....
,~
Q)
a.
I
I
o 0
579 579.5 580 580.5 581
579 579.5 580 580.5 581
time time

Figure A .14: Step flow trial 2, 1 August 1996: particle counter size di. iribu-
tions; time is in days since 1 January 1995 flow (dashed line) is multipli d by
100, the elapse of one travel time is in dicated by a vertical da shed-dott d lin .
both are inclu ded fo r reference. Inlet particle counter size distribution data is not
available for this step flow trial.

247
 Sample Point 1
100 r------.----~---~
en
Q)
Cl
~ 80
Q)
u
t 60 ~~ 1 ,.- · '" ~ -
<0 . I
a.
'0 I
Q) 40 I
Cl
<0 I
C
~ 20 I.
Qi ~-----l,.JY1\~~
a.
o'--_.....t:::lo.............._ ........--.;.. ~

585 585.5 586 586.5

time

Sample Point 2 Sample Point 3

100 100
en I en .I
~
Q) Q)
Cl ~ r""l\ Cl
c c
....
<0
Q)
80
lr'I <0
....
Q)
80

u u
t 60 r ~ ~~ t.. 0.: ..;.... ," .~ .... . t 60 . . .. ,.... ~ , J" L"..:.. "... ~ _ ... . ~
<0 I <0 . I ·
a. a.
I I I
'0 '0
Q) 40 I I Q) 40
Cl
<0 I
I
, Cl
<0
C I I'
C
Q) I Q)
o 20
o 20
.... I ·A." l ....
Q) "- I. Q)
a. I- '",-, i :
-vy'
~

- a. ~ - "-

o 0
585 585.5 586 586.5
585 585.5 586 586.5
time time

Figure A.I5: Step flow trial 3, 7 August 1996: particle counter size distribu-
tions; time is in days since 1 January 1995, flow (dashed line) is multipli d by
100, the elapse of one travel time is indicated by a ueriical dashed-dott d lin .
both are included fo r refer ence. Inlet particle counter size distribution data i not
available f or this st ep flow trial.

:..4
 Sample Point 1
100
C/)
Q)
OJ
c
Cll 80
....
Q)
C3
t 60
Cll
a.
'0
Q) 40
OJ
Cll
C
Q)
o.... 20
Q)
a. I
-- - - - \. - -- - -
0
600 600.5 601 601.5 602
time

Sample Point 2 Sample Point 3

100 100
C/) C/)
Q) Q)
OJ OJ
c 80 c 80
Cll
....
....
Cll
Q) Q)
13 I
13
.. ... ' . .. t 60
t 60 " I' , I'
Cll

-
Cll
a. I a. I
0 I, '0 I
Q)
OJ
Cll
40
I
" . . .. . ' ,'
Q)
OJ
Cll
40
I:' I
C I: I
C
Q) Q)
o 20 ~
20
....
Q) Q)
a. a.
0 0
600 600 .5 601 601 .5 602 600 600.5 601 601,5 602
time time

Figure A.16: St ep flow trial 4, 22 August 1996: particle counter siz di lrilni-
tions ; time is in days since 1 January 1995, flow (dashed line) is multipli d by
1DO, the elapse of one travel tim e is in dicated by a vertical dash d- doti d lin .
both are included f or refer ence. Inlet particle counter size distribution data i not.
available f or this st ep flow trial.

249
 Sample Point 1
100
en
Q)
0>
c: 80
...ctl
Q)
C3
t 60
ctl
a.
'0
Q) 40
0>
ctl
c:
Q)
o... 20
Q)
a.
0
606 606.5 607 607.5 608
time

Sample Point 2 Sample Point 3

100 ,------.~----.---~--___, 100
en
Q)
en
Q)
0> 0>
~ 80 c 80
1 .".........u..-~.V" ~
Q) Q)
C3 1 C3 I
... ...1. .1 . 1 I:
t 60 I" " '" . t 60
ctl
a. I i ctl
a.
1
I .
I
1 I
'0 1 I '0
1 I:
Q) 40 ..... 1· ... I. . . . . . . . . . .. '.' . . . . Q) 40 1 1
0> I 1 1 0> I: I
~ I ctl I
I c: I: I
ai~ 20 1 ~~-....""---- 1
I .'"""" ..... ~ Q)
...o 20 . I
Q) I_ ~ : _ Q)
1 ______ - - - -
a. a.
0
607 608 606 606.5 607 607.5 608
time time

Figure A.17: St ep fl ow tria lS, 28 Augu st 1996: particle counter siz distribu-
tions; time is in days since 1 Jan uary 1995, flow (dashed line) i muliipli d by
100, the elapse of one travel time is in dicated by a vertical dash d-dott d lin ,
both are included f or ref erence. Inlet particle counter data was not y t coni pui r
logged (hand-logged data is shown) . Inlet particle counter iz di iribuiion data
is not available f or this step fl ow trial.
 Inlet Shed Sample Point 1
100 ,-----,---~----.----~
l/) l/)
Q) Q)
OJ OJ
~ 80 c:
~
Q) 1 Q)
U 1 U
t 60 . . . . . . .1. . . .
ell 1 I
t 60
ell
a. a.
o 1 I
's
Q) 40 ····· 1··· ·:· · ·· 1· · ·· · · · · · · Q) 40
g> I I OJ
ell
C C
~ 20 1 ~ 20
s ~~~~~~~\iI~~~~ Q)
a.

647.5 648 648 .5 649 647 .5 648 648 .5 649

time time

Sample Po int 2 Sample Point 3

100 100 NY"" T
l/)
Q)
l/) !F,f :....",..,....
Q) ~. I~ L .. .. - ... .
OJ OJ
c:
ell
....
80 ~ 80 I
Q) Q) I I
U I U I I
t 60 t 60 ..
ell I ell 1 1
a. a.
0 1 i I
o I I
40 .. , . j . .:. .. . I · · · · · .. · 40 I 1
Q)
OJ
Q)
OJ
I
ell 1 I ell I I
I
C I
C 1 I
~
Q)
o 20 20 I
.... I " I
Q)
a.
Q)
a. .... - -L 1
.l . :
-
... _- -- - - --- - - 1-
' j
- -
0 o
647 647.5 648 648.5 649 647 647.5 648 648 .5 64 9
time time

Figure A.18: St ep flow tria l 6, 8 October 1996: particle counter siz di iiribu-
tions ; time is in days since 1 January 1995, flow (dashed line) i multipli d by
100, the elapse of one travel tim e is in dicated by a vertical da sh d-dott d lin .
both are included f or reference.
 Inlet Shed Sample Point 1
100 100
en en
Q) Q)
Cl Cl
c:: 80 c::
«l «l 80
"- "-
Q) Q)
u U
t 60 .I· ... . ...
t«l 60
«l
a. I a.
'0 I '0
Q) 40 . ( . . . .. ' . . . ... . "
OJ 40
Cl Cl
«l «l
1:: I
1::
OJ
20 .. . . . . . . . . . . OJ
20
o
"-
o"-
OJ OJ
a. a.
0 0
654 654 .5 655 655.5 656 654 654.5 655 655.5 656
time time

Samp le Point 2 Sample Point 3

100 100
en en
OJ OJ
Cl Cl
c:: 80 c:: 80
«l
"-
T 1 «l
"-
OJ I 1 OJ
U o · .
t«l 60 .I· ..1. ..' .
t«l 60 .I : •. 1
·.
-a.
0
Q)
Cl
40
I
. J. ...
I
1
1
I
... . . . ' .' . ...
a.
'0
OJ
Cl
40
I
I.
·.
. J:
·.
I
I
I
«l «l I
I
1:: 1::
OJ
o 20 . .~ . .". . . . I. . . .... ...... . . OJ
o 20
"- "-
I I OJ
OJ
a. . -- -- - - - --- a.
0 0
654 654 .5 655 655.5 656 654 654.5 655 655.5 656
time time

Figure A.19: St ep flow trial 7, 15 October 1996: particle counter tz di itri-

butions; time is in days since 1 January 1995, flow (dashed lin ) i multipli d
by 100, the elapse of one travel tim e is indicated by a vertical dashed-dolt d lin
both are included for reference.

252
 Inlet Shed Sample Point 1
100 100
(/) (/)
Q) Q)
Cl Cl
c 80 c 80
...ro
Q)
...ro
Q)
I
U U
.. . . .. .. ... ..
tro 60 I ··. tro 60
a. I a.
'0 .
. . . ". '
I . . . .. . '0
...
. . . .. ... . .. . . . .. . . .. ..
Q) 40 '. ,
Q) 40
I
Cl
ro , .. __ ................ ...... '-.' - - -- - Cl
ro _. ~-.- .. _...... ""'- - - --
"E "E
Q) . . . .. . . . Q)
...o
Q)
20 ...
o
Q)
20 ..

a. a.
0 0
663 663 .5 664 664 .5 665 663 663.5 664 664.5 665
time time

Sample Point 2 Sample Point 3

100 100
(/) (/)
Q) Q)
Cl Cl
c 80 c 80
...ro
Q)
~
Q)
u U
tro 60 ., ... . . t 60
ro
a. a.
'0 '0
40 ... '. 40
-
Q) Q)

,. -_ ...:.. . ... _. _......

-...
Cl
ro
c
Q)
o 20
-~ - - _... Cl
ro
"E
Q)
...o
Q)
20
J • • _ . , _ . _. __ ....... '- - - _... -

Q)
a. a.
0 0
663 663.5 664 664.5 665 663 663.5 664 664.5 665
time time

Figure A.20: St ep flow trial 8, 24 October 1996: particle counter siz disiri-
butions; time is in days since 1 January 1995, flow (dashed line) i multipli d
by 100, the elapse of one travel tim e is in dicated by a vertical dash d-dott d lin
both are included for reference.

253
 Inlet Shed Sample Point 1
100 r - - - - - . - - - - - - - - . - - - - - - - - , 100 r-----;---.---------~
l/l l/l
Q) Q)
Cl Cl
~ 80 ~ 80
Q) Q)
13 I 13 I .
.€ 60 '1 t 60 ' 1 1'
Cll Cll
0. 0.
I I I I
'0 . . . 1 . . . .. : . . '0

I', I. . __·
Q) 40 · 1 ·· · · ·· : · ·· ···· .. · ... Q) 40 . 1 1· · .
Cl
Cll -~ ..-.) : Cl
Cll -~ • • - .) I
---~ ---~_·· ·
C C
~ 20 ~ 20
Q) Q)
0. 0.

669 .5 670 670 .5 669.5 670 670.5

time time

Sample Point 2 Samp le Point 3

100 100
l/l
Q)
Cl
l/l
Q)
Cl
•
1

-or. I ..'...... . . ,
c 80 j : 1 c 80 I
...
Cll
Q)
~
Q)
I I
13 I I . I 13 I I
t 60 . ... ... I .. ! : ' 1. . ., . t 60 . . . .I . . .
I
Cll Cll
0. 0.
I I : I I I
'0 '0 . 1. . I
Q) 40 ....... 1. ·1 .. : · · · · · · 1 ····· Q) 40
~
Cl
- ~ •. -J I' :. II . --.. _.... - .. ..
.......---- Cl
Cll - ~-. -..,J' I
C
I .
C
Q)
Q)
20
...o
Q)
20 ~
Q)
0. 0.
o L-..:.:.....J~:±::::::l::::b:==io._ ___l
669.5 670 670 .5 669 669.5 670 670 .5
time time

Figure A.21: Step flow trial 9, 30 October 1996: particle count r iz di iri-
butions; time is in days since 1 January 1995, flow (dashed line) i muliipli d
by 100, the elapse of one travel time is indicated by a vertical dash d-dott d lin
both are included for reference.

254
 Inlet Shed Sample Point 1
100 100 r----~--~--~--~
en
Q) J. II 1 en
Q)
OJ .... .'1..... r: OJ
~ Fr' l , ·T·
Q)
80 ...~
Q)
80

U o
t 60 t 60
<0 <0
a. a.
'0 '0
Q) 40 '. '
Q) 40
OJ
<0
\ , _. J ..J .
I .
11
:
J•••• • • , .. .... _ ... _
:
. 1_.'............
: I
OJ
<0 •..:. ~ ..J •• J • • •••• • • __ " _ . 1_ .,6. ,
C C
~ 20 ~ 20 .. ...... :- 1.. ·
',
· ~ -:..J:l· ,J ' _.i ' j'
Q) Q)
a. I"'l'" u, ' ''1\'" a.
""-
o
704 704 .5 705 705 .5 706 704.5 705 705.5 706
time time

Sample Point 2 Sample Point 3

100 100 ,-----.---~----.-~--~--~
en en
Q)
Q)
OJ OJ
c 80 .. I ~ 80 i
...
<0
I ...
Q)
I
Q)
U i U i
t 60 . . . . . .'. . . .I t 60 i
<0 <0
a. i a. i
'0 '0 i
Q) 40 . ':' . . .fi
. : . . . . . . . ..: .. . .. Q) 40 .: I
OJ OJ I, _. J ..J... , J""
<0 I•..:. J ..J .~ J _._ • • _ .1 _ ~' 6." ..•. <0
I .
• •• . . . .... _ • • _ .. 1__ ' .. . ... ... 4 .

I
C I : I ' . I C
Q) ..... .. ..". .. ·1 .... .... . I ' ~ 20
...
o
Q)
20 I : . Q)
I I

a. a.
0 1;.""",;".:;.::;,...J....::::!:~_ ......_ _..........._ .....
0 706
704 704 .5 705 705.5 706 704 704.5 705 705.5
time time

Figure A.22 : St ep flow trial 10, 4-5 December 1996: particle count r siz dis-
tributions; time is in days sin ce 1 January 1995 flow (dashed lin ) i mul iipli d
by 1 DO, the elapse of one travel tim e is indicated by a vertical da li d-doii d lin .
both are included for reference.
 Inlet Shed Sample Point 1
100 100
(/) (/)
Q) Q)
Ol Ol
c 80 C
....
ltl
...ltl 80
Q) Q)
u .. . I. . .... _ .... : • • • • _. .. I~ • • ..L.J,_ •• • ~ . ...... . j .. . . U I
'e 60 t 60 " ~ • •• • • •• • ••• • • __.. . . . . .J.." ... ~ .•......l
ltl I ltl II I
0. 0.
I II
'0 . . . . . .. . . . ... . . . . . .. . : ... .. '1' '0 I I
Q) 40 Q) 40 . . . . . Ii " I
Ol Ol
ltl I ltl II I
C I C I
Q)
20 Q) I
...
U
Q)
...o
Q)
20
0. 0.

0 0
710 711 712 713 710 71 1 712 71 3
time time

Sample Point 2 Sample Point 3

100 100
(/) (/)
Q) Q)
Ol Ol
c 80 c 80
...ltl ...
ltl
Q) I
Q)
U I U I
t 60 -t.'. _...:.......... 1•• • ..1.- 1.... 'l • • ~ . . . . . . . . !.. . . t 60 I • • ", . . . . . . . . . . . . . .. ... .J..••• • • • • ~ • ••••• • .l
ltl ltl I I
0. 0.
I I
'0 .. ... . .. . . .. . .. . .
'0 I
Q) 40 Q) 40 I I
Ol Ol
ltl I I
ltl
C C
Q)
I
I
Q) I
o 20
.... ~
20
Q) Q)
0. 0.

0 0
710 711 712 713 710 711 712 71 3
time time

Figure A.23: St ep flow trial 11, 10-1 2 December 1996: par-tiel count r siz
distribution s; time is in days sinc e 1 January 1995, flow (dashed lin ) i multipli d
by 100, the elapse of one travel tim e is indicated by a vertical da h d-tlott d lin ~ .
both are in cluded f or ref erence.
 Inlet Shed Sample Point 1
100 ,------~----,...-----~
en
Q)
en
Q)
OJ OJ

...~
Q)
80 ~ 80 1
1
~
u CJ 1
't 60 . . 't 60 I
ell ell I
C.
'0 t--_ _ ~ _ _ IJ_ ••••_.:_ _ _ •• _ .. 1• o
C.
,i -···--··_···_····_u _.••._._.... .. _ ••_'.
I
~ 40 : .. .. .. .. .. .. .. . .. " 1 .. ~ 40 .. ,'1 " .. , 1
ell 1 ell 1 1
C C 1
Q) ~ .. . . . . . . . . . . : : I . Q) 20 ..... 1 . . . . . . . . . . . . . I
~ ~
Q) Q)
c. c.

718 719 720 718 719 720

time time

Sample Point 2 Sample Point 3

100 r----,---r-----.-------,----~ 100 ......
en en ;..... . ...
r "
T
r~
Q) Q) l'
OJ OJ
c
...~
Q)
80 ~
Q)
80 1
1
U U 1
t ..
t 60 60 . ·1
ell ell 1
C.
o
C.
t--, _ --
I
~
:
_.J_ __..:_ _ I' 0 ..
,--'1--"-" . r • •• • _ u _...._. _ ..... . _ .• - '.
Q) 40 . .... · 1·· ·· · ·· · ·· · · · ·· · ··· ··· · " 1 " Q) 40 '1 • 1
OJ
ell I I : : I
C
OJ
ell
1
I
I
1
1
C I I .
20 I' I Q) .1
Q)
~
. 1 : -
...
CJ 20
Q)
c. I
Q)
c. - I~
r
OL....l.o................- .........-..:.;.....;...--.:.......IIL..J o 719 720
717 718 719 720 717 718
time time

Figure A.24: Step flow trial 12, 17-1 9 December 1996: pariicl counter siz
distributions; tim e is in days since 1 January 1995, flow (dashed lin ) i inuli ipli d
by 1 DO, the elapse of one travel tim e is indicated by a dott d line, both ar incliul d
f or reference.
Appendix B

Calculations for the

Determination of k1(f), k2(f)

B.l Introduction

In section 6.5.3 an outline of the calculations that were used to determine t ho

parameters k 1 (1) and k 2 (1) of the model of the response of particle counts to
a step increase in flow is given. To maintain the flow of thought int crnu-diato

equations were omitted there, but are provided in this appendix.

Let us recall the particle counts model eq. (6.9) on page 149:

p(t) = v 1:0 b'(,)) (k,(I)e-al!-<J) + k,(I)) l(t - I)) til!. (B.l)

with

I if t > 0
1(t) ={ (8.2)
o otherwise

From this it follows that for t > T (1(t) not necessary] the parumotor- ('J (T)
and C2 (T) can be calculated,

c1(T) V k1 (f) 1:0 b* (19)eQd dJ) (B.3)

c2(T) vk 2 (f ) 1:0 b*(19) d19. (B.-t )

which can be found independent of these equations directly from the particle
counter data.

In addition, the inverse 8- 1(s) of the Laplace transform S(s) of the shear-off
function s(t) is used. The Laplace back-transform of S-1(S) is denoted by "'1P11.(t).
It is

b*('l9) = -
Iii)
V 0
p(t)Sinv('l9-t)dt. (B.5)

with

Sinv(t) = L -1 { 1 }
S(s)' (B.6)

B.2 Calculation of the Laplace Back-Transform

The aim is to find

With £'{j(t)} = F(s), the well known theorem [18,82] can be written as

259
with

I(V) _ li d" f(t)

Jo - Im-~
t---++o dt v

Thus (since the Laplace transform is linear)

£-1 {sF(s)} = £-1 {sF(s) - fo} + £-1 {fo}

= J'(t) + f 06(t) (B.IO)
£-1 {S2 F(s)} = £-1 {S2 F(s) - sfo - f~} + £-1 {sfo} + £-1 {f~}
= f"(t) + f06'(t) + f~6(t). (B.ll)

where prime denotes differentiation, so that j'(t) is the first and f"(t) the second
derivative of f(t) with respect to t: In eqs. (B.IO) and (B.II) Dirac's delta impulse
is used [70, 79]. The validity of the use of the Dirac impulse in this context can 1)('
seen from pp. 92/93 in [70] or p. 202 in [79]. However. it can also be calculated.

Consider the definition of the Laplace transform to find:

= Se-sa , (B.12)

where integration by parts and the fact that 6(t - a) = 0 for t = 0 and c:: = 0
for t ---t 00 is used. Hoskins [70] remarks at this point "... that strictly this last

step does involve a tacit appeal to the pointwise behaviour of the delta function.

However, all that we are really saying is that the operational property of I' is
confined absolutely to the point t = O. At any other points it has no pffpct. and
this is what is really meant by saying that it 'evaluates to zero' at such point s"

(p.53/54). Then by formally setting a = 0, the following is obtained:

£{8'(t)} = s.

260
Note that the respective integral is not defined for a = O~ but this result i~
consistend within the framework of the Laplace transform and therefore generally
adopted (see p. 202 in [79] or p. 82/83 in [70]).
With the help of on eqs. (B.IO) and (B.II) eq. (B.t) can be solved:

(B.1-1)

Therefore, the inverse shear-off function is obtained to be

as given in eq. (6.29) on page 161. This result was also double-checked with

Mathematica [158].

B.3 Calculation of the Full Integral

Now eqs. (B.3) and (BA) will be solved.

However, first the abbreviation is introduced

(13.lt;)

261
and calculate the following integral:

(B.17)

Note that the minus signs in front of p(t) and p'(t) within the integration by
parts originate from the argument of the delta impulse, which is negative in t.
The above result is only valid for p(O) = 0 and the interpretation of the limit at

t = {) follows the argumentation of eqs. (B.12) and (B.13).

Now, for i = 1,2, eqs. (B.3) and (B.4) can be written as

Eq. (B.18) is the same as eq. (6.31) on page 162.

262
B.4 A Test of Validity

A simple test of the validity of eq. (B.18) can be performed by calculating a

known case. This obviously does not prove that the equation is valid. however. it
gives some evidence to that effect. Take a step response of 5(s). i.e. a flat biofilm
profile with magnitude 1 is the input of the system 5(s). In that case, eq. (B.5)
leads to:

11{) p(t)Sinv('19 - t) dt =
- 1. (B.19)
v 0

Furthermore, the response of the system for t < T is known to be

1
p(t) = vk (f ) (1 - e- at ) l(t) + vk 2 (f )t l(t). t <T (B.20)
a
p'(t) =vk 1 (f )e- at l(t) + vk 2 (f ) l(t), t <T (B.21)

Note, that the condition p(O) = 0 is fulfilled in this case. Thus, for t < T (i.e.
'19 < T) 1

1 Note that eq. (B.I8) is only concerned with 0 ~ {) ~ T

263
Using integration by parts, the following is obtained

(B. 2:3)

With this result:

= 1, (B. 2-t)

as expected.

2G-t
Appendix C

Implementation in Matlab

e.l Overview

The Water Quality Toolbox written during the work for this PhD contains over
100 Matlab functions and scripts. Its current version is 2.3 and it is written for
Matlab 5. The purpose of this toolbox is to facilitate the analyses necessary for
modelling water quality in the "TORUS" pipe rig of Thames Water. It contains
function for reading data into Matlab, has its own time format, contains several
functions for visualisation and graphical input, includes the preprocessing (as
discussed in section 5.5), is capable of finding the parameters of the particle
counts model and the biofilm shedding profile (see section 6.10) and does the
calculations of various chlorine decay coefficients (cf. section 7.8).
A discussion of the functionality of the relevant functions is given at the end
of each chapter in part III (as indicated in the previous paragraph). In addition
a full on-line help is available for each function, a list of all functions is available
through help wqt. A list of required data files is provided through wqt_data, a
brief on-line introduction is given through wqt_info.
The following section explains briefly the most important functions and script.
It is not a full tutorial of the Water Quality Toolbox. For more information S('('

the on-line help.

265
C.2 Using the Water Quality Toolbox

C.2.1 Reading Data and Time Formats

An executable file, datapre (source in C++). does some initial conversions Oil the
ASCII output file of the pipe rig computer (extension wIf). The converted filt'~
have the extension swlf and are the input of the load function of the Water Qual-
ity Toolbox, wload. This function can load pipe rig data between January 1995
and July 1997. If new data is to be analysed, only wload of all the functions in
the Water Quality Toolbox needs to be modified.
Since the work on this Toolbox started in 1996, this function was originally
written for Matlab 4 and no cell arrays were available. This implies that then' is
a need for four data matrices, one for each sample point (Inlet, SP 1. SP 2. SP :3),
because the sizes do not always agree (mainly due to errors or meter failures in
single sample points). These variables are declared global and are called yrawO
(Inlet), yrawl (SP 1), yraw2 (SP 2) and yraw3 (SP 3). In addition there is yraw4
which contains the inlet flow readings.
In addition to the logged pipe rig data several bench analyses were done.
These results are provided in files with the extension dat (see the file readme or
wqt_data in Matlab for a complete list of swIf and dat files). These data files

are loaded through wxIoad.

The Water Quality Toolbox used two time formats of its own. The 'internal'
format is days in the year since 1 Januarv 1995 (used in swIf-files and all vi-
sualisation), the 'external' format is (generally) of the form yyddmm.hhmm wi t h
yy=year, dd=day, mm=month, hh=hour. mm=minute. The year is optional, tho
default is 1996. This format is used in all dat-files. The function wdate converts

into and from these time formats.

C.2.2 Processing of Raw Data, Visualisation and Graph-

ical Input
A lot of the data processing is done in wcreate. This function splits t he raw

266
data matrices yrawO ... yraw4 into columns, depending on which variable i~ of
interest. The columns are referred to by numbers. e.g. 3=total particle counts.
23=total chlorine, 24=temperature. A full list of column numbers and their
meaning is obtained with wcreate ('?'). Use any column>2-1 with care sinr«
calculations may take very long. It is generally better to calculate the chlorine
decay coefficients with wdecay.
It is also possible to generate a matrix of indices in data vectors created by
wcreate or in the raw data. These matrices are referred to as minmax-rnatrices
and are generated by wminmax.
Visualisation of data is frequently done with the script lookat. This script
uses wcreate internally. This script can plot swlf-data and external data (dat-
files), it can input via a crosshair, plot in subplots, do axes adjustments and
change the x-axis labels, among other things. The plot functions wplot or wlog
are used.
Other functions that take graphical input (generally a crosshair) are finddate,
findindex, findttime, finddistance and findstepheight.
The size distributions are no column in the raw data matrices, therefore thev
have to be calculated separately in the function sizedistr and are not included
in wcreate or lookat.

C.2.3 Data Analysis

The functions and scripts used for the analysis of data as discussed in part III
of this thesis are already given in the appropriate chapters (as indicated above).
Thus, the implementation of the preprocessing as presented in chapter 5 is given in
section 5.5 on page 120, the Matlab functions for the particle counter models and
their implications as shown in chapter 6 are provided in section 6.10 on page 183
and the part of the Water Quality Toolbox that is concerned with modelling
monochloramine decay (chapter 7) is discussed in section 7.8 on page 207.
These sections together with rest of the respect in' chapters and the on-lino
help should give sufficient information for the usage of t hesc functions and «-ript-.

-)6-
- I
C.2.4 Starting a Session

Typically a session of using the Water Quality Toolbox would be started by

invoking wload (obviously after starting Matlab), e.g. wload( [19 20 22]) to
load July, August and October 1996. Subsequently preprocessing is necessary,
e.g. it is necessary to run preproc (no input arguments). This will set the global
variable preprocessing to 1 if run successfully, or to 0 otherwise.
After this any of the above mentioned functions or script can be used, e.g.
type lookat and give as 'dmin' 572, as 'dmax' 573 and as 'col' 3 to see a plot
of the total particle counts of SFT 1. Type? at the prompt 'Do you wish to
see another plot?' to obtain a list of capabilities of lookat. Alternatively.
use wdecay or calcdecay for chlorine decay coefficients or bioprof i for the
particle counts model and the biofilm shedding profile. See the scripts calcmax
or bioscript for examples of how to use calcdecay or bioprofi.

C.3 Summary
This appendix gave a brief overview and introduction into the Water Quality
Toolbox written for Matlab 5. A full on-line help complements the above given
deliberations. These functions and scripts form a powerful tool for the analysis of
"TORUS" pipe rig data and possibly other water quality data. In addition they
can be used to reproduce all the data given in this thesis and most of the graphs.

268
Appendix D

Reynolds N umbers

This appendix gives in figure D.l the Reynolds numbers versus the volumetric
flow rate in litres/second for the "TORl'S" pipe rig. The Reynolds number Ht
is calculated as

Re = vd = f . (0.1)
V 2507rdv

where v=velocity of water in ta]«. d=pipe diameter in m, v=kilwmat i(' visrosi t~·
of water in m 2 Is and f =volumetric flow rate of water in lis. The pipe diamorr-r
is in the case of this work assumed to be uniformely 0.11 m.
The kinematic viscosity changes with temperature (se{' p.33i in [50]). Grn-
erally it is assumed that for Reynolds numbers below 2000 the flow is laminar
and for Reynolds numbers larger than 4000 it is turbulent (see p.142 in [102]),
al though these numbers may increase drastically for vorv smoot h pipes.

269
 12000

10000

,
~ 8000
.0 ,
8
::s
e ,
,
-
~
o
s:::
c-,
6000

". .'
, .
, .'
,

Q.)
, ., '
. :

cG 4000 , ., .. '
"

, , .,.. . :
.'

, .
:

2000 , ....
,:..
. , :~ . ,

,
.

, :.'
i(." •

0.2 0.4 0.6 0.8

volumetric flo w rate [lIs]

Figure D.l: R eyn olds numbers ve rsus uolume tric fl ow f or th iiTOR ). pip ,
ri g (diam et er 11 Omm). Chang e in R ey n olds number due to chan g in vi cosi ui
with temperature is indicated.