MEASUREMENT AND INSTRUMENTATION

BMCC 3743
LECTURE 4: EXPERIMENTAL
UNCERTAINTY ANALYSIS
Mochamad Safarudin
Faculty of Mechanical Engineering, UTeM
2010
andims - FKM -
UTeM
2
Contents
Propagation of uncertainties
Consideration of systematic and random
components of uncertainty
Sources of elemental error
Uncertainty of the final result
Design-stage uncertainty analysis
Applying uncertainty-analysis in digital data
acquisition system
andims - FKM -
UTeM
3
Propagation of uncertainties
Uncertainty analysis is important to identify
corrective measures while validating and
performing experiments.
Propagation of uncertainties => total
uncertainties, e.g. P = VI = n

Two important factors in uncertainty:
Random uncertainty (or precision uncertainty) :
imprecision in measurements
Systematic uncertainty (or bias uncertainty):
estimated maximum fixed error
P
w
andims - FKM -
UTeM
4
General consideration
If R is a function of n measured variables x
1
,
x
2
, . x
n
, i.e.


Then a small change in is due to small
changes in s in x
i
s via the differential
equations:


( )
n
x x x f R , ,
2 1
=
R o
'
i
x o

=
=
n
i
i
i
x
R
x R
1
o
o
o o
Sensitivity
coefficient
(1)
(2)
andims - FKM -
UTeM
5
General consideration
For calculated result based on
measured x
i
s, Eq. (2) can be rewritten
as

where | | is to make sure we dont get
zero uncertainty in R.
However, this can produce high
estimate for w
R
.

(3)

=
=
n
i
i
x R
x
R
w w
i
1
o
o
Uncertainty
in result
Uncertainty
in variables
andims - FKM -
UTeM
6
General consideration
Hence Eq. (3) is better represented by


=>root of the sum of the squares (RSS)
In this case, the confidence level must be the
same for all uncertainties (typically 95%).
Assumption is made that each measured
variables (hence, error) are independent of
each other.
(4)
2 / 1
1
2
|
|
.
|

\
|
(

=

=
n
i
i
x R
x
R
w w
i
o
o
andims - FKM -
UTeM
7
Exercise
To calculate the power consumption of
an electric circuit, we have P = VI where
V = 100 2 V and I = 10 0.2 A
Calculate the maximum possible error
(uncertainty) and best-estimate
uncertainty (RSS). Hint: Use Eq. (3) and
Eq. (4) respectively.

andims - FKM -
UTeM
8
Answer to Exercise
W x x
i
P
V
P
w
i V p
40 2 . 0 100 2 10
max
= + =
c
c
+
c
c
= e e
Because P=VI
dP/dV=I=10.0 A , dP/di=V=100.V then
( ) ( ) ( ) W x x
i
P
V
P
w
i V p
3 . 28 2 . 0 100 2 10
2 / 1
2 2
2 / 1
2 2
= + =
|
|
.
|

\
|
|
.
|

\
|
c
c
+
|
.
|

\
|
c
c
= e e
andims - FKM -
UTeM
9
Contents
Propagation of uncertainties
Consideration of systematic and
random components of uncertainty
Sources of elemental error
Uncertainty of the final result
Design-stage uncertainty analysis
Applying uncertainty-analysis in digital data
acquisition system
andims - FKM -
UTeM
10
Consideration of systematic and
random components of uncertainty
Random uncertainty depends on
sample size (usually large, n>30)
Systematic uncertainty is independent
of sample size & does not vary during
repeated reading
Need to separate for detailed
uncertainty analysis
andims - FKM -
UTeM
11
Random uncertainty
Using t-distribution, the random uncertainty
for all measurements is given by


where S
x
is the standard deviation of the
sample

For a single measurement (also for each
individual measurement), the random
uncertainty is
M
S
t P
x
x
=
(5)
x x
tS P =
(6)
andims - FKM -
UTeM
12
Systematic uncertainty
Sometimes assumed as level of
accuracy
Depends on manufacturers
specification, calibration tests,
mathematical modelling, considerable
judgement as well as comparisons
between independent measurements.
andims - FKM -
UTeM
13
Systematic uncertainty some
examples
Radiation heat transfer => lower
measured value
Instrument location => spatial error, e.g.
a single thermometer measures
temperature in a box oven
Dynamic errors
andims - FKM -
UTeM
14
Combining random & systematic
uncertainties
Total uncertainty is obtained, using RSS
(Eq. 4) for all measurements, is given by


For a single measurement of x,

( )
2 / 1
2 2
x x x
P B W + =
(7)
( )
2 / 1
2 2
x x x
P B W + =
(8)
andims - FKM -
UTeM
15
Contents
Propagation of uncertainties
Consideration of systematic and random
components of uncertainty
Sources of elemental error
Uncertainty of the final result
Design-stage uncertainty analysis
Applying uncertainty-analysis in digital data
acquisition system
andims - FKM -
UTeM
16
Sources of elemental error
Chain of uncertainties, e.g. A/D
converter would have quantisation
errors, sensitivity errors and linearity
errors. Each of these components
contribute to further errors.
Can be random or systematic error.
andims - FKM -
UTeM
17
Estimation of uncertainty
Systematic uncertainty: just combine all
elemental uncertainties
Random uncertainty: 3 approaches to
determine S
x

1. Run entire test in a sufficient number of times
2. Run auxiliary tests for each measured variable x.
3. Combine elemental random uncertainties
=> Based on experiment requirement.
andims - FKM -
UTeM
18
5 categories of elemental errors
Calibration Uncertainties: residual systematic errors
due to; uncertainty in standards, uncertainty in
calibration process, randomness in the process
Data-Acquisition Uncertainties: during measurement
due to; random variation of measurand, A/D
conversion uncertainties, uncertainties in recording
devices
Data-Reduction Uncertainties: due to interpolation,
curve fitting and differentiating data curves
Uncertainties Due to Methods: due to
assumptions/constant in calculation, spatial effects
and uncertainties due to hysterisis, instability, etc.
Other Uncertainties
andims - FKM -
UTeM
19
Combining elemental systematic &
random uncertainties (RSS)

=
=
=
=
m
i
i x
k
i
i x
S S
B B
1
2 2
1
2 2
Calibration
Uncertainties
Data-Acquisition
Uncertainties
Data-Reduction
Uncertainties
Uncertainties Due
to Methods
Other
Uncertainties
( )
2 / 1
2 2
y uncertaint
Variable
x x x
tS B w
x
+ =
Reproduced from Wheelers book:
ASME 1998
andims - FKM -
UTeM
20
Degrees of freedom, v
x
When sample size is large, v
x
is simply
number of sample, n, minus 1.
When sample size is small, then v
x
is given
by



=> Welch-Satterthwaite formulation (ASME
1998)
( )

=
=
=
m
i
i i
m
i
i
x
v S
S
v
1
4
2
1
2
/
(9)
Degrees of freedom
of individual
elemental error
andims - FKM -
UTeM
21
Contents
Propagation of uncertainties
Consideration of systematic and random
components of uncertainty
Sources of elemental error
Uncertainty of the final result
Design-stage uncertainty analysis
Applying uncertainty-analysis in digital data
acquisition system
andims - FKM -
UTeM
22
Uncertainty of the final result
(Multiple measurement)
Referring to Eq. 1, then for multiple
measurements, M, the mean results is given
by


Little exercise:
Derive the standard deviation (S
R
) and
random uncertainty ( ) of R.

=
=
M
j
j
R
M
R
1
1
(10)
R
P
andims - FKM -
UTeM
23
Uncertainty of the final result
(Multiple measurement)
Rearranging Eq. 4 (RSS), we get the
systematic uncertainty in terms of the
combination of elemental systematic
uncertainties, given by
2 / 1
1
2

|
|
.
|

\
|
=

=
n
i
i
i R
x
R
B B
o
o
(11)
andims - FKM -
UTeM
24
Uncertainty of the final result
(Multiple measurement)
Therefore, the total uncertainty estimate of the mean
value of R is


To estimate random uncertainty for multiple
measurements, results are more reliable using the
test results themselves, compared to auxiliary tests
or combination of elemental uncertainties.
Practical applications: The life of a light bulb, the life
span of a certain brand of tyre or car engine
( )
2 / 1
2 2
R
R
R
P B W + =
(12)
andims - FKM -
UTeM
25
Uncertainty of the final result
(Single measurement)
To deal with uncertainty of a single test
result only
Practical applications: measuring blood
pressure/ heartbeat, speed of car, etc
To estimate random uncertainty of the
result, must use or combine auxiliary
tests and elemental random
uncertainties.
andims - FKM -
UTeM
26
Uncertainty of the final result
(Single measurement)
Similar to Eq. 11, standard deviation of
the result is given by


Hence, the total uncertainty in the final
result is given by
2 / 1
1
2

|
|
.
|

\
|
=

=
n
i
i
i R
x
R
S S
o
o
( ) | |
2 / 1
2 2
R R R
tS B w + =
(13)
(14)
andims - FKM -
UTeM
27
Uncertainty of the final result
(Single measurement)
For a large n, then t is independent of v,
the degree of freedom, (and has a value
of 2.0 for a 95% confidence level).
For a small n, again using Welch-
Satterthwaite formulation, we get

( )

=
|
|
.
|

\
|
(

=
n
i
i
i i
R
R
S
x
R
v
S
v
1
2
2
2
1
o
o
(15)
andims - FKM -
UTeM
28
Example
The manufacturer of plastic pipes uses a scale with an
Accuracy of 1.5% of its range of 5 kg to measure the
Mass of each pipe the company produces in order to
Calculate the uncertainty in mass of the pipe. In one batch
Of 10 parts, the measurements are as follows:
1.93, 1.95, 1.96, 1.93, 1.95, 1.94, 1.96, 1.97, 1.92, 1.93 (kg)
Calculate
a. The mean mass of the sample
b. The standar deviation of the sample and the standar deviation
of the mean
c. The total uncertainty of the mass of a single product at
a 95% confidence level
d. The total uncertainty of the average mass of the product at a 95%
confidence level
andims - FKM -
UTeM
29
Solution:
10
93 . 1 92 . 1 97 . 1 96 . 1 94 . 1 95 . 1 93 . 1 96 . 1 95 . 1 93 . 1 + + + + + + + + +
= =

n
m
m
i
av
kg 94 . 1 =
(a)
(b)
2 / 1
2
]
1
) (
[

=

n
m m
S
av i
samp
( ) ( ) ( )
kg
n
S
S
kg
Samp
mean
2
2
2
2 / 1
2 2 2
10 52 . 0
10
10 * 70 . 1

10 * 65 . 1
]
9
.......... 02 . 0 01 . 0 01 . 0
[

= = =
=
+ + +
=


) Confidence (95% k 10 * 73 . 3
2
g tS P
samp Single

= =
) % 95 ( 084 . 0 ) ( : y uncertaint Total
075 . 0 5 015 . 0 015 . 0 B
6.6 Table t - Student from freedom of degrees 9 1 - n for is 2.262 t
2 / 1 2
sin
2
sin single
single
confidence B P w
kg kg Range
gle gle
= + =
= = =
= = = u
(c)
andims - FKM -
UTeM
30
) % 95 ( 076 . 0 ) ( : y uncertaint Total
075 . 0 B
) Confidence (95% 10 18 . 1
2 / 1 2 2
mean
single mean
2
mean
confidence B P w
kg B
kg tS P
mean mean
mean
= + =
= =
= =

(d)

As can be seen, the dominant factor is the systematic uncertainty.
andims - FKM -
UTeM
31
Contents
Propagation of uncertainties
Consideration of systematic and random
components of uncertainty
Sources of elemental error
Uncertainty of the final result
Design-stage uncertainty analysis
Applying uncertainty-analysis in digital data
acquisition system
andims - FKM -
UTeM
32
Design-stage uncertainty analysis
(Based on ASME 1998)
1. Define the measurement process
State test objectives, identify independent parameters and
their nominal values, etc
2. List all elemental error sources
To do a complete list of possible error sources for each
measured parameter.
3. Estimate the elemental errors
Estimate the systematic uncertainties and standard
deviations. If error is random in nature and/or data is
available to estimate the std dev. of a parameter, then
classify it as random uncertainties, which must have the
same confidence level. For small samples, to determine
degrees of freedom. Refer Table 1.
andims - FKM -
UTeM
33
Guideline to assign elemental
error (Table 1), from Wheeler
ERROR ERROR TYPE
Accuracy
Common-mode volt
Hysterisis
Installation
Linearity
Loading
Noise
Repeatability
Resolution/scale/quantisation
Spatial variation
Thermal stability (gain, zero, etc.)
Systematic
Systematic
Systematic
Systematic
Systematic
Systematic
Random*
Random*
Random*
Systematic
Random*
*
assume no. of samples > 30

andims - FKM -
UTeM
34
Design-stage uncertainty analysis
(Based on ASME 1998)
4. Calculate the systematic and random uncertainty
for each measured variable
Use the RSS formulation with data & procedure in Step 3.
5. Propagate the systematic uncertainties and
standard deviations all the way to the result(s)
Use the RSS formulation to find the final test results, with
the same confidence level in all calculations.
6. Calculate the total uncertainty of the results
Use the RSS formulation to find the total uncertainty of the
result(s).
andims - FKM -
UTeM
35
Contents
Propagation of uncertainties
Consideration of systematic and random
components of uncertainty
Sources of elemental error
Uncertainty of the final result
Design-stage uncertainty analysis
Applying uncertainty-analysis in
digital data acquisition system
andims - FKM -
UTeM
36
Applying uncertainty-analysis in
digital data acquisition system
A digital DAS typically consists of sensor,
sensor signal conditioner, amplifier, filter,
multiplexer, A/D converter, Data reduction
and analysis
Problem may occur due to sequential
components which may have different range
from adjacent components.
So, adjustment to uncertainty data must be
done.
andims - FKM -
UTeM
37
Another example
In using a temperature probe, the following uncertainties
were determined:
Hysteresis 0.1
0
C
Linearization error 0.2% of the reading
Repeatability 0.2
0
C
Resolution error 0.05
0
C
Zero offset 0.1
0
C

Determine the type of these error (random or systematic) and
the total uncertainty due to these effects for a temperature
reading of 120
0
C
andims - FKM -
UTeM
38
C
B
28 . 0
) 1 . 0 )] 120 )( 002 [(. 1 . 0 (
2 / 1 2 2 2
=
+ + =
C P 21 . 0 ) 2 . 05 (.
2 / 1 2 2
= + =
C w
w
tS B p B w
35 . 0
] ) 21 . 0 ( ) 28 . 0 [(
] ) ( [ ] [
2 / 1 2 2
2 / 1 2 2 2 / 1 2 2
=
+ =
+ = + =
Assuming that the random errors have been determined with samples>30,

So total uncertainty
hysteresis
+ 01 . C


systematic
Lineariz.error
0 2% . of reading


systematic
Resolution error 0.05C random
zero off set 0.1C systematic
repeatability
+0 2 . C
random
andims - FKM -
UTeM
39
Two resistors, R
1
=100.0 0.2 and R
2
=60.0 0.1
are connected (a) in series and (b) in parallel.
Calculate the uncertainty in the resistance of the resultants
circuits. What is the maximum possible error in each case?
andims - FKM -
UTeM
40
R R R
R
R
R
R
w
R
R
R
w
R
R
R
w
R
w
R
R
R
w
R
R
R
w
R
= + = =
= +
= + =
= +
= + =
1 2
1
1
2
1
1
1
2
1
2
2 1 2
0 2 1
2
01 1
2 1 2
0 22
1
1
1
2
0 2 01 0 3
c
c
c
c
c
c
c
c
c
c
c
c
,
[( ) ( ) ]
/
[( . * ) ( . * ) ]
/
.
,max
. . .
O
O
R
RR
R R
R
R
R
R R
R
R
R
R R
w
w
R
R
=
+
=
+
=
+
= =
= =
= + =
= + =
1 2
1 2 1
2
2
1 2
2
2
1
2
1 2
2
2
2
2
2
2 2 1 2
50
150
100
150
011 0 44
011 0 2 0 44 01 0 05
011 2 0 44 01 0 07
c
c
c
c ( )
,
( )
( )
. .
[( . * . ) ( . * . ) ] .
. *. . * . .
/
,max
O
O
(a) In series
(b) In parallel
andims - FKM -
UTeM
41
Another example:
A mechanical speed control system works on the basis of
centrifugal force, which is related to angular velocity
through the formula:

F=mre
2
where F is the force, m is the mass of the rotating weights,
r is the radius of rotation, and w is the angular velocity of
the system. The following values are measured to determine
e :
r=20 0.02 mm, m=100 0.5 g and F=500 0.1%N
Find the rotational speed in rpm and its uncertainty. All
measured values have a confidence level of 95%.
andims - FKM -
UTeM
42
) % 95 ( sec / 30 . 1
) % 95 ( % 26 . 0 , 10 * 60 . 2
] 10 10 10 * 25 [
2
1
] )
500
5 .
( )
20
02 .
( )
100
5 .
[(
2
1
] ) ( ) ( ) [(
2
1
6 . 7 .
sec / 500 , 000 , 250 , ) (
3
2 / 1 6 6 6
2 / 1 2 2 2
2 / 1 2 2 2
2 2 / 1
level confidence rad w
level confidence
w
w
F
w
r
w
m
w w
Eq
rad
mr
F
F r m
+ =
+ =
+ + =
+ + =
+ + =
= = =

e
e
e
e
e
e
e
e e e
Solution
andims - FKM -
UTeM
43
Next Lecture

Signal Conditioning


End of Lecture 4