DEVELOPMENT OF A DECISION-

MAKING TOOL FOR BRIDGE

PRESERVATION AND MAINTENANCE
DR. RUEL SABELLANO, PE (MSU, MD
USA)

1
 • A bridge collapse along Forbes
Avenue near Pittsburgh’s Frick Park
PITTSBURGH’S BRIDGE on Friday, Jan. 28, 2022
COLLAPSE • At least 10 people were injured
• This was a warning call to bridge
managers across the US to repair
bridges that are in poor condition
• 53% of collapsed bridges were in
poor condition
• There is a backlog of about $125
billion of bridge repairs
• PA was set to receive more than
$327M in federal funding yet
Figure 1 Fern Hollow Bridge Collapse(Rittmeyer, collapsed bridge was not included
Guza, & Cato, 2022)
in the list of priorities 2
 There are more than 617,000 bridges
across the United States.
• 42% of all bridges are at least 50
ASCE REPORT CARD 2021 years old
• 46,154, or 7.5% of the nation’s
bridges, are considered structurally
deficient, meaning they are in
“poor” condition
• Nearly 231,000 bridges in all 50 states
still need repair and preservation
work
• Construction spending needs to be
increased from $14.4 billion annually
to $22.7 billion annual to increase
the condition
• States need more efficient tools in
bridge management
Figure 2 ASCE Infrastructure Report Card
2021(ASCE, 2021) • Need more refined deterioration
models 3
 2022 BRIDGE REPORT BY AMERICAN ROAD AND
TRANSPORATION BUILDERS ASSOCIATION (ARTBA)
US Bridge Inventory, by Rating • The number of bridges in
fair condition grew by
2020 2021 Change 20 to % Change 20 2,916 in 2021.
21 to 21 • Total for all state-focused
formula programs to
Good 278,427 278,122 -305 -0.1% increase by $14B
Fair 294,972 297,888 2,916 1.0%
• Bridges in fair condition
need to be preserved
Poor 45, 023 43,578, -1,445 -3.2%
and maintained,
otherwise there will be a
Total 618,422 619,588 1,166 0.2% massive number of
bridges that will become
poor in the next few
Table 1 US Bridge Inventory Condition Rating Change, 2020
to 2021(ARTBA, 2022)
years
4
 MOVING AHEAD FOR PROGRESS IN THE 21ST CENTURY
ACT (MAP-21) • Each state is required to develop a
risk-based asset management plan
• To maintain the highway
infrastructure asset system in a state
of good repair
• To strengthen America’s highway
systems
• National Bridge Inspection
Standards (NBIS) regulation was
enacted on April 27, 1971, which
requires periodic and thorough
Figure 3 MAP 21(ASCE, 2021) inspections of all bridges in the US
5
 NATIONAL BRIDGE INSPECTION
STANDARDS (NBIS) • Bridge Inspector’s Reference Manual (BIRM), is
a comprehensive manual on programs,
procedures, and techniques for inspecting and
evaluating a variety of in-service highway
bridges.
• Routine inspections interval not to exceed 24
months
• States collect NBI data and additional
component and more detailed element level
data for their bridge management programs
• States to prepare and maintain inventory of all
bridges subject to NBIS
Figure 4 Bridge Inspector’s Reference
6
Manual (FHWA, 2012)
 STRUCTURE INVENTORY AND
APPRAISAL (SI&A CODING GUIDE)
• States collect and retain SI&A data
according to the Coding Guide (1995)
• SI&A data that are significantly used in
this study includes structure number
(Item 8), detour length (Item 19),
maintenance responsibility (Item 21),
owner (Item 22), year built (Item 27),
average daily traffic (Item 29),
structure type (Item 43), structure
length (Item 49), deck width (Item 52),
condition ratings (Items 58-62),
structure length (Item 49)and year
Figure 5 Coding Guides 1995 and 2022 reconstructed (Item 106)
(FHWA, 2012) 7
 COMPONENTS OF A STEEL BRIDGE
• Deck is the surface of the
bridge. It maybe constructed of
concrete, steel, open grating or
wood
• Superstructure is the portion of
the structure that is the span
and directly receives the live
load
• Substructure consists of piers
and abutment shafts or walls,
Figure 6 Components of a steel bridge hammerhead, bed block,
(SSSBA, 2020)
pedestals and bearings
8
 METHODS OF INSPECTIONS

Hands-On Inspection (HOI)is an inspection technique with a high level of

scrutiny, in which access within arm’s length to each component of the
bridge is required.
Risk-Based Inspection (RBI)will utilize HOI techniques for specific defects of
priority. Generally, an RBI is required when defects in primary structural
components and safety concerns are deemed to be severe enough to
warrant interim inspection until the next round of hands-on inspections are
performed or the repairs associated with the defects are completed.
Visual Inspection (VI)for a structure is intended to identify the overall
condition of the structure and does not require a hands-on inspection.

9
 NBI Condition Ratings
Code Description Code Description
N NOT APPLICABLE 3 SERIOUS CONDITION - loss of section,
deterioration, spalling or scour have seriously
9 EXCELLENT CONDITIONS affected primary structural components. Local
8 VERY GOOD CONDITION - no problems noted. failures are possible. Fatigue cracks in steel or shear
cracks in concrete may be present.
7 GOOD CONDITION - some minor problems.
2 CRITICAL CONDITION - advanced deterioration
6 SATISFACTORY CONDITION - structural elements of primary structural elements. Fatigue cracks in steel
show some minor deterioration. or shear cracks in concrete may be present or scour
5 FAIR CONDITION - all primary structural elements may have removed substructure support. Unless
are sound but may have minor section loss, cracking, closely monitored it may be necessary to close the
spalling or scour. bridge until corrective action is taken.

4 POOR CONDITION - advanced section loss, 1 "IMMINENT" FAILURE CONDITION - major

deterioration, spalling or scour. deterioration or section loss present in critical
structural components or obvious vertical or
horizontal movement affecting structure stability.
Bridge is closed to traffic, but corrective action may
Table 2 NBI Condition Rating Scale(FHWA, 2012) put back in light service.
0 FAILED CONDITION - out of service - beyond
corrective action.
10
 In addition, the State DOTs has
started collecting total quantities
ELEMENT LEVEL DATA and quantities associated with
condition states for the National
Bridge Elements (NBEs), Bridge
Management elements (BMEs),
Agency Developed Elements or
also known as element level data
since 2015.
Each element is coded for the
following condition states:
CS 1 means good condition
CS 2 means fair condition
Figure 7 National Bridge Elements CS 3 means poor condition
(Soden, 2013)
CS 4 means severe conditions
11
 PRESERVATION ACTIONS OR REPLACEMENT
Bridge preservation encompasses both
rehabilitation and preventive
maintenance activities. Bridge
preventive maintenance includes
condition-based activities and cyclical
activities.
Bridge rehabilitation involves major
work required to restore the structural
integrity of a bridge as well as work
necessary to correct major safety
defects.
Bridge replacement as the total
replacement of a bridge with a new
facility constructed in the same general
traffic corridor.
12
Figure 8 Common recommended actions (FHWA, 2018)
 BENEFITS OF PRESERVATION ACTIONS
• Bridge preservation actions are
implemented to slow down the
deterioration rate and prolong
the service life of the structure.
• Extending the service life can be
translated into lower
maintenance, rehabilitation, or
replacement.
• Bridge preservation can be
costly, but it lowers annualized
Figure 9 Benefits of preservation costs over the life of the structure.
(FHWA, 2018) Programs for preservation in
each State DOT varies
depending on budget
availability
13
 MARYLAND DEPARTMENT OF TRANSPORTATION (MDOT)
CONSOLIDATED TRANSPORTATION PLAN (CTP)

MDOT manages the State

Transportation Trust Fund, which
gives Maryland the flexibility to
fund priority transportation
needs that support a balanced
statewide multimodal
transportation system. Funds go
the MTA, MAA, SHA, MVA, MPA
and Local Governmental
agencies. The MDTA is self-
sufficient and receives no gas
tax, motor vehicle fees, or other
revenue in the Transportation
Figure 10 MDOT Transportation Trust Fund
(FHWA, 2023) Trust Fund. 14
 MDOT CTP PROGRAM LEVELS

The proposed tool

can help decision
makers to predict
bridge condition and
effectively allocate
funds for the
maintenance,
rehabilitation, and
repair of these
bridges. This figure
shows the increasing
MDOT CTP Budget
since 2013.
Figure 11 MDOT Total Capital Program Levels (MDOT, 2023) 15
 PROBLEM STATEMENT

For years, bridge owners all over the country have been trying to find more
efficient ways to manage bridges they owned. The Maryland Department of
Transportation has recently updated the Transportation Asset Management
Plan (TAMP) and Lifecycle Management Plan (LMP). The various committees
involved in the development of these documents have expressed the need
of a decision-making tool to help them in the yearly or quadrennial
updating of these documents especially in predicting future conditions and
to determine the needs based on projected condition ratings. More refined
deterioration models are needed to identify future needs. Currently, there
are no deterioration models for each type of material for all 5,402 bridges in
Maryland. This development of a decision-making tool will significantly
impact the way bridge asset management will be handled.

16
 RESEARCH OBJECTIVES

1. To collect the bridge inspection data for 5,402 bridges in Maryland.

2. To develop a deterioration model using Normal and Weibull Distributions
as well as Machine Learning approaches.
3. To compare various lifespans derived from the deterioration models
4. To determine the factors that significantly influence the condition ratings
5. To investigate the accuracy of four Machine Learning Applications
6. To develop a decision-making tool which will allow users to predict future
conditions and prioritizing the bridges for appropriate preservation,
rehabilitation, and replacement plans with the state’s limited budget.

17
 Structure Kind (SI&A item 43A) Number of Structures

DATA COLLECTION Concrete 1300

A total of 5,402 bridges were used in Concrete continuous 124

this study. The breakdown per the
kind of material is shown in Table 3 Steel 2199
Out of the 5,402 bridges in Maryland,
2,199 or about 41% are steel bridges. Steel continuous 1025
This is followed by 1,300 concrete
bridges which 24% of the total Prestressed concrete 489
number and steel continuous which is
about 19%. The rest of the 16% are Wood/timber 180
prestressed concrete, concrete
continuous, wood or timber, masonry, Masonry 40

aluminum, and other kind of

Aluminum/wrought iron/cast iron 39
constructions materials. Most bridges
were built between 1950’s and 2000’s
Other 6
as shown in the next slide.
Table 3 MD Bridges Inventory(FHWA, 2021) Total 5,402 18
 DISTRIBUTION OF BRIDGES ACCORDING TO YEAR
BUILT

Figure 12 Distribution of bridges according to year built 19

 DATA CENSORING There is a total of 2199
steel structures, and
NBI Ratings - DECK
BIN LIST
2019 2018 2017 2016 2015 2014 2013 2012 2011 2010 2009 2008 2007 only 653, 697, and 548
200000B-0097010 6 6 6 6 6 6 7 7 7 7 7 7 7 for deck,
NBI Ratings - DECK (cont…) superstructure, and
2006 2005 2004 2003 2002 2001 2000 1999 1998 1997 1996 1995 1994 1993 1992 substructure
7 7 7 7 7 7 7 7 7 7 8 8 8 8 8 respectively have
Figure 13 Usable data usable data, resulting
in about 30% of
Various cases of unusable data include the structures with usable
following: any case where a structure stays in the data.
same rating from 1992 thru today, data with no
start or end points, data interrupted by
replacement or rehabilitation and abnormal drop
in ratings 20
DETERIORATION MODEL USING Normal Distribution
NORMAL AND WEIBULL • One-side normal distribution is used,
DISTRIBUTIONS or no negative standard deviations
will be used
• Use assumption that no structure will
deteriorate faster than the mean
• In normal distributions, the definition
of a Standard Deviation is that within
one Standard Deviation from the
mean, 68% of structures should be
accounted for. Two (2) Standard
Deviations will account for 95% of
Figure 14 Normal Distribution standard bell curve structures
21
 Weibull Distribution
DETERIORATION MODEL • A value of K>1indicates that the failure
rate increases with time
USING NORMAL AND
WEIBULL DISTRIBUTIONS • This happens if there is an aging process
• Generally used for deterioration models
• Line of best fit is for each material type is
used to find the scale factor
• Using the method of moments, the
Weibull Distribution deterioration model is
calculated. The values within the rating
drop tables each represent a different
data point that corresponds to the total
amount of time a structure was in that
certain rating. The more lines, the more
data points.
Figure 15 Normal Distribution using adjusted scale factor 22
 DETERIORATION MODEL USING NORMAL AND WEIBULL
DISTRIBUTIONS These calculated values
are the mean, and do not
Material Type: Component: Weibull Normal +/-

CONCRETE - Weibull vs. Normal

have any lower or upper
47.00 46.31

Deck 1.50% 35.73

Sub
44.00

CONCRETE 47.50 41.13

Super
41.13
47.50
bounds. This was done for
Super 15.48% Deck
46.31
47.00 the comparison, as the
range should only be
0.00 10.00 20.00 30.00 40.00 50.00
44.00 35.73 Normal Weibull

affected by the individual

Sub 23.14%

CONCRETE CONT. - Weibull vs. Normal

Deck
54.50 44.40

22.74% Sub
49.26
50.00
model. For example, in
CONCRETE
48.00 46.82
Super
46.82
48.00
timber, the deck has an
CONTINUOUS
Super 2.53%
Deck
44.40
54.50 expected lifespan of 38
years using the Weibull
0.00 10.00 20.00 30.00 40.00 50.00 60.00

50.00 49.26 Normal Weibull

Sub 1.51%
model, and 36.76 years
PRESTRESSED CONC. - Weibull vs. Normal

Deck
30.50 33.25

-8.28% Sub
40.45
using the Normal model.
37.50

PRESTRESSED Super
37.17
34.50
34.50 37.17
CONCRETE
Super -7.20% Deck
30.50
33.25
Figure 16a Lifespan Comparison (concrete,
37.50 40.45
0.00 5.00 10.00 15.00 20.00 25.00 30.00 35.00 40.00 45.00
concrete continuous, prestressed
concrete
Normal Weibull
23
Sub -7.28%
 DETERIORATION MODEL USING NORMAL AND WEIBULL
DISTRIBUTIONS Also note that the
STEEL - Weibull vs. Normal

Deck
37.00 38.93

-4.96% Sub
42.14
expected lifespan isn’t the
expected overall lifetime of
39.50

41.46
Super
STEEL 38.50 41.46 38.50

Super -7.14% Deck

37.00
38.93
the bridge, the expected
39.50 42.14
34.00 35.00 36.00 37.00 38.00

Normal
39.00

Weibull
40.00 41.00 42.00 43.00
lifespan is the expected
Sub -6.26%
number of years it will take
35.50 37.56
STEEL CONT. - Weibull vs. Normal
for a structure to go from
NBI CR 9 to NBI CR 3 (‘4 to
Deck -5.48% Sub

STEEL Super

CONTINUOUS
Super
43.50 38.91

11.80% Deck
Fail’) condition. This is
45.00 39.44
0.00 10.00 20.00 30.00 40.00 50.00 assuming no work is done
over the life.
Normal Weibull

Sub 14.09%

38.00 36.76
TIMBER - Weibull vs. Normal

Deck 3.37% 31.84

Sub
35.50

37.25
Super
TIMBER 40.00 37.25 40.00

Super 7.39% Deck

36.76
38.00 Figure 16b Lifespan Comparison
35.50 31.84
0.00 5.00 10.00 15.00 20.00 25.00 30.00 35.00 40.00 45.00 (steel, steel continuous, timber)
Normal Weibull
24
Sub 11.50%
 MACHINE LEARNING
Machine learning is widely used in many fields throughout the world including
the healthcare sector, transportation, advertisement, economics, and image
recognition. Machine learning is an application of artificial intelligence (AI)
that provides systems the ability to automatically learn and improve from
experience without being explicitly programmed. Furthermore, machine
learning at its most basic level is the practice of using algorithms to parse data,
learn from it, and then make a determination or prediction about something in
the world. There are two major categories of problems often solved by
machine learning i.e., regression and classification. The regression algorithms
are used for numeric data and classification problems include binary and
multi- category problems. Machine learning algorithms are further divided into
two categories including supervised learning and unsupervised learning
algorithms. The supervised learning algorithm is performed by using prior
knowledge in output values whereas the unsupervised learning algorithm does
not have predefined labels. In this study, the supervised machine learning
algorithms are used 25
 DETERIORATION MODEL USING
MACHINE LEARNING
• Data collected for 5402 bridges ( 69 of
which are not suitable) Most of the data sets
contain missing values. Therefore, in the
data processing step, the missing values are
replaced with meaningful values without
changing the structure of the data sets.
• Data analyzed using Jupyter Lab of Python
• Effects of each factors on the bridge
conditions were determined(13 factors)
• Model development using SVM, KNN,
Decision Trees and Random forest
algorithms
Figure 17 Machine learning approaches 26
MODEL DEVELOPMENT

• To develop the model, the data set is split into training and test
sets. 80% of the data set is considered for the training purpose and
the other 20%, which was not used in the training process, is used
for the evaluation process. The problem is considered as a
supervised classification problem, meaning, the data set is labeled,
the input variables are known, and the outcome, which is the
bridge rating, is a multi-class variable. National Bridge Inventory
(NBI) rating system classifies the bridge condition into 9 classes
which can be predicted by the proposed model based on FHWA
data. Some of the well-known machine learning classification
methods including Support Vector Machines, Decision Trees, K-
Nearest Neighbors, and Random Forest were employed.

27
 MACHINE LEARNING CLASSIFICATION
METHODS
• Support Vector Machine (SVM) is a supervised learning algorithm used for
regression, classification, and outlier detection. The main objective of SVM is to
create a line or a hyperplane which separates the data into classes.
• K-Nearest Neighbors (KNN) is a type of supervised machine learning (ML)
algorithm that can be used for both classification and regression predictive
problems. However, it is primarily used in the industry for classification and
prediction problems.
• Decision Trees are tree-based models that help in making decisions in both
regression and classification problems. To decide, they use a hierarchical
structure and split the dataset into smaller subsets.
• Random Forest is a commonly used machine learning algorithm that is
trademarked, which combines the output of multiple decision trees to reach a
single result. Its ease of use and flexibility have fueled its adoption, as it
handles both classification and regression problems.
28
 ARTIFICIAL NEURAL NETWORK (ANN)
ALGORITHM

In addition to the classifiers, an ANN algorithm is used which is

a parallel information-processing system that has certain
performance characteristics similar to biological neural
networks. A neural net consists of large numbers of simple
processing elements called neurons. Each neuron is
connected to other neurons by means of directed links, and
each directed link has a weight associated with it. These are
used to address problem that are intractable or cumbersome
with traditional methods

29
 VALIDATION METHODS
To evaluate the models, several accuracy measurements were
examined which are defined as follow.
Accuracy is a metric for classification models that measures the
number of predictions that are correct as a percentage of the
total number of predictions that are made.
Accuracy is a useful metric only when there is an equal
distribution of classes in our classification. If the classes are
unbalanced, accuracy is no longer a useful metric.
F1 Score is a simple way to solve unbalanced class problems
which takes into account not only the number of prediction
errors made by the model, but also the type of errors that are
generated during the model development process 30
 • True Positive (TP): the actual value and predicted
CONFUSION MATRIX value are the same. For instance, when the
model predicts the bridge has a condition rating
5 and it is rated 5 by NBI rating.
• True Negative (TN): the sum of all the values
which are predicted incorrectly for that specific
class. It is the sum of all the bridges whose
predicted condition rating differs from the actual
rating.
• False Positive (FP): the sum of values of
corresponding columns except the TP value (Type
I error). For instance, for the bridge with NBI rating
Figure 18 Confusion matrix 5, false positive is the sum of all the bridges which
Confusion matrix represents counts were rated 5 but their actual NBI rating is not 5.
from predicted and actual values, • False Negative (FN): the sum of values of
which are utilized for the performance corresponding rows except the TP value (Type II
evaluations of the methods used after error). For instance, for a bridge with a NBI rating
the classification. For binary 5, false negative, is the sum of all the bridges
classification, the scheme of the which were predicted to have a NBI rating not
confusion matrix is shown in Figure 18. equal to 5.
31
 Variable Variable Label
Name

INPUT VARIABLES Year built

Year
YEAR_BUILT_027
YEAR_RECOSTRUCTED_106
reconstructed
Years in Yr_in_serv
The thirteen (13) factors and the service
dependent variable which represents Traffic ADT_029
the deterioration grades were used to Truck traffic Percet_ADT_Truck
construct machine learning models. Bridge kind STRUCTURE_KIND_043A
Optimal configuration for the machine Bridge type STRUCTURE_TYPE_043B
learning models were obtained by Structure STRUCTURE_LE_MT_049
length
trying different methods for scaling the
variables, namely standardization and Deck width DECK_WIDTH_MT_052

normalization, and by using nonlinear Maximum Max_Temp

temperature
activation functions. Correlation
Minimum Min_Temp
between the input variables was
temperature
examined and the highly correlated
Mean Mean_Temp
variables are eliminated. temperature
Range Range_Temp
Table 4 Input variables temperature 32
 CORRELATION MATRIX (WITH HIGHLY CORRELATED VARIABLES)
Optimal configuration
for the machine
learning models were
obtained by trying
different methods for
scaling the variables,
namely
standardization and
normalization, and by
using nonlinear
activation functions.
Correlation between
the input variables
was examined and
the highly correlated
variables are
Figure 19 Highly correlated variables eliminated. 33
 DISTRIBUTION OF DATA

Figure 20 Distribution of data, Percentages of the

bridges with certain CR

Based on the data distribution shown in the Figure 20., there are only a few
bridges with ratings of 1, 2, 3, 4, 8 and 9 in the data set, more than 2,000
bridges have a rating of 6, and more than 2,000 have ratings of 5 and 7
combined.
34
 MODEL VALIDATION AND COMPARISON
Model Accuracy Accurac
Score y Score
F1
Score
F1
Score
F1 Score for
CR 7 for the
• Training set consisting of 4266 bridges
Training on Test for CR 5 for CR 6 Validation Set of the data set, and they have been
Set Set for the for the
Validati Validati validated using the proposed
Support 55% 53%
on Set
56%
on Set
79% 60% methods on the test set including 1067
Vector
Machine
bridges.
K-nearest 67% 50% 61% 77% 73%
neighbors • SVM, KNN, and ANN Models
Decision
Tree
99% 60% 61% 75% 66% employed in this study were
Random 99% 60% 52% 74% 66% evaluated and results show that they
Forest
ANN 60% 50% NA NA NA all perform similarly on the training
and test sets.
Table 5 Model comparison
• The Decision Tree and Random Forest
models perform better with an
• These models can predict bridge accuracy of 99% on training set as
condition rating with an excellent compared to the accuracy on the
accuracy for rating 5, 6 and 7 test set.
35
 FEATURE IMPORTANCE
• The estimated numbers indicate
the relative importance of each
factor affecting a bridge
condition rating. A larger value
shows that the factor has a larger
impact in accelerating the
deterioration. The following
factors, numbers of years in
service, ADT, structure length,
Figure 21 Feature importance
Factors Description Importance structure width and the
Yr_in_Srv
ADT_029_2021
Years in service
Average annual daily traffic
0.25677
0.17928 temperature have a significant
STRUCTURE_LE_MT_049 Length 0.14523
impact on the deterioration of
DECK_WIDTH_MT_052 Width 0.12518
Range_temp_2021 Difference between max and min temperature 0.09796 the bridges with the highest
STRUCTURE_KID_043A Bridge structure kind 0.08496
PERCET_ADT_TRUCK_109 Annual daily truck traffic 0.07002 importance rates.
STRUCTURE_TYPE_043B Bridge structure type 0.04058

Table 6 Importance features 36

 The lifespan for
CALCULATED LIFESPAN SUMMARY concrete
continuous bridges
for deck,
superstructure and
substructure is the
highest. Followed by
steel continuous,
then prestressed
bridges. The
presence of joints
has always been
known as the major
issue in the
deterioration of the
Table 7 Calculated lifespan for deck, superstructure and substructure bridge components. 37
 CALCULATED LIFESPAN FOR DECK PER MATERIAL
140

The lifespan of decks for

126
122
120 115

continuous concrete
107
104
98 101
100 95

bridges can reach up to

89 90
86 86
82 82 81 82 80
Lifespan (yr)

80 76

126 years. This is possibly to

69 71 70 71
67 65
60 58
55
the absence of deck joints
60 52 53 52
49 50
41 39

which is a common
40 35
29

20
problem in simple span
0
bridges. This can be
observed between
concrete and concrete
Material
continuous or steel and
0 S.t.D. 1 S.t.D. 1.5 S.t.D. 2 S.t.D.
steel continuous.
Figure 22 Calculated lifespan for deck per material
38
 CALCULATED LIFESPAN FOR SUPER PER MATERIAL
140

120
126 127

113
The lifespan of
112
superstructure for
108 109
106
100 96 96
92
continuous concrete
89 91 90 88
86
79 79 80 80
Lifespan (yr)

80 74 76

bridges can reach up to

72 70
66 64
63
57 59

127 years. This is possibly to

60 52 54
47 46
42 40 39 37
40
the absence of deck joints
20
which causes various issues
0
in the superstructure. This
can be observed between
concrete and concrete
Material
continuous or steel and
0 S.t.D. 1 S.t.D. 1.5 S.t.D. 2 S.t.D.
steel continuous.
Figure 23 Calculated lifespan for superstructure per material
39
 CALCULATED LIFESPAN FOR SUB PER MATERIAL
140
123
130
The lifespan of substructure
for continuous concrete
120 114 112 112
111
105 104

bridges can reach up to

98 96 95 95
100 92 93
90 91
86 84
81
130 years. This is possibly to
79 80
Lifespan (yr)

78 77
80 75
72
66 67

60
49
53
60

49 48 50
the absence of deck joints
46 45
40
40
which is a common
20
problem in simple span
0
bridges. This can be
observed between
concrete and concrete
Material
continuous or steel and
0 S.t.D. 1 S.t.D. 1.5 S.t.D. 2 S.t.D.
steel continuous.
Figure 24 Calculated lifespan for substructure per material 40
DECISION-MAKING FLOWCHART

Figure 25 Decision-making framework

41
 UPDATE NBI data from FHWA website
• Tool can download new NBI data from FHWA website for
future years in a mostly automated way
• User just needs to save data file locally and then tool
automates the rest

Figure 26 Update data

42
 INITIAL PROJECT INPUTS

The inputs include certain information

determined by the user or the bridge
owner such as budget, project duration,
and year. Each agency has specific
budget for every fiscal year and is the
primary limiting factor when selecting
multiple projects. Duration is either withing
a 10-year plan or a 25-year long range
plan. The year usually starts from the
current fiscal year.
Figure 28 Initial project inputs
43
 PRIORITY SCORE CALCULATION
Using several factors such as
time/duration in current rating, CS3/CS4
Element Level Data values, detour length,
bridge age, and AADT were determined
to affect any given structure. Along with
the NBI condition ratings, a priority score
can be calculated for each structure in
the decision-making tool. Each sub-
component (deck/super/sub) has a
corresponding weight that governs how
important each sub-component is to the
overall health of the structures. Figure 26
shows the top 20 worst ranked bridges out
Figure 27 Prioritize bridges
of 5,402 bridges in Maryland.
44
 The costs for all repair types are found in
SELECTION OF STRUCURES the same ways for all structures. It uses the
Deck Area of any given structure and unit
costs developed by a state agency to
assign a repair/replacement cost for each
of the different repair types. Each repair
type can be selected by the user for
repair recommendations in the cash flow
tables. Structural types will alter the NBI
Ratings in the year after
repair/replacement.
Cyclical types will not alter NBI Ratings in
the year after cleaning/sealing/painting.
Using item 8 from SIA, the following can be
determined:
Figure 29 Ownership sorting
• Responsible for maintenance
• What county
• Ownership
45
 ITEM 8 - SIA

Figure 30 SI&A Item 8

46
 PRIORITY RANKINGS AND PROJECT RECOMMENDATIONS
TOP 20 PRIORITY RANKINGS BASED ON USER SELECTION • Worst ranked structures according
STRUCTURE NAME
300000CHZ301010
TYPE
BRIDGE
MATERIAL
STEEL
RANKING
158 to ownership selected
300000BCZ472010 BRIDGE STEEL CONTINUOUS 259
300000BCY127010
300000BCY080010
BRIDGE
BRIDGE
STEEL
STEEL
358
458
• Once structures are verified, the
300000BCY102010
300000BCY129010
BRIDGE
BRIDGE
STEEL
STEEL
558
658
next step is to make repair
300000BCY125010
300000BCY118010
BRIDGE
BRIDGE
STEEL
STEEL
758
858
recommendations
300000BCY121010 BRIDGE STEEL 958
300000BCY085010 BRIDGE STEEL 10
58
300000BCY082010 BRIDGE STEEL 11
58
300000B-MT01020 CULVERT CONCRETE 12
56
300000B-MT02020 CULVERT CONCRETE 13
56
300000B-NR02010 BRIDGE STEEL 14
58
300000BCY126010 BRIDGE STEEL 15
58
300000B-X625020 CULVERT CONCRETE 16
56
300000BCY112010 BRIDGE STEEL 17
58
300000BCW586010 BRIDGE PRESTRESSED CONCRETE 18
60
300000BCY078010 BRIDGE STEEL 19
58
300000H-X765013 BRIDGE STEEL 20
58

Figure 31 Top 20 MDTA bridges and work

recommendations

47
 COMPARE SELECTIONS
After repair recommendations,
the proposed decision-making
tool can generate a pdf report
that will include all relevant
historical and selection
information from User Interface.
Multiple scenarios can be
created using this functionality.
A Project Summary page is
available where all structures and
their corresponding repair
recommendations will be
summarized including costs and
NBI Rating changes.

Figure 32 Project summary and over budget detection 48

 GENERATE REPORT

Overall budget needs

report/page will enable the
Figure 33 Generate report for evaluation
user to make some
adjustments on the
Once contracts are created cashflow or the selection of
and the total $ per year has repair types/actions based
been verified final selection
on the yearly budget. The
can then be printed into PDF.
decision-making tool will
Print out includes:
enable the bridge owners
• Summary to prioritize projects
• Cash flow
• Select projects summary 49
 DASHBOARD (HISTORICAL AND INVENTORY)

Figure 34 Dashboard (historical and inventory section)

50
 DASHBOARD (GENERAL SUMMARY SECTION)

Figure 35 Dashboard (general summary section)

51
 DASHBOARD (SELECTION SECTION)

Figure 36 Dashboard (selection section)

52
 DECISION-MAKING TOOL VERSION 1.0
Current functionality
• User interface will lock out the user
until correct log in credentials are
provided
• Using macros new data can be
downloaded easily from FHWA
website
• Tool can prioritize bridges
according to ownership
• Repairs recommendations will be
summarized including costs and
NBI rating changes
• Reports can be printed
• Dashboard summarizes the history
Figure 37 Decision-making tool version 1.0 for alpha testing and selected projects 53
 CONCLUSION
• In this study, a methodology to determine deterioration models using
Normal and Weibull Distribution as well as Machine Learning models were
developed to predict bridge condition rating based on data extracted
from the FHWA website for 5,402 bridges located in Maryland and were
used in the development of a decision-making tool.
• Various lifespans of different types of materials used for bridge
construction were calculated such as steel bridges which are the most
common in Maryland have projected lifespan of up to 90 years, concrete
continuous bridges have the longest lifespan of up to 130 years. The other
material types have the following lifespans: up to 123 years for concrete,
up to 114 years for steel continuous, up to 112 years for prestressed
concrete. Other material types ranges from 90 to 100 years. The lifespan
calculations is the backbone of the decision-making tool.
54
 CONCLUSION

• Various machine learning algorithms provided for assurance of

effectiveness of factors affecting deterioration. The models
developed were used to estimate the level of importance of the
influential factors.
• The developed Machine Learning models used 13 influential factors
as input variables and the bridge condition rating as the output
variable. The models predicted bridge condition ratings in the range
between 5 (fair) to 7 (good) with an accuracy up to 79%.
• From the estimated importance factors, it was observed that the
number of years in service (25.6%), ADT (17.9%), structure length
(14.5%), structure width (12.6%), and temperature (9.8%)have
significant impacts on the deterioration of bridges.
55
 CONCLUSION
• The decision-making tool can download NBI CR directly from FHWA
website by implementing macros. This will enable the user to
recalculate priorities periodically.
• The ownership sorting feature of the tool is very helpful since MD
bridges are owned by various agencies or entities.
• Using MDTA as a case study, the top 10 recommended structures were
verified and found to be have been included in the CTP or Long-
Range Needs.
• This project has received support and interest for implementation of
the tool from various agencies such as Baltimore County, Howard
County, MDTA and MDOT SHA

56
LETTER OF SUPPORT FOR IMPLEMENTATION

Figure 37 Expression of support from Howard County 57

LETTER OF SUPPORT FOR IMPLEMENTATION

Figure 36 Expression of support from MDOT SHA 58

 PEER-REVIEWED PUBLICATIONS AND GRANT PROPOSALS
• A paper with the same title of this dissertation was presented on July 22, 2022, at
the Short to Medium Span Bridge Conference (SMSB) 2022 in Toronto, Ontario,
Canada.
• A related paper entitled Evaluating and Predicting Deterioration of Bridges Using
Machine Learning Applications has been accepted and has been published in
the Structures Congress 2023 conference proceedings. The conference was held
from May 3-7, 2023, in New Orleans, Louisiana, USA
• Submitted a proposal to the Center for Integrated Asset Management for Multi-
modal Transportation Infrastructure Systems (CIAMTIS) and proposal is currently
funded (USDOT UTC); Presented at TAIM (PennState Univ.) on September 1, 2023
• A related paper entitled Developing of Bridge Deterioration Predictive Models has
been accepted and will be published in the International Bridge Conference 2024
proceedings. The conference was held from June 3 to 5, 2024, in San Antonio,
Texas, USA
59
 FUTURE IMPROVEMENTS

• NBI CR historical data was found to be subjective and full of missing

information. The newer NBE or Element Level Data which started in
2015 will provide more accurate deterioration models especially for
the Machine Learning approaches in which this adaptive tool can
be implemented
• The cost information can also be improved by using machine
learning applications in unit cost determination based on historical
construction costs. These costs can be broken down into cost per
element to match with the new deterioration model which will use
element level data.

60
Q&A

61
DR. RUEL SABELLANO, PE,
M. ASCE
[email protected]
[email protected]
[email protected]
FB/MESSENGER ACCOUNT: CIVIL YANO