Engineering with Computers

https://doi.org/10.1007/s00366-019-00874-2 (0123456789().,-volV)(0123456789().
,- volV)

ORIGINAL ARTICLE

Can past failures help identify vulnerable bridges to extreme events?

A biomimetical machine learning approach
M. Z. Naser1

Received: 5 August 2019 / Accepted: 9 October 2019

 Springer-Verlag London Ltd., part of Springer Nature 2019

Abstract
With limited resources to properly maintain and upgrade transportation infrastructure, bridges often end up exceeding their
expected service lifespan; thus, becoming vulnerable to the adverse effects of aging and extreme loading conditions. In
order to better assess the vulnerability of these structures, this study showcases the outcome of an observational analysis
that utilizes biomimetical (bio-inspired) machine learning algorithms to predict the vulnerability and expected degree of
damage in bridges in the aftermath of an extreme loading event (such as fire, flood, earthquake, etc.). These algorithms
comprise deep learning, decision tree, genetic algorithm and genetic programing and were trained and validated using 299
international incidents covering a wide variety of bridge systems/configurations, traffic demands, etc. Based on this
analysis, user-friendly assessment tools that can be used to evaluate propensity of a given bridge to undergo high levels of
damage and/or collapse are developed. These tools can aid designers and decision-makers in evaluating performance of
new or existing bridges against a variety of hazards, as well as in developing relevant design strategies for mitigating
disaster-induced failures as to minimize disruptions to supply chain operations and/or evacuations during an emergency.

Keywords Bridges  Machine learning  Collapse  Extreme events  Classification

1 Introduction varying from unintentional errors (barge/tanker impact,

vehicle collision) to those that occur intentionally (such as
Bridges are critical infrastructure that provide commuters arson, sabotage, terrorism, etc.) [2, 3].
with a primary mean for ground transportation. Bridges In a nutshell, it is due to the rapid urbanization and
also play other key roles especially in supply chain oper- increasing frequency and intensity of hazards that bridges
ations or mass deployment/evacuation during disasters. are experiencing far more damaging effects than those
Due to their strategic importance, bridges are often being traditionally designed for. Furthermore, due to their
designed to endure a multitude of loading conditions as remoteness and in order to satisfy logistical and working
they are expected to serve over a number of decades [1]. operational requirements, bridges are easily accessible and
Giving this relatively long service lifespan, bridges must are open to public, with minimum or no security at all
withstand a continuous increase in demand (population [4, 5]. In lieu of the above two factors, one should keep in
size), exposure to harsh seasonal weathering (snow, rain, mind that the majority of bridges are not only deteriorating,
etc.) as well as one-of-a-kind extreme loading events. Such but are also inadequately maintained [6]. It goes without
events can occur in response to climate change (i.e., tsu- saying that the combination of all of the aforementioned
namis, wildfires) or natural occurrences (earthquakes, factors seem to yield unfavorable conditions that jeopar-
mudslides) and can also be triggered by human elements dize the integrity of bridges. This has been duly noted in
recent and notable studies [7, 8].
This has been also documented by the American Society
& M. Z. Naser of Civil Engineers (ASCE) in its 4-year report card which
[email protected]; [email protected]
http://www.mznaser.com depicts the condition and performance of bridges in the
United States. The latest report card, published in 2017,
1
Glenn Department of Civil Engineering, Clemson University, scored US bridges with a ‘‘C?’’ and showed that 9.1% of
Clemson, SC 29634, USA

123
 Engineering with Computers

available bridges are structurally deficient1 and 13.6% of Whenever a bridge is damaged (whether due to fire or
the same bridge population are functionally obsolete2 any other incident), carrying out a proper investigation,
(these percentages translate to about 56,007 and 83,556 out inspection and maintenance is required before the bridge is
of the 614,387 bridges that were in service in 2016). The put back into operation. Closing a bridge for maintenance
same report card also shows that 40% of all bridges are at triggers a domino effect that starts by detouring traffic to
least 50 years of age and an additional 15% are between nearby routes and disrupting supply chain operations—thus
the ages of 40 and 49. Interestingly, the ASCE’s report imposing significant traffic delays. This adversely stresses
highlights three alarming figures: (1) about 188 million the flow of traffic and adversely affect commuters’/busi-
daily trips take place across structurally deficient bridges, ness’ convenience [18, 19].
(2) rehabilitation cost for bridges is estimated at $123 Such disruptions can be mitigated by identifying vul-
billion, and (3) the typical bridge remains in service for nerable bridges that could undergo severe damage or would
about 43 years. require shutting down/prolonged repairs post a traumatic
The convergence of aforementioned effects has led to an event. Such identification can come in handy in a number
increase in bridge failure incidents over the last decades of practical scenarios. For example, identifying vulnerable
[5, 9–11]. A close examination of such incidents reveals bridges to scour or flooding can provide insights into
that the majority of working bridges failed as a result of a optimal preparation for an upcoming disaster (flood/storm)
direct exposure to an extreme event (i.e., fire) or due to an season. In this scenario, vulnerable bridges can either be
event-triggered incident (i.e., scour/hydraulic failure during structurally strengthened before the start of the disaster
a storm, etc.) [5, 12]. It is interesting to note that these season or closed for public (during the season). Not only
failures continue to take place despite the use of modern identifying such a bridge may minimize commuter losses,
construction materials (e.g., high-performance concretes, but it is also be of importance to authorities while planning
high-strength low-alloy steels) as well as implementation routes for emergency response and evacuation. In a similar
of improved knowledge in planning and structural engi- fashion, pinpointing a given bridge’s susceptibility to a
neering (as evident in the regularly updated bridge design particular hazard of low probability (i.e., earthquake or fire)
codes and standards). may lead to implementing upgrades and/or policy changes.
However, one should still note that while a good amount For example, bridges prone to damage from an earthquake
of work has been carried out to advance the state of could be upgraded with energy absorbing dampers. Simi-
knowledge on traditional loading events (i.e., wind, storm, larly, structural members in bridges susceptible to fire
earthquake, and impact) [13–15], notable studies have hazard (i.e., timber/steel girders) can be installed with fire
pointed out the lack of codal provisions in the case of fire insulation. These bridges may also enforce certain policy
hazard [1, 8, 13]. This has led to an exponential increase in changes, i.e., heavily loaded fuel tankers may only pass
bridge failures due to fire incidents [1, 12, 16]. Bridge fires during certain hours (outside rush hours) or weather con-
can occur due to wildfire, arson, or primarily as a result of ditions (by avoiding foggy/rainy days) [4].
vehicles collision, i.e., barge/fuel tankers [2–6]. Since As one can imagine, identifying vulnerable bridges is
collisions occur at high speeds, the impact of such incident expected to be a hectic and tedious task. This is given the fact
releases dense thermal energy in a short burst of time and that the outcome of a bridge incident (viz. expected damage)
causes burning of fuels. Gasoline fuels are of low flash is a governed by a variety of factors. These factors cover a
points and extremely flammable and hence could reach number of fronts such as: bridge characteristics (i.e., structural
800–900 C within few minutes of burning [17]. The system, type of construction materials, age), incident features
combination of mechanical (i.e., impact force) and thermal (e.g., type, intensity, duration), etc. Even in the case where
loadings can negatively affect the structural integrity of a information on all of these factors are collected, few questions
bridge; often developing thermally induced forces/phe- arise: (1) what is the proper mean to analyze these factors? (2)
nomena (i.e., spalling, buckling) leading to major damages How to validate the outcome of such analysis? (3) How such
or collapse [11]. an analysis be scaled for bridges of different characteristics
and/or for various extreme conditions? and (4) How can this
process be automated?
1
Structurally deficient bridges are those that have been restricted to While a practicing engineer/decision-maker may strug-
light vehicles, closed to traffic or require rehabilitation. A structurally gle to answer some/all of these questions, this work
deficient bridge is one for which the deck, the superstructure or the hypothesizes that it is possible to arrive at reliable answers
substructure are rated in condition 4 out of 10 or less [57].
2
to these questions by adopting an untraditional analysis
Functionally obsolete bridges are those built to standards that are
scheme. This analysis leverages novel and bio-inspired
not used today. Such bridges do not have adequate lane/shoulder
widths or vertical clearance to serve current traffic demand, or those machine learning techniques that are specifically tailored to
that may be occasionally flood [57]. comprehend the highly complex nature of relations

123
Engineering with Computers

Table 1 A list of notable bridge failures due to extreme events

Bridge Date Cause Structural damage Human
losses

United States
I-85 in Atlanta, GA [33] March 30, Vandalism/burning of stored 28-m section collapsed due to severe –
2017 PVCs heat
I-10 Twin Span Bridge, LA [34] August 29, High wind (Hurricane Katrina) Significant damage to superstructure 8 injures
2005
Sunshine Skyway Bridge, FL [35] May 9, 1980 Barge collision (MV Summit Over 370 m span collapsed into the 35 deaths
Venture) river
Worldwide
Beaver River Trestle Bridge, June 12, 2012 Arson Major damage/bridge had to be shut –
Canada [36] down
Jiujiang Bridge, China [37] June 15, 2007 Struck by vessel Partial collapse 9 deaths
Valigonda Railway, India [38] October 29, Flooding Bridge collapsed 114 deaths
2005

between all of the aforementioned factors. Such tools have 2 Analysis on recent bridge failures due
been recently used to unlock hidden mechanisms in com- to extreme loading events
plex problems in the areas of materials science [20–22],
forensic engineering [23, 24], transportation planning There has been a spike in bridge failures in recent years [5].
[25, 26], etc. A number of notable works have advocated Though these failures tend to vary in aspects with regard to
the potential of artificial intelligence into modern con- bridge configurations, cause of failure, type of event, etc., a
struction and planning and as such is the main drive behind common feature in these failures clearly show the vulner-
this work [27, 28]. ability of bridges to extreme loading conditions. This
The development of such assessment tools would section highlights well-documented statistics and bridge
greatly enhance the state of planning, design and mainte- failures that occurred due to the aforementioned events
nance of bridges. In order to pursue such goal, this paper over the past few decades.
explores utilizing artificial intelligence to develop bio-in- To start, the New York state department of transporta-
spired machine learning-powered expert systems that can tion conducted a comprehensive survey on failed bridges
comprehend the multi-dimensional relations governing across 18 states in the United States (US) covering the
bridges’ vulnerability to extreme events. These expert period of 1960–2008 [29]. A summary of this survey shows
systems comprise deep learning, decision tree, genetic that a total of 1746 bridges collapsed/deemed unservice-
algorithm and genetic programing and were trained and able. Of these bridges, about 1006 bridges collapsed due to
tested using 299 actual bridge failure incidents. These hydraulic issues (i.e., scour, flooding) and 228 bridges
incidents were specifically picked to cover a wide variety failed due to collision/impact incidents. Further, 52 bridges
of structural systems/configurations, event causes/types, collapsed due to fire (about half of these were timber
magnitude of damage, etc. The result of this intelligent bridges that were damaged by wildfires), and only 19
analysis helped in developing user-friendly assessment collapsed due to earthquakes. The outcome of this survey
tools that can be used to evaluate propensity of a given seems to fall in line with those carried out by US-based
bridge to experience high level of damage and/or collapse researchers such as Harik et al. [30] (covered incidents
in the aftermath of a fire, high wind, storm, earthquake, or from 1951 to 1988 and showed that 37% of bridge failures
impact/collision event. Further, these tools can be used were due to collision), Wardhana and Hadipriono [31]
during the design stage of new bridges as well as for (covered bridges in the period from 1989 to 2001 and
existing and/or historical bridges. These tools take into pointed out that 53% of investigated failures were due to
account geometric, material, age and design features of hydraulic issues followed by 12% due to collision/impact,
bridges and also the critical nature of the bridge from traffic 3.4% and 3.2% due to earthquake and fire, respectively), as
functionality (i.e., geographic location and expected daily well as the more recent survey conducted by Cook et al.
traffic). The outcome of this work shows the practicality of [32] (92 bridge failures were identified and were primarily
machine learning tools in assessing damage as well as in attributed to hydraulic causes (52%) and collisions (20%)).
paving the way towards developing relevant design Table 1 lists some of the notable bridge failures in recent
strategies for mitigating disaster-induced bridge failures. years.

123
 Engineering with Computers

A closer look into the open literature shows that there is incidents covered bridges in the US, and 88 bridges were
very limited information and statistics on bridge failures in collected from around the world). This information covers
the US; with a similar trend observed in bridges in Europe, various aspects on characteristics of bridges, associated
Asia, and South America. Only few studies have properly traffic demands as well as event type and damage magni-
documented some bridge failure incidents that occurred tude. Each of these bridges was selected on the following
outside the US (see Table 1) [1, 5, 11, 37, 39]. In fact fewer basis: (1) it was subjected to an extreme event, and (2) had
works have reported statistics on overall bridge dam- sufficient amount of information on key aspects of bridge
ages/collapses that occurred in the aftermath of an extreme and incident available. It should be noted that the complete
event (as opposed to other issues such as fatigue, over- list of these bridges is provided in the Appendix. Figure 1
loading, construction errors, etc.) [40, 41]. In one early shows key statistics obtained from this database.
study, Smith [42] covered failed bridges in Europe in the Out of the massive amount of data collected, a challenge
period between 1847 and 1975 and identified 70 bridge is to identify key characteristics and features in order to
hydraulic failures and another 4 due to high wind. In a arrive at optimally designed algorithms that better com-
more recent work, Fu et al. [41] reviewed a total of 157 prehend the nature of event-induced bridge damage as well
bridge collapses that were reported by the Chinese media as link each bridge’s characteristics and features with
between 2000 and 2012. In their review, these researchers expected magnitude of damage during an extreme event.
showed that the majority of these collapses were caused by From construction and structural engineering points of
natural disasters (with flooding being the leading view, the response of a given bridge to an extreme event is
cause * 45.8%—or 72 bridges out of all examined a function of the structural characteristics of such a bridge,
bridges). Xu et al. [37] also reviewed 171 bridges during together with traffic features (from geographical and traffic
their service lifespan in China (between 2000 and 2014) functionality perspectives), and type of event the bridge is
and showed how an average of 20 bridges may collapse exposed to. The validity of these characteristics and fea-
annually. Biezma and Schanack [43] carried out a world- tures was also cross-checked by examining recent works,
wide survey that examined failure of steel bridges and as well as consultation/recommendations with practicing
noted how 80% of identified failures (in about 350 bridges) engineers [1, 10, 11, 40, 44]. The rationale behind the
occurred due to earthquake, sabotage, scour or selected characteristics and features is further outlined
impact/collision incidents. below.
The outcome of the above surveys stresses the: (1)
limited works in this research area; (2) vulnerability of 3.1 Structural characteristics
bridges to severe loading conditions; (3) rise in bridge
failure frequency; (4) globality of the issue of bridge fail- The structural characteristics that govern the vulnerability
ures, and (5) dire need for assessment tools to better assess of a bridge arise from the type of structural system and
bridges and mitigate unforeseen failures. This work aims to construction material used in load bearing elements, as
narrow some of these knowledge gaps. well as the span and age of the bridge. These were arrived
at by noting how the structural capacity of a given bridge is
influenced by these characteristics (i.e., sectional capacity
3 Data collection and development of a load-bearing member in a bridge is a function of the
of database above factors). For example, typical ‘‘load-bearing sys-
tems, S’’ in bridges are assembled under truss/arch, girder-
In order to successfully apply biomimetical machine type (box or I-shaped), cable-stayed and suspension. The
learning algorithms/tools, one must develop a properly last two types of bridges are notorious for having complex
designed database. Thus, a literature survey was first car- load-paths, comprising larger spans and attend to large
ried out to identify well-documented bridge failures that volumes of traffic. Thus, these bridges are very susceptible
occurred as a result of an extreme event. As discussed to lateral and sudden loadings due to the use of steel cables
above, five different extreme events were considered here: and slender elements. On the other hand, bridges with more
fire, high wind, scour/flood, earthquake or impact/colli- traditional structural systems (i.e., trusses, arches, girders)
sion.3 The outcome of this extensive survey led to col- are often designed to be compact and hence have higher
lecting information on 299 international bridges (with 211 redundancy and are easier to design/construct.
Every load-bearing member in bridges is either fabri-
3
cated using concrete, steel or timber materials. While most
Unlike other works [30–32, 37, 40–43], many of which were cited
‘‘construction materials, M’’ seem to perform well under
in the previous section, bridges that failed during construction/
demolition, or due to overloading, fatigue, and other similar causes normal/ambient environments, some materials may per-
were not considered herein. form better under certain conditions. For instance,

123
Engineering with Computers

(a) (b)

(c) (d)

(e) (f)

(g) (h)
Fig. 1 Statistics of the developed database

123
 Engineering with Computers

structural members made of steel and timber are suscepti- noted that the number of lanes in a bridge indirectly reflects
ble to fire damage due to the high conductance of steel and the average number of vehicles traveled per day as well as
combustibility of timber. This is unlike concrete bridges, bridge’s load carrying capacity and thus is of significance
especially those without propensity to fire-induced spalling from this work’s perspective. The average number of lanes
[2, 29]. Concrete bridges can be classified under two shown in Fig. 1f is 2.9 lanes with a maximum and mini-
groups, i.e., reinforced concrete, or pre-stressed concrete. A mum of 11 lanes and 1 lane, respectively.
hybrid combination between steel and concrete bridges is
that they comprise steel–concrete composite construction. 3.3 Type of extreme event
Since bridges are massive structures, then the ‘‘span, P’’
of a bridge primarily governs the associated magnitude of As discussed earlier, five different ‘‘types of extreme
the level of loading it can carry (i.e., traffic volume) and events, T’’ were considered here: fire, high wind, scour/
hence indirectly its structural capacity and design. Thus, an flood, earthquake or impact/collision. While the magnitude
assumption is made that the vulnerability of a bridge to an of an event (i.e., bridge fire) may not share exact features to
extreme event is directly correlated to its span length (e.g., all other fire events collected in the compiled database, the
large span bridges have high dead to live load ratio and adverse effects of a fire (by imposing thermal and
serve heavy traffic). Also, large spanned bridges usually mechanical loadings) on the structural response of a given
span over natural obstacles (e.g., rivers, valleys, etc.) and bridge is more or less similar in all cases. The same also
these require special construction and attention. The aver- goes for other types of events (e.g., high wind/earthquake
age span of all bridges in this database was estimated at generates a combination of vertical and lateral dynamic
103.4 m, with a maximum of 1900 m and a minimum span loadings, impact and scour incidents generally affect sub-
of 4.8 m. structure of a bridge, etc.). It needs to be stressed that this
While bridges are built to continuously be in service for work does not specifically account for the magnitude of an
a number of decades, load-bearing components in these extreme event (i.e., 2 MW fire, 180 mph wind, etc.), but
structures undergo constant deterioration resulting from rather hypothesizes that the type of the event is of impor-
environmental effects (corrosion), overloading effects (fa- tance.4 This assumption is only valid if the collected data
tigue), etc., thus requiring regular inspection and preser- points also reflect upon this assumption. As such, bridge
vation. As an example, existing bridges often suffer from incidents of similar characteristics were only collected
excessive cracking in concrete, corrosion in steel, or herein. That is to say, the only incidents included are those
delamination/rotting in timber and this reduces the struc- of an extreme nature (i.e., an incident wherein a bridge
tural capacity (moment or shear capacity) and integrity undergoes a fire due to a motorcycle accident or a minor
(causing sagging) of such bridges. To account for these earthquake (i.e., of a magnitude of 2.0) was not included
effects, the ‘‘age, A’’ of a bridge is considered of impor- into this database). To better illustrate the magnitude of
tance. Figure 1d shows that the average age of bridges at each type of events used in this study, Table 2 describes
the time of incident = 44.6 years with standard devia- the key aspects behind each event type.
tion = 35.8 years (with a maximum and minimum ages of It could be argued that other structural characteristics
reviewed bridges as 149 years and 1 year, respectively). and traffic features can also be added into the compiled
database such as bridge rating, average daily traffic,
3.2 Traffic features expected human losses, etc. Some of these factors were not
explicitly included herein due to the lack of information on
Traffic demand is another factor that governs the design, such factors (especially for older bridges and those of
and by extension the structural response of a bridge. Two international origin). In any case, the compiled database
main factors are identified herein, ‘‘location/geographical can be extended (both horizontally: by adding new factors,
significance, L’’ and ‘‘number of lanes, N’’. as well as vertically: by adding new bridge incidents) upon
To distinguish between location/geographical signifi- the availability of new incidents. All that is needed is to
cance of different bridges, these structures are grouped collect information on a new factor (i.e., bridge rating) and
under common, landmark and prestigious bridges. Com- add this as a new input parameter to all bridges. In the case
mon bridges are those located in rural areas and those of
low traffic density. Sub-urban bridges present landmarks to 4
A keynote to remember is that due to lack of proper documentations
their region, and urban/prestigious bridges are of historical on incident magnitudes, an accurate and quantitative estimation of the
or national importance (i.e., Golden Gate bridge, CA). size/intensity of a particular event may not be obtained (i.e. media
outlets and public reports do not usually provide the exact magnitude
Eighty-nine bridges were classified under ‘‘urban’’ cate-
of a fire, speed of collision, flooding volume, etc.). This aspect,
gory while the rest of the 110 bridges were evenly split together with others, are further discussed in a subsequent section
under ‘‘rural’’ and ‘‘sub-urban’’ categories. It should be towards the end of this paper.

123
Engineering with Computers

Table 2 Description of magnitude of extreme events

Event type Description of considered events

Fire Resulting from collision of vehicles, fuel tankers, barges, and full-sized fires due to arson or wildfires/lightning
High wind Effects arising from severe storms, tsunamis, level 3–5 hurricanes (i.e., wind in excess of 110 mph), and exceptional
Scour/flood precipitation (exceeding expected annual average)
Earthquake Of magnitude 5.0 and higher
Impact/collision Crashes involving single or multi-vehicles (cars, semi-trucks, tractors, ships and airplanes) either with other vehicles or with
bridge components (i.e., cables, girders, piers)

Table 3 Performance of
DL DT GA GP
selected algorithms (%)
Training Testing Training Testing Training Testing

No/minor 85.5 82.4 62.2 50.0 See Table 4 89.4 85.0

Major 33.8 20.0 54.6 41.6 80.5 79.2
Collapse 84.8 82.6 59.5 68.9 84.4 83.3

where information on such a factor is not accessible, then a predictive models/tools that allow proper assessment of
proper value can be assumed. Overall, the principles and bridges against a variety of extreme loading conditions.
methodology presented in this work will still hold true. Detailed descriptions of these algorithms are provided
herein.
3.4 Damage magnitude
4.1 Deep learning (DL)
In the aftermath of an extreme event, both bridge’s integ-
rity and competence of its surrounding transportation net- The conceptual design of a deep learning (DL) architecture
work can be jeopardized depending on the damage mimics the biological neural network and functionality of
magnitude the bridge experiences. For example, if a bridge the brain and hence comprises a unique topology. An
does not suffer from any damage post an incident, then this optimal topology is characterized by a number of layers
bridge is expected to continue to perform normally. On the wherein each layer serves as a set of parallel nodes. The
other hand, a bridge undergoing significant damage is first layer receives the information required for analysis
expected to be closed for maintenance and repairs. To and hence is referred to as the ‘‘input’’ layer. Once inputs
allow such repairs, traffic would need to be detoured to (representing structural, traffic and event features) are fed
nearby routes. From the point of view of this work, the into the first layer, data flow towards successive layers
‘‘damage magnitude, D’’ can be ‘‘no/minor’’ (that does not (known as hidden layers). By using neurons (processing
require shutting down a bridge), ‘‘major’’ (where a bridge units) in the hidden layers, the received inputs are analyzed
would need to undergo major repairs), and ‘‘collapse’’ through specific weightages (connections). As such, the
(where a bridge partially or fully collapses). Overall, 86 of network can learn and recognize the relevant patterns in
the collected bridges underwent minor/no damage, 91 inputs and map such patterns into meaningful relations
experienced major damage, and 122 collapsed as can be using transformation operations. The process of forward
seen in Fig. 1h. flowing of inputs as a means to train a DL algorithm is
known as ‘‘feed-forwarding’’. The training process con-
tinues for a pre-defined number of iterations and/or
4 Description of biomimetical machine reaching a pre-specified error tolerance between DL pre-
learning algorithms dictions and actual observations (i.e., outcome of a real
bridge incident). The most commonly used optimization
This section summarizes both mathematical and compu- method in DL is called Levenberg–Marquardt, which
tational backgrounds of biomimetical machine learning evaluates the error in terms of mean squared error (MSE).
algorithms used in this study. Four algorithms, namely: In this method, if z is the experimental dataset, then MSE
deep learning (DL), decision tree (DT), genetic algorithms can be calculated using the following equation:
(GA) and genetic programing (GP) were used to develop

123
 Table 4 GA-derived expressions to assess vulnerability of bridges
Case Derived expressionsa R2 MAE

123
Bridges with a span larger than 150 m
5
    
3:08
  
No/ Damage ¼ logistic  MS þ cos NL þ factorialðsinð345T 2 ÞÞ þ atan min sin 353P1:2610 ; 96.6 0.04
A46:9
minor    
sin 4:98 þ 2:23  103 abs 0:158P56:3
A46:9 ; 2:57  103 Þ  1:77:ÞÞ
Major 98.4 0.02
42:2L
Damage ¼ logisticð850þ 7S3þ 3:93MTS þ 30:4  ð0:00375P  1:33  roundð0:00375P  1:33Þ
þ sinhð8  N Þ  exp A46:8  916SÞÞ

Collapse McoshðMS Þ 6:35M 2 logðM Þ6:35M 2 logðSÞ 98.2 0.001
Damage ¼ logistic 25:9 sinð16:9TÞ  S þ 56:10:158P
A46:8 þ S2
N 
þ tan L  15:6  atan(96 þ 101 logðMÞ  101 logðSÞÞÞ
Bridges with a span less between 50 m and 150
m Damage ¼ logisticð33:35M þ 0:026A þ 49ifð0:98S  3:08; 0:95M  3:52; 1:15Þ þ 1:58ifð0:026A  1:06; 53:32; ifð0:038P  3:27; 20; 36:48ÞÞ þ 539:13ifð0:623T  1:77; 99.1 0.002
No/ 1:23L  2:45; 0:026A  1:06Þ þ 6:65LAifð0:62T  1:77; 1:23L  2:45; 0:026A  1:06Þ þ 3:21LP; ifð0:62T  1:77; 1:23L  2:45; 0:026A  1:06Þ þ 0:15Pif ð0:62T  1:77;
minor 1:23L  2:45; 0:026A  1:064Þ þ modð0:95M  3:52; 62:99Þ  69:61  1:38P  0:65Nifð0:026A  1:064; 53:32; ifð0:038P  3:27; 20; 36:48ÞÞ  6:38Pifð0:62T  1:77;
1:23L  2:45; 0:026A  1:06Þ  13:20Aifð0:62T  1:77; 1:23L  2:45; 0:026A  1:064Þ  271:45Lifð0:62T  1:77; 1:23L  2:45; 0:026A  1:064Þ  0:078LPAif
ð0:62T  1:77; 1:23L  2:45; 0:026A  1:06ÞÞ
Major DMJ ¼ logisticð0:038P þ 1:06atan(25:61T  72:93Þ þ 2:95Tless or equalð0:95M  3:52; 0:038P  3:27Þ þ 2:45 sinhð0:62T  1:77Þ sinhð0:038P  3:27Þ 91.0 0.09
2
896:10287:48LS90:95S2
þ 570:97Sþ451:18Lþ45:79LS
A40:81 þ sinð5:40  2:44AÞ þ roundð0:659N  1:58Þ þ asinh(factorial(0:62T  1:77ÞÞ þ atan2ð0:95M  3:52; 0:95M  3:52Þ
3:75  1:23L  0:026Aatan(25:61T  72:93Þ8:41less or equalð0:95M  3:52; 0:038P  3:27Þ  1:23L sinhð0:62T  1:77Þ sinhð0:038P  3:27ÞÞ
Collapse Damage ¼ logisticð463:89 þ 2:71T þ 1:23L þ 0:026A þ 0:0014 expð0:077PÞ þ 1794:79 stepð1:23L  2:45Þ þ 1483:87 cosð2:76 þ 0:95MÞ þ 1097831:90 lessð0:81; 0:95M 93.4 0.02
3:52Þ þ 6:76SP stepð1:23L  2:458Þ þ 326:15M 2 cosð2:76 þ 0:95MÞ158:40S  171:67round(0:65N  1:58Þ  21:24P stepð1:23L  2:45Þ  571:79S stepð1:23L
2:45Þ  1204:95M cosð2:76 þ 0:95MÞ  29:42M 3 cosð2:76 þ 0:95MÞÞ
Bridges with a span less between 25 m and 50 m
3:12

No/ Damage ¼ 146:22T þ 24:33M þ 4:37A þ 0:27P þ 12:25T 3 þ 2:3MAT þ 25:64MT 2 þ 3:54AT 2 þ 0:085TA2 þ 0:019MA2 þ 0:18MAT 3 þ 0:82M4:13 þ 0:006A2 T 3 95.6 0.04
minor þ0:013MA2 T 2 þ 0:19P6:9
T2:13 þ factorialð0:7T  1:63Þ þ sinhðroundð0:03A  1:36ÞÞ þ atan2ðtanhð1:44L  2:65Þ; floorð1:44L  2:65ÞÞ  84:22  tanðroundð0:79  S  3ÞÞ
minðroundð0:46N  1:29Þ; 1:41Þ1:43MA  7AT  47:86MT  0:06A2  78:34T 2  0:55AT 3  4MT 3  0:028MTA2  1:16MAT 2  0:04A2 T 2  0:002MA2 T 3 Þ
Major Damage
0 ¼ logistic 87.2 0.06
7:66Lþ
roundð0:14P5:31Þ þminðtanð0:03A1:36Þ;roundða tan 2ð0:17;0:03A1:36ÞÞÞ
@ if ð0:14P5:31;0:10P3:72;0:31Þ þ 7:66Lminð6:03 roundð1:44L  2:66Þ; factorialð0:29P  10:51ÞÞ if ð0:82M  2:51; if ð1:44L
modðif ðgreaterð0:03A1:36;0:53Þ;0:03A1:36;tanð0:03A1:36ÞÞ;2:01Þ
2:66; 74; 0:79S  3:0Þ; if ð0:46N  1:29; 0:82M  2:51; if ð1:44L  2:66; 74; 0:29P  10:51ÞÞÞ þ 0:38Pminð6:03 roundð1:44L  2:66Þ; factorialð0:29P  10:51ÞÞ
if ð0:82M  2:51; if ð1:44L  2:66; 74; 0:79S  3:0Þ; if ð0:46N  1:2; 0:82M  2:51; if ð1:44L  2:65; 74; 0:29P  10:51ÞÞÞ þ cosð0:115 þ 2:02M Þ þ tanð0:70N  1:94Þ
þminð21:43S þ 0:49S3  27:08  5:65S2 ; 4:29 tanhð3:14S  11:91ÞÞ  12:07  2:07L2  1:94 tanð0:77T  1:63Þ  14:14minð6:03 roundð1:44L  2:65Þ;
factorialð0:29P  10:51ÞÞ if ð0:82M  2:51; if ð1:44L  2:65; 74; 0:79S  3:0Þ; if ð0:46N  1:29; 0:82M  2:51; if ð1:44L  2:66; 74; 0:29P  10:51ÞÞÞ
0:21LPminð6:03 roundð1:44L  2:65Þ; factorialð0:29P  10:51ÞÞ if ð0:82M  2:51; if ð1:44L  2:65; 74; 0:79S  3:0Þ; if ð0:47N  1:29; 0:82M  2:51;
if ð1:44L  2:66; 74; 0:29P  10:51ÞÞÞÞ

Collapse Damage ¼ logistic 0:62S2 þ 0:001 expð2:25M Þ þ 248:62N maxðQÞþ225:37M maxðQÞ688:50maxð1:290:46N;maxQÞ81:38MN maxðQÞ þ 88.1 0.05
A44:14
atan2ð0:76T  1:63; 327358439062562Þ þ maxð2:65  1:44L; 1:44L  2:65Þþatan2ðfactorialð0:03A  1:36Þ; 121:74 sinð4:92 þ 0:03AÞÞ þ cosð0:14Pif
ð5:31  0:14P; 63:39; 0:146163664957489P  5:31Þ  5:31 if ð5:31  0:14P; 63:39; 0:14P  5:31ÞÞ  1:98  4:75SÞ
where, Q ¼ 1:29  0:46N; maxð0:79S  3:0; 0:82M  2:51Þ
Engineering with Computers
Engineering with Computers

1X 1X

materials include: prestressed concrete (1.0), reinforced concrete (2.0), composite members (3.0), steel (4.0) and timber (5.0); arbitrary numeric for extreme events are: fire (1.0), wind (2.0),
Structural systems are assigned an arbitrary numeric value, i.e., cable-stayed (1.0), suspension (2.0), truss/arch (3.0), box girders (4.0), and I-girders (5.0); arbitrary numeric for construction

storm/scour (3.0), earthquake (4.0), and collision/impact (5.0); arbitrary numeric for bridges according to their location/geographic significance are given as: common (1.0), landmark (2.0) and
z z
MAE

0.03

0.06

0.01
MSE ¼ ðei Þ2 ¼ ð m i  pi Þ 2 ; ð1Þ
z i¼0 z i¼0

where z is the total number of datasets, ei is the error for

each input set, mi is the measured output, and pi is the
88.5

86.0

94.2
R2

estimated output (Fig. 2).

It should be noted that a pre-sensitivity analysis showed
how using 31 hidden layers and a validation criterion of
root mean square error (RMSE) led to realizing an optimal
topology of DL with a good balance between accuracy and
Damage ¼ logistic M 3:52 þ 225floorðsinð1:51N ÞÞ þ tanð0:788T Þ þ factorialð0:0589A  3:24Þ þ if ð16 þ 0:0407P2  1:62PÞ\ð0:202P  4:01Þ; 50:3P  998; 48:7Þ

processing time. The outcome of analysis in these layers in

Damage ¼ logisticð87:24floorðsinð17:3N ÞÞ þ expðM Þ þ coshðtanð0:8365T ÞÞ þ if ðless than or equalð16:05 þ 0:04073P2 ; 0:2018P  4:006Þ; 22:17P  440:2; 22Þ

sent to the final (output) layer for visualization.

4.2 Decision tree (DT)

Damage ¼ logisticð6:16 þ 16:4N  1:47S þ 5:86gaussð3:94  M Þ þ 21:3minðceilingð0:0227AN  1:06N Þ; L þ 0:4T Þ þ factorialð0:412P  8:19Þ

A decision tree (DT) is a support tool that is inspired by the

form of full-grown trees. This tool utilizes a tree-like
model comprising decisions and their possible conse-
quences. Such a tree is produced by splitting a dataset into
branch-like arrangement where a DT starts at a root node.
9:218  S if ð0:03189A  1:492; N; 0:2018P  4:006Þ isðorðL  N; N  LÞ; 3:762; 24sgnð0:03189A  1:49ÞÞ  12:81SÞ

Decision trees are favorable in classification problems as

31  Sðif ð0:0319A  1:49; N; 0:202P  4:01Þif ðxorðL  N; N  LÞ; 8:57; S if ð0:0319A  1:49; 12:2; 17ÞÞ  26:3SÞ

they provide a schematic representation of the outcome of

analysis, thus becoming of high value to trace specifics in a
given DT simulation [45]. Depending on the observations
in hand, impurity measures (e.g., Gini) can be used to
þ sinð5:79 þ absð1:35  103 A  6:32  104 ÞÞ  atanhðsinð4:16  4:44N ÞÞ  5:71M  0:35ANÞÞ

process datapoints. For example, for a node t, Gini index

g(t) is defined as [46]:
X
gðtÞ ¼ pðjjtÞpðijtÞ; ð2Þ
j6¼i

where i and j are target field categories.

pðj; tÞ pðjÞNj ðtÞ X
pðj; tÞ ¼ ; pðjtÞ ¼ ; and pðtÞ ¼ pðj; tÞ:
pðtÞ Nj j

ð3Þ
It should be noted that the developed tree had a maxi-
mum depth of 7, with a confidence level = 0.1, minimum
Derived expressionsa


Bridges with a span less than 25 m
Table 4 (continued)

prestigious (3.0)
Collapse
minor
Major
No/
Case

Fig. 2 Representation of a typical DL network topology

a

123
 Engineering with Computers

among others. On the other hand, the terminal (T) com-

prises arguments as well as numerical constants and/or
Square root (√)
variables. Both, functions and terminals are first randomly
generated and then joined to make a model. Hence, a
developed model has a tree-like configuration in which
branches can extend from a function and end in a terminal
Subtraction (-) as shown in Fig. 3. This tree can be converted into a
‘‘Karva-expression’’. (The first position in this expression
denotes the root of the tree and the transformation process
starts from the root and reads through the string layer by
layer.)
x Division (/) Once a set of models is arrived at, these algorithms
evaluate the fitness (accuracy) of each model for repro-
duction. The fitness of a model is defined as a value that
best reflects how far the model’s predicted results are from
that observed in real life. The fittest models are then
2 y selected and further manipulated by a number of bio-in-
spired operations, i.e., reproduction, crossover and muta-
tion [49, 51]. While the reproduction operation assigns a
qffiffiffiffiffiffiffiffiffiffi
Fig. 3 Typical tree representation for Karva-expression: x  2y higher probability of selection to successful models, the
crossover operation on the other hand warrants the
exchange of genetic material code between evolved mod-
leaf size and maximum size for split equals to 2 and 4, els. In the mutation operation, the algorithm randomly
respectively. These parameters are similar to those sug- selects a function (or terminal) from a model to mutate.
gested by a number of researchers/simulation tools [47]. The used values for these operations were 0.2%, 0.1% and
0.04%, respectively [50]. Finally, the fitness for all of the
4.3 Genetic algorithm (GA) and genetic processed models is calculated and is terminated once a
programing (GP) convergence condition is met. A flowchart representing the
process of arriving at a suitable solution is shown in Fig. 4.
These two algorithms are evolutionary techniques that
were introduced by Holland [48], Koza [49] and Ferreira
[50] and utilize supervised learning processes to attempt to 5 Development and validation of bridge
express relations hidden between a number of factors assessment tools
through mimicking the natural selection process and fol-
lowing principles of Darwinian evolution. In these tech- Now that the database is compiled, this database is ready to
niques, predefined strings of expressions as well as be analyzed using the above discussed algorithms. First,
programs strive to arrive at mathematical representations to the database was randomly arranged to eliminate any
express a certain phenomenon. The main difference biasness arising from a particular bridge incident or feature.
between GA and GP lies in their representation of the final Then, the database was split into a model development set
relation between inputs and output(s). While GA creates a (for training and validation purposes * 60%) and a testing
binary string (i.e., equation/expression) that represent the set (* 40%) which was used for evaluating performances
solution, on the other hand, GP develops computer codes; of applied algorithms after the training is completed (i.e.,
expressed in a functional programming language (such as bridge incidents in this set was not used in the training
C??, Matlab, Fortran, etc.) that can be run to solve a stage) [57]. The database was then analyzed using com-
phenomenon. monly available machine learning code environments
In both of these algorithms, a random population of Matlab [52], or GMDH Shell DS [53].
individuals often referred to as ‘‘trees’’ is created to house a The performance of the carried out analysis is listed in
number of possible solutions through structural ordering of Table 3. This table shows the accuracy of each of the
mathematical symbols. Thus, a possible solution in GA/GP applied algorithms. It can be seen that both GA and GP
is a ranked tree consisting of functions and terminals. For algorithms managed to outperform the other algorithms,
instance, a function (F) may contains basic mathematical and achieve a high accuracy for identifying bridges with
operations (?, 9, etc.), power functions (^, log, exp), No/Minor damage. The lowest accuracy in these models
conditional and logic functions ([, B, AND, OR, etc.), was 80.5% and this was for the case of bridges expected to

123
Engineering with Computers

Start of analysis

Create random populaon

Arrive at possible expressrion trees

Evaluate fitness of expression

If not, apply genetic

Terminate analysis (if fitness
operation (mutation,
criteria is met)
crossover, etc.)

Fig. 4 A flowchart of analysis procedure

undergo major damage. GA and GP were followed by the 5.1 Illustrative example
DL algorithm, which also achieved good training accuracy
(except for the case of bridges with propensity to undergo To further illustrate the application of the developed
major damage). However, the performance of DL (as well assessment tools, one example is carried out herein. This
as DT) was comparatively low when it was tested against example covers an actual bridge that was subjected to fire.
bridges that were not part of the training process This incident broke out on March 23, 2003 at the I-95
(* 40–60%).5 A common observation between all cases Howard Avenue Overpass in Bridgeport, CT, when a
shows that all algorithms seem to struggle when analyzing vehicle crashed into a fuel tanker transporting 30,000 L
bridges that are expected to undergo major damage. (equivalent to 30 m3) of gasoline. As a result of this col-
It is worth noting that the GA algorithm did not yield a lision, the gasoline spilled and ignited, starting a fire. The
positive outcome initially. As such, the compiled database fire lasted for 120 min and reached 1100 C. This intense
was further split into 4 mini databases. These databases heat caused the 23-m-long steel girders to buckle and then
were arranged by span such that bridges with a span less triggered the collapse of northbound and southbound seg-
than 25 m, between 25–50 m, 50–150 m and more than ments (each of which was carrying 5 lanes). As such,
150 m were lumped together. After this, the GA algorithm nearby traffic had to be detoured for a number of weeks [1].
successfully managed to arrive at proper expressions with
Step 1: collect related information:
high accuracy. As a result, 12 different expressions were
derived (4 groups, each with the outputs No/Minor, Major Structural characteristics:
and Collapse). These expressions, together with their
Load-bearing systems, S ? I-girders = 5
coefficient of determination (R2) and mean average error
(MAE) are listed in Table 4. For simplicity, all 12 Construction materials, M ? composite (steel and
expressions derived using GA algorithm, along with col- concrete) = 3
lected data points on bridge incidents, can be found in a Span, P ? 23 m
spreadsheet that is accompanying this study. Further, the Age, A ? 10 years
developed computer codes that were arrived at using GP
are listed in the Appendix in Matlab’s programing Traffic features:
language.
Location/geographical significance, L ? Sub-urban
(based on visuals collected from Google maps) = 2
Number of lanes, N ? Lanes = 10

Type of extreme event:

5
It should be noted that description on DT is provided is Table 6 (in
the ‘‘Appendix’’).
Types of extreme events, T ? Fire = 1

123
 Engineering with Computers

Step 2: apply assessment tools developed using literature/media. From this point of view, it is unfortunate
biomimetical machine learning: that such observations are (1) very limited, (2) not widely
Based on DL tool available nor easily accessible, and (3) may lack com-
Predicts the bridge will collapse. pleteness on aspects such as (event intensity, magnitude,
Based on DT tool etc.). Furthermore, the fact that there are few similarities
Predicts the bridge will undergo No/Minor damage. between collected bridge failures impose complexities
Based on GA-derived expressions for bridges with span upon machine learning algorithms (as these algorithms,
less than 25 m: when applied to solve complex phenomena, seem to

No/ Damage ¼ logisticð87:24floorðsinð17:3  10ÞÞþ

minor  ð3Þ þ coshðtanð0:8365
exp   1ÞÞþ  
if less than or equal 16:05 þ 0:04073ð23Þ2 ; 0:2018  23  4:006 ; 22:17  23  440:2; 22  9:218
5  if ð0:03189  10  1:492; 10; 0:2018  23  4:006Þisðorð2  10; 10  2Þ; 3:762; 24sgnð0:03189  10  1:49ÞÞ
12:81  5Þ = 1.0 (the bridge will undergo some degree of damage)

Major Damage ¼ logistic 6:16 þ 16:4  10  1:475 þ 5:86gaussð3:94  3Þ þ 21:3 minðceilingð0:0227  10  10  1:06  10Þ; 2 þ 0:4  1Þþ
factorialð0:412  23  8:19Þ þ sinð5:79 þ absð1:35  103  10  6:32  104 ÞÞatanhðsinð4:16  4:44  10ÞÞ  5:71  3  0:35
10  10Þ = 0.0 (does not belong to this class)
Collapse Damage ¼ logistic(33:52 þ 225floorðsinð1:51  10ÞÞ þ tanð0:788  1Þþ
factorialð0:0589  10  3:24Þ þ if(16 þ 0:0407ð23Þ2  1:62  23Þ\
ð0:202  23  4:01Þ; 50:3  23  998; 48:7Þ  31  5  ðifð0:0319  10  1:49; 10; 0:202  23  4:01Þ if(xor(2  10; 10  2Þ; 8:57; 5
if(0:0319  10  1:49; 12:2; 17ÞÞ  26:3SÞ = 1.0 (the bridge will collapse). A similar outcome can also be arrived at using the GP-based
Matlab codes tabulated in the Appendix

A comparison of the outcomes obtained from that above smoothly perform in settings with sufficient data points
assessment tools shows that three out of these four tools with some level of resemblance) [54]. Fortunately, some of
agree on the fact that this bridge will collapse (under the these challenges can be overcome.
aforementioned fire event). This agrees with documenta- For example, bridge failures can be re-constructed using
tions and observations from the actual incident [1]. Overall, advanced finite element (FE) simulations. Once properly
one is expected to follow recommendations from GA and validated, FE models can be used to examine key charac-
GP algorithms, as these two achieved the highest accuracy teristics and features of failed bridges by means of para-
in training and validation. The outcome of the developed metric studies [4, 55]. These studies can explore the
DL and DT tools needs to be examined with caution, until outcome of a particular event by varying key characteris-
improved tools are realized. tics (i.e., changing the main construction material from
The use of the developed tools could be of importance if timber to concrete, changing the load bearing structural
the performance of this bridge (or any other bridge for that system from cable-stayed to truss, etc.) and then re-ana-
matter) was to be assessed against a similar fire incident lyzing the same bridge under similar conditions to the
(during the design/construction phase, or a few years into event that struck the bridge in the first place. The outcome
service—as a result of an annual inspection or due to of such studies not only adds an additional set of obser-
changes in traffic dynamicality, i.e., increase in commuters vations but also provides a machine learning algorithm
population, directing traffic towards this bridge as a result with completely new data points resembling a slightly
of closure somewhere else, or heavily using this bridge for modified bridge’s response to a pre-defined level of
fuel transportation due to planning for an oil refinery in the intensity. The newly obtained data points can then be used
vicinity of the bridge, etc.). to improve the accuracy of the developed assessment tools.
While this option was not explored herein, this will be
showcased in a future study.
6 Closing remarks Further, the outcome of parametric studies can also
provide designers/authorities with insights to mitigate
As one would expect, the level of accuracy in predictions damage/collapse in bridges [56]. For instance, in open and
obtained from biomimetical machine learning algorithms remote locations, a well-designed light-frame truss bridge
heavily relies on the availability of actual observations, i.e., could collapse if exposed to a hurricane (as can be seen in a
properly documented bridge failures reported in number of cases listed in the database). However, the same

123
Engineering with Computers

bridge may not have collapsed under the same conditions if 7 Conclusions
it is to be made from heavy concrete systems. The use of
similar tools to those developed herein can provide some Based on the information presented in this paper, the fol-
guidance (i.e., rough assessment) into how to avoid failures lowing conclusions can be drawn.
by identifying characteristics/features that if changed, can
• Recent incidents have noted that bridges are becoming
lead to improved performance under extreme events. An
vulnerable to extreme loading events. This vulnerability
example of such a change in an existing bridge may include
arises from various factors such as, poor/lack of
reducing working lanes from 8 to 4 during periods of heavy
adequate upkeep, increasing intensity of natural/man-
wind. This reduction in lanes would reduce the overall
made disasters, etc.
loading on a particular bridge and hence minimize its
• Biomimetical machine learning-based algorithms are
vulnerability (by lowering live-to-dead load ratio which
used to develop assessment tools that can be used to
lowers magnitude of stresses in load bearing structural
evaluate propensity of bridges to various events. These
members and hence free up sectional capacity reserve to
tools account for structural characteristics and traffic
resist the effect of high wind). It can be seen that these
features of bridges and show high potential in accu-
tools can act as ‘‘expert systems’’ to supplement human-
rately identifying vulnerable bridges.
governed decision-making process and may in fact com-
• The performance of the developed tools (as well as
pensate deficiency in engineering experience to some
future tools to be realized through machine learning)
extent.
can be further improved with the availability of
Finally, it goes without saying that a unique feature of
properly documented incidents, together with bridge
machine learning algorithms is that they are flexible
characteristics and traffic features, and intensity of
enough to evolve/improve overtime; once additional data
extreme events).
points are available. Hence, the presented expressions/tools
are expected to undergo a series of improvements and
calibrations before being officially inducted into in prac- Acknowledgements The author would like to acknowledge the rec-
tical applications. An emphasis should be made with regard ommendations provided by Amir Alavi during the GP analysis. A
special thanks goes to Ignacio Paya-Zaforteza and Guillem Peris-
to producing leaner/shorter expressions and those with Sayol for sharing the majority of incidents covering bridge fires. The
more aesthetic appeal. Future tools are also expected to be authors would like to acknowledge the support and help of the editor
able to identify the magnitude of damage in a quantitative and reviewers.
manner, i.e., pinpoint how many load-bearing members are
expected to fail, predict expected duration of shut down
and impact on supply chain operations, etc. (as opposed to Appendix
the qualitative description used in the study).
See Tables 5, 6 and 7.

Table 5 Database
Bridge Year of City/country Location System Material Span Age Lanesa Type
incident (m)

Beaver Dam Bridge 1963 Canada Rural Cable- Reinforced 37 30 2 Storm/

stayed concrete scour
Puente Rı́o Abajo 2017 USA Rural Cable- Reinforced 22 43 2 Storm/
stayed concrete scour
44th Street Bridge 2018 USA Rural Cable- Steel 12 36 1 Wind
stayed
Bell Ford Covered Bridge 1999 USA Rural Cable- Timber 90 114 1 Storm/
stayed scour
Allenville Covered Bridge 1945 USA Rural Cable- Timber 27 80 1 Storm/
stayed scour
Dewey Bridge Grand 4/6/2008 Moab, Rural Suspension Timber 152 93 1 Fire
County UT, USA

123
 Engineering with Computers

Table 5 (continued)
Bridge Year of City/country Location System Material Span Age Lanesa Type
incident (m)

Almö Bridge 1980 Sweden Rural Truss/arch Prestressed 278 20 4 Collision/

concrete impact
Plaka Bridge 2015 Greece Rural Truss/arch Reinforced 61 149 1 Storm/
concrete scour
US50 Cheat River Covered 5/17/1950 West Virginia, USA Rural Truss/arch Composite 103 32 2 Fire
Bridge
Mungo River bridge 7/1/2004 Cameroon Rural Truss/arch Steel 25 43 2 Fire
Big Four 1893 KY, USA Rural Truss/arch Steel 165 1 2 Wind
Utica Swing Bridge 1906 USA Rural Truss/arch Steel 115 1 2 Wind
Viaduc d’Evaux-les-Bains 1884 France Rural Truss/arch Steel 104 1 1 Wind
Island Park Bridge 1952 USA Rural Truss/arch Steel 40 45 1 Wind
Burr Oak Bridge 2004 USA Rural Truss/arch Steel 36 61 1 Wind
Northport Bridge 1949 USA Rural Truss/arch Steel 80 52 1 Storm/
scour
L’Erable Bridge 2005 USA Rural Truss/arch Steel 61 102 1 Storm/
scour
White Wagon Bridge 1972 USA Rural Truss/arch Steel 61 82 1 Storm/
scour
New Hope Through Truss 2018 USA Rural Truss/arch Steel 48 134 1 Storm/
Bridge scour
McGregor Bridge 1936 USA Rural Truss/arch Steel 43 55 2 Storm/
scour
Elkhorn River Bridge 2010 USA Rural Truss/arch Steel 40 92 2 Storm/
scour
Whangaehu River bridge 1953 New Zeeland Rural Truss/arch Steel 40 45 1 Storm/
scour
Newville Iron Bridge 1912 USA Rural Truss/arch Steel 35 37 2 Storm/
scour
Big Bureau Creek Bridge 2013 USA Rural Truss/arch Steel 33 69 1 Storm/
scour
Okoboji Bridge 1926 USA Rural Truss/arch Steel 25 17 1 Storm/
scour
Millville Road Bridge 2017 USA Rural Truss/arch Steel 18 101 2 Storm/
scour
Big Bayou Canot 1993 USA Rural Truss/arch Steel 59 30 1 Collision/
impact
Butler Bridge 1930 USA Rural Truss/arch Steel 53 23 2 Collision/
impact
Jelly’s Ferry Road Bridge 1949 USA Rural Truss/arch Steel 40 67 1 Collision/
impact
Bridge 12 1974 Canada Rural Truss/arch Steel 25 65 2 Collision/
impact
Brown Station Way 2018 USA Rural Truss/arch Steel 23 68 1 Collision/
impact
Turkey Creek Bridge 4/12/2002 Sharon Springs, KS, Rural Truss/arch Timber 92 30 1 Fire
USA
Old Bridgeton Covered 4/29/2005 Bridgeton, IN, USA Rural Truss/arch Timber 81 137 1 Fire
Bridge
The Jeffries Ford Covered 4/2/2002 Bridgeton, IN, USA Rural Truss/arch Timber 62 87 1 Fire
Bridge
Whites Bridge 7/7/2013 Smyrna, MI, USA Rural Truss/arch Timber 36.5 144 1 Fire
Gudgeonville Covered 11/8/2008 Girard, PA, USA Rural Truss/arch Timber 26 140 1 Fire
Bridge
Portageville Railroad 1875 USA Rural Truss/arch Timber 24.3 23 1 Fire
Bridge

123
Engineering with Computers

Table 5 (continued)
Bridge Year of City/country Location System Material Span Age Lanesa Type
incident (m)

Delta Covered Bridge xx/08/ Delta, IA, USA Rural Truss/arch Timber 24 136 1 Fire
2003
Brookside Train Trestle 5/23/2006 Brookside, AL, USA Rural Truss/arch Timber 20 103 1 Fire
160th Street Bridge 6/18/2008 Page County, IA, USA Rural Truss/arch Timber 16.4 83 1 Fire
SR 241 Dry Creek Bridge 8/21/2009 Yakima County, WA, Rural Truss/arch Timber 15.8 109 2 Fire
USA
Dumbarton Rail Bridge 1/2/1998 Palo Alto, CA, USA Rural Truss/arch Timber 13.5 88 1 Fire
Viaduct San Saba 5/19/2013 San Saba, TX, USA Rural Truss/arch Timber 5 77 1 Fire
Kinzua Bridge 2003 USA Rural Truss/arch Timber 25 121 1 Wind
Hoskins Covered Bridge 1963 USA Rural Truss/arch Timber 22 25 1 Wind
Waterworks Covered 1957 USA Rural Truss/arch Timber 66 98 1 Storm/
Bridge scour
Blenheim Bridge 2011 USA Rural Truss/arch Timber 35 79 1 Storm/
scour
128th Avenue Bridge 2005 USA Rural Truss/arch Timber 21 56 1 Collision/
impact
128th Avenue Bridge 2005 USA Rural Truss/arch Timber 21 56 1 Collision/
impact
Highway 13 bridge 2016 USA Rural I-girder Prestressed 40 25 2 Storm/
concrete scour
Carrabassett River Bridge 2011 USA Rural I-girder Reinforced 20 53 2 Storm/
concrete scour
Valdese 7/24/1998 Valdese, NC, USA Rural I-girder Composite 29.5 39 2 Fire
Ranchero Overpass 5/5/2014 Hesperia, CA, USA Rural I-girder Steel 40 0 8 Fire
Puente Rı́o Grande de 2017 USA Rural I-girder Timber 19 39 2 Storm/
Manatı́ scour
Panther Creek Bridge 1938 USA Sub- Cable- Prestressed 41 15 1 Storm/
urban stayed concrete scour
Cần Thơ Bridge 2007 Vietnam Sub- Cable- Composite 90 1 4 Storm/
urban stayed scour
Showa Bridge 1964 Japan Sub- Cable- Steel 24 30 4 Earthquake
urban stayed
Rafiganj Rail Bridge 2002 India Sub- Cable- Steel 30 50 1 Collision/
urban stayed impact
Granville Bridge 1977 Australia Sub- Cable- Steel 28 25 4 Collision/
urban stayed impact
Rio Chiriquı́ 1999 Panama Sub- Suspension Steel 200 72 2 Collision/
urban impact
Hoover Dam Bypass 2006 USA Sub- Truss/arch Prestressed 323 1 2 Wind
urban concrete
River Rega Bridge 1913 Germany Sub- Truss/arch Reinforced 65 30 1 Storm/
urban concrete scour
Haderslis Bridge 1987 Switzerland Sub- Truss/arch Reinforced 25 18 2 Storm/
urban concrete scour
Belle Isle 1905 MI, USA Sub- Truss/arch Steel 56 16 2 Fire
urban
Honeymoon Bridge 1938 USA Sub- Truss/arch Steel 256 40 2 Wind
urban
Omaha Railroad Bridge 1877 USA Sub- Truss/arch Steel 76 4 1 Wind
urban
Wailuku River Bridge 1946 USA Sub- Truss/arch Steel 42 23 1 Wind
urban
Fremont Wagon Bridge 1912 USA Sub- Truss/arch Steel 60 41 2 Storm/
urban scour

123
 Engineering with Computers

Table 5 (continued)
Bridge Year of City/country Location System Material Span Age Lanesa Type
incident (m)

Popo Agie River Bridge 2010 USA Sub- Truss/arch Steel 50 25 2 Storm/
urban scour
I-5 Coalinga Bridge 1995 USA Sub- Truss/arch Steel 30 25 2 Storm/
urban scour
Ely Street Bridge 1891 USA Sub- Truss/arch Steel 27 123 1 Storm/
urban scour
Galveston Bay Wagon 1900 USA Sub- Truss/arch Steel 24 7 2 Storm/
Bridge urban scour
Elm River Bridge 2011 USA Sub- Truss/arch Steel 17 80 1 Storm/
urban scour
Oder bridge 1945 Germany Sub- Truss/arch Steel 370 53 2 Collision/
urban impact
Chesapeake Bay Bridge 1942 USA Sub- Truss/arch Steel 90 16 3 Collision/
urban impact
Skagit River 2013 USA Sub- Truss/arch Steel 85 41 2 Collision/
urban impact
Freeman Bridge 1947 Germany Sub- Truss/arch Steel 75 2 1 Collision/
urban impact
Laasan Bridge 1945 Germany Sub- Truss/arch Steel 50 149 2 Collision/
urban impact
Zhuoshui Fengyu Bridge 11/27/ Chongqing, China Sub- Truss/arch Timber 30 14 1 Fire
2013 urban
Tehama Bridge 1881 USA Sub- Truss/arch Timber 40 10 2 Wind
urban
Bartonsville Covered 2017 USA Sub- Truss/arch Timber 43 29 1 Storm/
Bridge urban scour
I-85 3/30/2017 GA, USA Sub- I-girder Prestressed 30 64 6 Fire
urban concrete
Highway 310 and U.S. 175 10/13/ Dallas, TX, USA Sub- I-girder Prestressed 20 21 4 Fire
2008 urban concrete
Hatchie River Bridge 1989 USA Sub- I-girder Prestressed 31 15 2 Storm/
urban concrete scour
Wu-Shi 1999 China Sub- I-girder Prestressed 35 37 2 Earthquake
urban concrete
I-Chang 1999 China Sub- I-girder Reinforced 24 39 2 Earthquake
urban concrete
Howard Avenue 3/25/2004 Bridgeport, CT, USA Sub- I-girder Composite 23 10 10 Fire
urban
Schoharie Creek Bridge 1987 USA Sub- I-girder Steel 30 32 4 Storm/
urban scour
Don Patricio Causeway 1933 USA Sub- I-girder Timber 56 6 2 Storm/
urban scour
Morandi Bridge 2018 Italy Urban Cable- Prestressed 210 51 4 Storm/
stayed concrete scour
Hanshin Expressway 1995 Japan Urban Cable- Prestressed 36 25 4 Earthquake
stayed concrete
Nienburg Bridge 1825 Germany Urban Cable- Composite 73 1 2 Collision/
stayed impact
Sava Bridge 2009 Croatia Urban Cable- Steel 40 29 6 Storm/
stayed scour
Tacoma Narrows Bridge 1940 USA Urban Suspension Steel 853 1 2 Wind
Lewiston-Queenstown 1864 USA Urban Suspension Steel 318 13 2 Wind
suspension bridge
Silver Bridge 1967 USA Urban Suspension Steel 214 39 2 Wind
Wheeling Bridge 1854 WV, USA Urban Suspension Steel 308 5 2 Storm/
scour

123
Engineering with Computers

Table 5 (continued)
Bridge Year of City/country Location System Material Span Age Lanesa Type
incident (m)

Roche-Bernard 1852 France Urban Suspension Steel 198 13 2 Storm/

scour
Basse-Chaı̂ne Bridge 1850 France Urban Suspension Steel 105 11 2 Storm/
scour
Allen Street Bridge (Old) 1923 USA Urban Suspension Timber 182 17 2 Storm/
scour
Dryburgh Abbey chain 1817 Scotland Urban Suspension Timber 79 1 1 Storm/
scour
Chester Bridge 1944 USA Urban Truss/arch Steel 204 2 2 Wind
Chain suspension br. In 1833 Germany Urban Truss/arch Steel 75 7 2 Wind
Nassau, wooden deck
Main Street Bridge 1913 USA Urban Truss/arch Steel 143 26 2 Storm/
scour
Walnut Street Bridge 1996 USA Urban Truss/arch Steel 73 106 2 Storm/
scour
Melan Bridge 1965 USA Urban Truss/arch Steel 38 67 3 Storm/
scour
Hardy Bridge 1982 USA Urban Truss/arch Steel 30 66 2 Storm/
scour
Carmi Bridge 1916 USA Urban Truss/arch Steel 25 38 2 Storm/
scour
Fassett Street Bridge 1957 USA Urban Truss/arch Steel 88 61 2 Collision/
impact
Calumet River Bridge 1988 USA Urban Truss/arch Steel 71 74 2 Collision/
impact
Drake’s Creek Bridge 1947 USA Urban Truss/arch Steel 53 70 2 Collision/
impact
Münchenstein Bridge 1891 Switzerland Urban Truss/arch Steel 42 16 1 Collision/
impact
Ashtabula River Bridge 1957 USA Urban Truss/arch Steel 20 25 1 Collision/
impact
Pico Rivera 11/3/2005 Pico Rivera, CA, USA Urban Truss/arch Timber 68 15 5 Fire
Panjagutta Flyover 2007 India Urban Box-girder Prestressed 20 1 2 Wind
concrete
Brighton chain pier 1836 England Urban Box-girder Timber 78 13 2 Storm/
scour
California Cypress Freeway 1989 USA Urban I-girder Reinforced 23 38 8 Earthquake
Overpass concrete
MacArthur Maze 4/29/2007 Oakland, CA, USA Urban I-girder Composite 24.5 37 3 Fire
Hazel Park 7/15/2009 Detroit, MI, USA Urban I-girder Composite 24 1 5 Fire
Al-Sheikh Mansour bridge 2/11/2014 El Cairo, Egypt Urban I-girder Composite 20 30 2 Fire
NYS Thruway 10/9/1997 Yonkers, NY, USA Urban I-girder Composite 18.3 9 7 Fire
Second Yibin Bridge 1997 China Urban Truss/arch Reinforced 160 1 2 Wind
concrete
General Rafael Urdaneta 1964 Venezuela Rural Cable- Prestressed 235 2 4 Collision/
Bridge stayed concrete impact
Sidney Lanier Bridge, 1972 USA Rural Cable- Prestressed 76 24 4 Collision/
Brunswick stayed concrete impact
Mezcala 3/17/2007 Guerrero, Mexico Rural Cable- Steel 313 14 4 Fire
stayed
Jose Urquiza 3/29/2010 Entre Rios, Argentina Rural Cable- Steel 220 33 3 Fire
stayed
Cedar Bridge 1963 Madison County, IA Rural Truss/arch Prestressed 20 13 1 Fire
concrete
Garvin Brook Bridge 2017 USA Rural Truss/arch Reinforced 14 111 1 Storm/
concrete scour

123
 Engineering with Computers

Table 5 (continued)
Bridge Year of City/country Location System Material Span Age Lanesa Type
incident (m)

Ankeny Luten Arch 2018 USA Rural Truss/arch Reinforced 12 7 1 Storm/

concrete scour
Norfolk Southern Train 1999 New Brighton, PA, USA Rural Truss/arch Steel 49.5 80 1 Fire
Opatovice 1945 Opatovice, Czech Rural Truss/arch Steel 30 40 2 Fire
Republic
Little Black River 4/6/2008 Ripley County, MI,USA Rural Truss/arch Steel 25 66 1 Fire
Newcastle Bridge 1980 USA Rural Truss/arch Steel 73 90 2 Wind
Hurricane Creek Bridge 2015 USA Rural Truss/arch Steel 36 98 2 Wind
Elrod Bridge 5/17/1950 USA Rural Truss/arch Steel 35 105 1 Wind
Sangamon River Bridge 7/1/2004 USA Rural Truss/arch Steel 60 9 2 Storm/
scour
CRANDIC Cedar River 1893 USA Rural Truss/arch Steel 48 25 1 Storm/
Bridge scour
Huey P. Long Bridge 1906 USA Rural Truss/arch Steel 240 42 4 Collision/
impact
Hamilton Street Bridge 1884 USA Rural Truss/arch Steel 20 104 2 Collision/
impact
Delville Covered Bridge 1952 Wheatfield Township, Rural Truss/arch Timber 53 125 1 Fire
Pennsylvania, USA
Frankenlust township bridge 2004 Frankenlust, MI, USA Rural Truss/arch Timber 5 30 1 Fire
Kinzua Viaduct 1949 USA Rural Truss/arch Timber 626 103 1 Storm/
scour
Wiehltalbrücke 2005 Gummersbach, Germany Rural Box-girder Steel 60 33 8 Fire
Puyallup River 1972 Tacoma, WA, USA Rural I-girder Prestressed 44.5 5 4 Fire
concrete
Norwalk River Bridge 2018 Ridgefield, CT, USA Rural I-girder Prestressed 22 48 4 Fire
concrete
Fridley Bridge 1936 USA Rural I-girder Prestressed 20 25 1 Storm/
concrete scour
Queen Isabel Causeway 2010 USA Rural I-girder Prestressed 24 28 4 Collision/
concrete impact
Bolton Road Bridge 1953 USA Rural I-girder Prestressed 10 25 2 Collision/
concrete impact
I-75 Big slough Canal 1912 North Port, FL, USA Rural I-girder Reinforced 56 30 2 Fire
concrete
Dismal Bridge 2013 Alexandria, Tennessee, Rural I-girder Reinforced 28 30 2 Fire
USA concrete
Rockaway River Bridge 1926 USA Rural I-girder Reinforced 20 25 3 Storm/
concrete scour
Harrisburg (1) 2017 Harrisburg, PA, USA Rural I-girder Composite 44 43 2 Fire
Birmingham 2004 1993 Birmingham, AL, USA Rural I-girder Composite 40 30 3 Fire
Harrisburg (2) 1930 Harrisburg, PA, USA Rural I-girder Composite 28 43 1 Fire
Birmingham 2002 1949 Birmingham, AL, USA Rural I-girder Composite 25 40 4 Fire
I-40 bridge 1974 USA Rural I-girder Composite 176 30 4 Collision/
impact
Tay Bridge 2018 Scotland Rural I-girder Steel 75 2 1 Wind
Quinebaug River Bridge 4/12/2002 USA Rural I-girder Steel 25 50 1 Storm/
scour
Waubonsie Creek Bridge 4/29/2005 USA Rural I-girder Steel 18 91 1 Storm/
scour
Bass Point Creek Trestle 4/2/2002 USA Rural I-girder Timber 243 11 1 Storm/
scour
Sealy Bridge 7/7/2013 USA Sub- Cable- Reinforced 20 25 2 Collision/
urban stayed concrete impact

123
Engineering with Computers

Table 5 (continued)
Bridge Year of City/country Location System Material Span Age Lanesa Type
incident (m)

Benjamin Harrison 8/2008 USA Sub- Cable- Composite 111 11 2 Collision/

Memorial urban stayed impact
Bridge
Charilaos Trikoupis Bridge 1875 Greece Sub- Cable- Steel 560 1 4 Fire
urban stayed
Hidalgo-Reynosa 8/2003 USA Sub- Suspension Steel 137 7 1 Storm/
International Bridge urban scour
Hintze Ribeiro disaster 5/23/2006 Portugal Sub- Truss/arch Composite 48 30 2 Storm/
urban scour
James Street Swing Bridge 6/18/2008 Thunder Bay, Canada Sub- Truss/arch Steel 40 105 2 Fire
urban
Elbe River Bridge 8/21/2009 Germany Sub- Truss/arch Steel 98 30 2 Storm/
urban scour
Fraser Bridge 1/2/1998 USA Sub- Truss/arch Steel 45 38 2 Storm/
urban scour
South Front Street Bridge 5/19/2013 USA Sub- Truss/arch Steel 42 44 1 Storm/
urban scour
US52 Running Slough 2003 USA Sub- Truss/arch Steel 36 33 2 Storm/
Bridge urban scour
San Francisco–Oakland Bay 1963 USA Sub- Truss/arch Steel 23 53 10 Earthquake
Bridge urban
Sunshine Skyway Bridge 1957 USA Sub- Truss/arch Steel 366 9 4 Collision/
urban impact
Eggner’s Ferry Bridge 2011 USA Sub- Truss/arch Steel 170 80 4 Collision/
urban impact
Judge William Seeber 2005 USA Sub- Truss/arch Steel 110 36 4 Collision/
Bridge urban impact
Kattwykbrücke 2005 Germany Sub- Truss/arch Steel 109 18 2 Collision/
urban impact
Second Narrows Bridge in 2016 Canada Sub- Truss/arch Steel 91 2 2 Collision/
Vancouver urban impact
Bridge in Shepherdsville 2011 USA Sub- Truss/arch Steel 46 30 1 Collision/
near Louisville urban impact
Denville 7/24/1998 Denville, NJ, USA Sub- Box-girder Prestressed 27 40 9 Fire
urban concrete
Elysian Park 5/5/2014 Glendale, CA, USA Sub- Box-girder Reinforced 26 49 4 Fire
urban concrete
I-270/i-70 2017 Columbus, Ohio, USA Sub- Box-girder Reinforced 20 59 2 Fire
urban concrete
Ferdan Bridge 1938 Egypt Sub- I-girder Prestressed 79 12 2 Collision/
urban concrete impact
Manchac Bridge 2007 USA Sub- I-girder Prestressed 33 1 4 Collision/
urban concrete impact
Franklin (I) 1964 Franklin, Tennessee, Sub- I-girder Reinforced 40 30 2 Fire
USA urban concrete
Montebello 2002 Montebello, CA, USA Sub- I-girder Reinforced 38 44 6 Fire
urban concrete
Manatee 1977 Tampa, FL, USA Sub- I-girder Reinforced 31 27 4 Fire
urban concrete
Highway M1 1999 Mill Hill, London, UK Sub- I-girder Reinforced 13 56 6 Fire
urban concrete
Franklin (II) 2006 Franklin, Tennessee, Sub- I-girder Composite 40 0 6 Fire
USA urban
Riderwood 1913 Baltimore, MD, USA Sub- I-girder Composite 30 30 7 Fire
urban

123
 Engineering with Computers

Table 5 (continued)
Bridge Year of City/country Location System Material Span Age Lanesa Type
incident (m)

I-285 over GA400 1987 Atlanta, GA, USA Sub- I-girder Composite 30 32 4 Fire
urban
I-30 in Forth Worth 1905 Forth Worth, TX, USA Sub- I-girder Composite 24 54 8 Fire
urban
Tasman Bridge 1938 Australia Sub- I-girder Steel 95 15 5 Collision/
urban impact
Emery Crossing Road 1877 USA Sub- I-girder Steel 18 111 1 Collision/
Overpass urban impact
14th Street bridge 1946 USA Urban Cable- Prestressed 23 40 4 Collision/
stayed concrete impact
Lacey V. Murrow Memorial 1912 USA Urban Cable- Reinforced 50 50 4 Storm/
Bridge stayed concrete scour
Kapellbrücke Bridge 2010 Switzerland Urban Cable- Timber 204 30 1 Fire
stayed
Puente Marcial Candioti 1995 Argentina Urban Suspension Steel 147 55 2 Storm/
Bridge scour
Liberty Bridge 1891 PA, USA Urban Truss/arch Steel 137 89 4 Fire
Webster Street Bridge 1900 USA Urban Truss/arch Steel 107 28 2 Collision/
impact
Severn Railway Bridge 2011 England Urban Truss/arch Steel 100 81 2 Collision/
impact
John Grace Memorial 1945 USA Urban Truss/arch Steel 67 17 2 Collision/
Bridge impact
Stubenrauch Bridge 1942 Germany Urban Truss/arch Steel 56 37 4 Collision/
impact
Patullo Bridge 2013 Surrey, British Urban Truss/arch Timber 18 72 4 Fire
Columbia, Canada
Great Belt Fixed Link 1947 Denmark Urban Box-girder Prestressed 110 9 4 Collision/
concrete impact
U.S. 175 at Bexar Street, 1945 Dallas, TX, USA Urban Box-girder Reinforced 23.5 30 4 Fire
Dallas concrete
Guokeng Town 11/27/ China Urban Box-girder Reinforced 20 30 2 Fire
2013 concrete
Autobahn 57 1881 Dormagen, Germany Urban Box-girder Reinforced 20 30 4 Fire
concrete
Pont du Mathilde 2017 Rouen, France Urban Box-girder Composite 38.5 38 6 Fire
Lazienkovski Bridge 3/30/2017 Warsaw, Poland Urban Box-girder Steel 80 41 6 Fire
St. Petersburg 10/13/ St. Petersburg, FL, USA Urban I-girder Reinforced 33 28 2 Fire
2008 concrete
Jeffersoncity 1989 Jeffersoncity, MO, USA Urban I-girder Reinforced 15 30 2 Fire
concrete
Hood Canal Bridge 1999 USA Urban I-girder Reinforced 183 18 4 Storm/
concrete scour
Chester Creek 1999 Chester Creek, PA, USA Urban I-girder Composite 21 33 3 Fire
Nueces Bay Causeway 3/25/2004 USA Urban I-girder Steel 25 42 2 Storm/
scour
Big Four 1987 Louisville, KY, USA Rural Truss/arch Steel 167 115 1 Fire
Puente sobre el canal de 1933 Palencia, Spain Rural Truss/arch Steel 80 30 4 Fire
Castilla
Stop thirty Road 2018 Nashville, TN, USA Rural Box-girder Reinforced 28 30 2 Fire
concrete
Sweetwater 1995 Sweetwater, TX, USA Rural I-girder Prestressed 57 1 1 Fire
concrete
Bill Williams Bridge 1825 Parker, AZ, USA Rural I-girder Prestressed 23 30 2 Fire
concrete

123
Engineering with Computers

Table 5 (continued)
Bridge Year of City/country Location System Material Span Age Lanesa Type
incident (m)

Washington County 2009 Washington County, OR, Rural I-girder Prestressed 23 134 1 Fire
USA concrete
Rosario 1940 Rosario, Argentina Rural I-girder Reinforced 350 14 4 Fire
concrete
Valdosta 1864 Valdosta, GA, USA Rural I-girder Reinforced 33 30 4 Fire
concrete
Millersville 1967 Millersville, TN, USA Rural I-girder Reinforced 17.5 30 2 Fire
concrete
California Avenue Overpass 1854 Spartanburg, SC, USA Rural I-girder Reinforced 12.3 47 2 Fire
concrete
Elkridge 1852 Elkridge, MD, USA Rural I-girder Composite 48 47 1 Fire
Indianapolis 1850 Indianapolis, IN, USA Rural I-girder Composite 18.5 41 4 Fire
RI37 Expressway 1923 Cranston, RI, USA Rural I-girder Steel 25 37 3 Fire
Grenzwald Bridge 1817 Germany Sub- Cable- Composite 125 4 4 Fire
urban stayed
Xupu Bridge 1944 China Sub- Cable- Steel 590 14 8 Fire
urban stayed
Hedong Cable-Stayed 1833 China Sub- Cable- Steel 360 10 4 Fire
urban stayed
Walt Whitman 1913 New Jersey, USA Sub- Suspension Steel 610 61 6 Fire
urban
Chain suspension bridge 1996 Wales Sub- Suspension Steel 177 1 3 Wind
urban
Ambassador Bridge 1965 Detroit, MI, USA Sub- Truss/arch Steel 560 89 4 Fire
urban
Belle Vernon Bridge 1982 Belle Vernon, PA, USA Sub- Truss/arch Steel 170 66 4 Fire
urban
Pulaski Skyway 1916 Jersey, New Jersey, USA Sub- Truss/arch Steel 168 82 4 Fire
urban
Britannia Tubular Bridge 1957 Wales Sub- Truss/arch Steel 140 120 4 Fire
urban
Chesapeake Bay Bridge 1988 Sandy Point, MD, USA Sub- Truss/arch Steel 90 41 3 Fire
urban
Pasarela F1 1947 Valencia, Spain Sub- Truss/arch Steel 30 5 1 Fire
urban
Sidney Lanier Bridge 1891 USA Sub- Truss/arch Steel 380 27 4 Collision/
urban impact
Eko Bridge 1957 Nigeria Sub- Box-girder Prestressed 62 38 2 Fire
urban concrete
Sanguenay’s Dubuc Bridge 11/3/2005 Saguenay, Montreal, Sub- Box-girder Composite 60 65 2 Fire
Canada urban
Port St. Lucie 2007 Port St. Lucie, FL, USA Sub- I-girder Prestressed 38 51 2 Fire
urban concrete
Belle Isle 1836 Oklahoma, OK, USA Sub- I-girder Prestressed 33.5 26 3 Fire
urban concrete
Burbank 1989 Burbank, CA, USA Sub- I-girder Prestressed 26 50 2 Fire
urban concrete
Motorway bridge nr. 4/29/2007 Austria Sub- I-girder Prestressed 102 30 2 Storm/
Kufstein urban concrete scour
Shi-Nan 7/15/2009 China Sub- I-girder Prestressed 45 2 2 Earthquake
urban concrete
Yen-Feng 2/11/2014 China Sub- I-girder Prestressed 35 15 4 Earthquake
urban concrete
Ling-Wei 10/9/1997 China Sub- I-girder Prestressed 30 12 2 Earthquake
urban concrete
Yang-Ping 1997 China Sub- I-girder Prestressed 13 13 4 Earthquake
urban concrete

123
 Engineering with Computers

Table 5 (continued)
Bridge Year of City/country Location System Material Span Age Lanesa Type
incident (m)

Bartow 1964 Bartow, West Virginia, Sub- I-girder Reinforced 37 30 2 Fire

USA urban concrete
Atterbury Bridge 1972 Johannesburg, South Sub- I-girder Reinforced 33 30 10 Fire
Africa urban concrete
I-65 Franklin 3/17/2007 Franklin, Tennessee, Sub- I-girder Reinforced 32 30 2 Fire
USA urban concrete
Overpass Cypress/Spring 3/29/2010 Spring, TX, USA Sub- I-girder Reinforced 23 14 10 Fire
urban concrete
Bayway ramp 1963 Mobile, Alabama, USA Sub- I-girder Reinforced 11 30 1 Fire
urban concrete
Chang-Geng 2017 China Sub- I-girder Reinforced 4.8 41 2 Earthquake
urban concrete
Coronado Bridge 2018 San Diego, California, Sub- I-girder Composite 200 46 4 Fire
USA urban
Gilmore Bridge 1999 Cambridge, MA,USA Sub- I-girder Composite 42 38 4 Fire
urban
North Little Rock 1945 North Little, AK, USA Sub- I-girder Composite 40 50 3 Fire
urban
I-91 North Meriden 4/6/2008 Meriden, Connecticut. Sub- I-girder Composite 40 45 2 Fire
USA urban
Hagerstown 1980 Hagerstown, Maryland, Sub- I-girder Composite 32 55 2 Fire
USA urban
Robbinsville 2015 Robbinsville, NJ, USA Sub- I-girder Composite 25.5 30 2 Fire
urban
Davison Township 5/17/1950 Davison Township, MI, Sub- I-girder Composite 18 1 4 Fire
USA urban
Mao-Luo-Shi 7/1/2004 China Sub- I-girder Composite 55 0 2 Earthquake
urban
General W.K. Wilson Jr. 1893 Baldwin County, Sub- I-girder Steel 64 24 4 Fire
Bridge Alabama, USA urban
Zolotoy Bridge 1906 Vladivostok, Russia Urban Cable- Steel 737 0 6 Fire
stayed
Zakim Bridge 1884 Boston, MA, USA Urban Cable- Steel 227 11 10 Fire
stayed
Throgs Neck Bridge 1952 New York, NY, USA Urban Suspension Steel 550 48 6 Fire
Brooklyn Bridge 2004 New York, NY, USA Urban Suspension Steel 486 143 6 Fire
Brooklyn Bridge 1949 New York, NY, USA Urban Suspension Steel 486 145 6 Fire
Brooklyn Bridge Jul. 13 2005 New York, NY, USA Urban Suspension Steel 486 146 6 Fire
Manhattan Bridge 1972 New York, NY, USA Urban Suspension Steel 451 100 11 Fire
Benjamin Franklin Bridge 2018 New York, NY, USA Urban Suspension Steel 271.9 72 7 Fire
Akashi Kaikyo Bridge 1936 Japan Urban Suspension Steel 1900 7 4 Earthquake
Stone Arch Bridge 2010 USA Urban Truss/arch Reinforced 197 81 2 Storm/
concrete scour
Queensboro Bridge 1953 New York, NY, USA Urban Truss/arch Steel 360 104 4 Fire
Queensboro Bridge 1912 New York, NY, USA Urban Truss/arch Steel 360 104 4 Fire
Queensboro Bridge 2013 New York, NY, USA Urban Truss/arch Steel 360 104 4 Fire
Queensboro Bridge 1926 New York, NY, USA Urban Truss/arch Steel 360 104 4 Fire
Queensboro Bridge 2017 New York, NY, USA Urban Truss/arch Steel 300 96 4 Fire
Alexandra Bridge 1993 Canada Urban Truss/arch Steel 172 46 4 Fire
Harlem River Bridge 1930 New York, NY, USA Urban Truss/arch Steel 124 110 2 Fire
Tappan Zee Bridge 1949 New York, NY, USA Urban Truss/arch Steel 68 52 7 Fire
Münchenstein Bridge 1974 Switzerland Urban Truss/arch Steel 42 6 1 Storm/
scour
Bay Bridge 2018 San Francisco, CA, USA Urban Box-girder Prestressed 704 82 10 Fire
concrete

123
Engineering with Computers

Table 5 (continued)
Bridge Year of City/country Location System Material Span Age Lanesa Type
incident (m)

Puente del Arenal 4/12/2002 Cordoba, Spain Urban Box-girder Composite 110 20 4 Fire
Puente de Ventas 4/29/2005 Madrid, Spain Urban Box-girder Composite 45 12 8 Fire
Puente de Ventas 4/2/2002 Madrid, Spain Urban Box-girder Composite 45 14 8 Fire
East London 7/7/2013 London, UK Urban Box-girder Steel 16 82 2 Fire
Shuong-Tung 11/8/2008 China Urban I-girder Prestressed 25 37 2 Earthquake
concrete
M4 Sudáfrica 1875 Kwazulu-Natal, South Urban I-girder Reinforced 20 30 2 Fire
Africa concrete
Puente Santa Rosa 08/2003 Lima, Peru Urban I-girder Composite 60 30 6 Fire
Ballard bridge 5/23/2006 Seattle, WA, USA Urban I-girder Composite 33 93 4 Fire
Lime Kiln Overpass 6/18/2008 Louisville, KY, USA Urban I-girder Composite 30 37 2 Fire
Commerce 8/21/2009 Commerce, CA, USA Urban I-girder Composite 27 49 10 Fire
Ethanol Baltimore 1/2/1998 Baltimore, MD, USA Urban I-girder Composite 26 26 2 Fire
Eastex Freeway with North 5/19/2013 Houston, TX, USA Urban I-girder Composite 26 75 2 Fire
Loop
Grand Rapids 2003 Grand Rapids, MI, USA Urban I-girder Composite 23 30 8 Fire
Brucknen Expressway 1963 New York, NY, USA Urban I-girder Composite 21 32 5 Fire
Brooklyn-Queens 1957 New York, NY, USA Urban I-girder Composite 13 46 4 Fire
Expressway 08
Tobin Bridge 2011 USA Urban I-girder Composite 25 23 2 Collision/
impact
Inputs in italics were assumed as they were not verified through multiple references
In some instances, actual names of bridges were not identified and hence the incidents were named according to the nearest known town/area
a
Total number of lanes carried by a bridge

123
 Engineering with Computers

Table 6 Illustration and annotation of developed DT

P > 331.500
| A > 9.500: No/Minor {No/Minor=11, Collapse=0, Major=0}
| A ≤ 9.500
| | L = Sub-urban: Major {No/Minor=0, Collapse=0, Major=2}
| | L = Urban: No/Minor {No/Minor=2, Collapse=0, Major=0}
P ≤ 331.500
| N > 9.500: No/Minor {No/Minor=2, Collapse=0, Major=0}
| N ≤ 9.500
| | A > 135: Collapse {No/Minor=0, Collapse=2, Major=0}
| | A ≤ 135
| | | T = Earthquake
| | | | A > 20
| | | | | N > 3: Collapse {No/Minor=0, Collapse=3, Major=0}
| | | | | N ≤ 3: No/Minor {No/Minor=1, Collapse=1, Major=0}
| | | | A ≤ 20: No/Minor {No/Minor=3, Collapse=0, Major=0}
| | | T = Fire
| | | | S = Box girder
| | | | | P > 27.500: No/Minor {No/Minor=3, Collapse=0, Major=2}
| | | | | P ≤ 27.500: Major {No/Minor=0, Collapse=0, Major=5}
| | | | S = Cable: Major {No/Minor=0, Collapse=0, Major=3}
| | | | S = Girder: No/Minor {No/Minor=20, Collapse=5, Major=11}
| | | | S = Truss/Arch
| | | | | P > 65: No/Minor {No/Minor=7, Collapse=3, Major=0}
| | | | | P ≤ 65: Collapse {No/Minor=0, Collapse=9, Major=3}
| | | T = Impact/Collision
| | | | N > 1.500: Major {No/Minor=0, Collapse=6, Major=17}
| | | | N ≤ 1.500: Collapse {No/Minor=0, Collapse=6, Major=0}
| | | T = Scour/Flood
| | | | P > 16: Collapse {No/Minor=2, Collapse=27, Major=8}
| | | | P ≤ 16: Major {No/Minor=0, Collapse=0, Major=2}
| | | T = Wind
| | | | A > 75.500: Major {No/Minor=0, Collapse=0, Major=2}
| | | | A ≤ 75.500: Collapse {No/Minor=1, Collapse=11, Major=0}

123
Engineering with Computers

Table 7 Matlab codes obtained from GP analysis (these three codes can be run simultaneously and the model with the lowest outcome governs)

Case for No/Minor damage

%-------------------------------------------------------------- function result = gepLT2B(x, y)
---------- if (x < y),
function result = gepModel(d) result = 1;
else
G1C7 = 4.2784570698464; result = 0;
G1C4 = 8.27053834650716; end
G2C3 = -5.76746116122928;
G4C1 = 8.26579149754326; function result = gepGOE2C(x, y)
G4C7 = -8.91926175725578; if (x >= y),
G4C0 = -1.70445875423444; result = (x+y);
G4C3 = 8.10541093173009; else
G4C4 = 2.71000427259133; result = (x-y);
G5C7 = 2.02978606524857; end
G5C1 = -4.64712265502701;
G6C0 = 4.97909482100894; function result = gepGOE2E(x, y)
G7C7 = 1.97526557206946; if (x >= y),
G8C2 = 9.00570696127201; result = (x+y);
G8C0 = -4.46355055544862; else
result = (x*y);
y = 0.0; end

y = function result = gepGOE2G(x, y)

gepNET2G(((gepGOE4B(gepNET2A(d(2),G1C4),G if (x >= y),
1C7,d(1),G1C4)+(d(7)-d(7)- result = (x+y);
d(7)))/2.0),(gepLOE4L(d(6),d(6),d(1),d(4))- else
gepGOE2E(d(7),d(6)))); result = atan(x*y);
y = y + end
gepAMax3(((gepAMin2(d(1),d(3))+min(d(3),d(2))+g
epRamp3(d(3))+gepNET4A(d(4),d(5),d(7),d(1)))/4.0 function result = gepLogi(x)
),((d(1)+G2C3+d(6)+d(7))/4.0),gepAMin2(d(5),d(4)) if (abs(x) > 709.0),
); result = 1.0 / (1.0 + exp(abs(x) / x * 709.0));
y = y + else
gepGOE2E(ceil(gepLT2B(gepRamp1(d(5)),gepNET result = 1.0 / (1.0 + exp(-x));
4J(d(4),d(7),d(5),d(2)))),gepMap5A(gepGOE4K(d(1) end
,d(1),d(5),d(2)),(d(7)*d(7))));
y = y + function result = gepGau3(x, y, z)
gepMap6A(gepLOE3G(gepLT3D(d(2),G4C1,G4C7), result = exp(-((x+y+z)^2));
gepLT4K(d(4),G4C0,G4C3,d(2)),gepGOE4D(d(4),d(
6),d(4),d(7))),gepGau3(gepGOE4D(d(1),d(6),d(6),d( function result = gepRamp1(x)
5)),gepNET4J(d(7),G4C4,d(3),d(4)),d(1))); if (x > 0.0),
y = y + result = x;
gepNET2C(gepAMax3(gepGOE2C(d(6),d(7)),gepL else
OE2C(d(2),d(7)),gepLOE3E(d(6),G5C7,d(1))),abs(ge result = 0.0;
pNET2E(d(6),G5C1))); end
y = y +
gepNET3G((gepGOE3J(d(6),d(3),d(7))*gepCL3A(d( function result = gepRamp3(x)
6),d(2))),gepNET2C(reallog(d(3)),min(G6C0,d(7))),g if (x > 0.0),
epLT4G(d(7),d(3),d(6),d(7))); result = 0.0;
y = y + else
gepLT4E(gepLT2B(gepLOE4L(d(5),d(6),d(3),d(5)), result = -x;
gepGOE4B(d(5),d(7),d(6),d(2))),gepLOE4G(gepLO end
E4G(d(2),d(7),G7C7,d(2)),d(2),G7C7,d(5)),gepStep4
(d(3)),d(1)); function result = gepStep4(x)
y = y + if (x >= 1.0),
(gepGOE2G(gepLT4G(gepMap6B(d(2),d(3),d(7)),se result = 1.0;
ch(G8C0),d(5),d(3)),gepGOE3J(d(5),d(6),d(4)))- elseif (x <= 0.0),
gepLogi(gepGOE2F(d(4),G8C2))); result = 0.0;
else
SLOPE = 0.318032302310422; result = x;
INTERCEPT = -2.30860900366104; end

probabilityOne = 1.0 / (1.0 + exp(-(SLOPE * y + function result = gepCL3A(x, y)

INTERCEPT))); if (x > 0.0 && y < 0.0),
result = probabilityOne; result = 1.0;

123
 Engineering with Computers

Table 7 continued

elseif (x < 0.0 && y > 0.0), if (temp < y),

result = -1.0; temp = y;
else argMax = 1.0;
result = 0.0; end
end if (temp < z),
argMax = 2.0;
function result = gepMap5A(x, y) end
SLACK = 15.0; result = argMax;
if (y < (x - SLACK)),
result = 0.0; function result = gepNET2A(x, y)
elseif (y >= (x - SLACK) && y < (x - SLACK/3.0)), if (x ~= y),
result = 1.0; result = x;
elseif (y >= (x - SLACK/3.0) && y < (x + else
SLACK/3.0)), result = y;
result = 2.0; end
elseif (y >= (x + SLACK/3.0) && y < (x +
SLACK)), function result = gepLOE2C(x, y)
result = 3.0; if (x <= y),
elseif (y >= (x + SLACK)), result = (x+y);
result = 4.0; else
end result = (x-y);
end
function result = gepMap6A(x, y)
SLACK = 10.0; function result = gepNET2C(x, y)
if (y < (x - SLACK)), if (x ~= y),
result = 0.0; result = (x+y);
elseif (y >= (x - SLACK) && y < (x - SLACK/2.0)), else
result = 1.0; result = (x-y);
elseif (y >= (x - SLACK/2.0) && y < x), end
result = 2.0;
elseif (y >= x && y < (x + SLACK/2.0)), function result = gepNET2E(x, y)
result = 3.0; if (x ~= y),
elseif (y >= (x + SLACK/2.0) && y < (x + result = (x+y);
SLACK)), else
result = 4.0; result = (x*y);
elseif (y >= (x + SLACK)), end
result = 5.0;
end function result = gepGOE2F(x, y)
if (x >= y),
function result = gepMap6B(x, y, z) result = (x+y);
minValue = min(x,y); else
maxValue = max(x,y); result = sin(x*y);
midrange = (minValue + maxValue)/2.0; end
midpoint1 = (minValue + midrange)/2.0;
midpoint2 = (midrange + maxValue)/2.0; function result = gepNET2G(x, y)
if (x ~= y),
if (z < minValue), result = (x+y);
result = 0.0; else
elseif (z >= minValue && z < midpoint1), result = atan(x*y);
result = 1.0; end
elseif (z >= midpoint1 && z < midrange),
result = 2.0; function result = gepLT3D(x, y, z)
elseif (z >= midrange && z < midpoint2), if ((x+y) < z),
result = 3.0; result = (x+y);
elseif (z >= midpoint2 && z < maxValue), else
result = 4.0; result = (x-z);
elseif (z >= maxValue), end
result = 5.0;
end function result = gepLOE3E(x, y, z)
if ((x+y) <= z),
function result = gepAMin2(x, y) result = (x+y);
if (x < y), else
result = 0.0; result = (x*z);
else end
result = 1.0;
end function result = gepLOE3G(x, y, z)
if ((x+y) <= z),
function result = gepAMax3(x, y, z) result = (x*y);
temp = x; else
argMax = 0.0; result = (x+z);

123
Engineering with Computers

Table 7 continued

end end

function result = gepNET3G(x, y, z) function result = gepGOE4K(a, b, c, d)

if ((x+y) ~= z), if ((a+b) >= (c+d)),
result = (x*y); result = sin(a*b);
else else
result = (x+z); result = sin(c*d);
end end

function result = gepGOE3J(x, y, z) function result = gepLOE4L(a, b, c, d)

if ((x+y) >= z), if ((a+b) <= (c+d)),
result = (x*y); result = atan(a*b);
else else
result = (x/z); result = atan(c*d);
end end
Case for Major damage
function result = gepNET4A(a, b, c, d) %--------------------------------------------------------------
if (a ~= b), ----------
result = c; function result = gepModel(d)
else
result = d; G1C9 = 5.91296121097446;
end G1C5 = -0.53994232001709;
G1C8 = 6.22432111575671;
function result = gepGOE4B(a, b, c, d) G3C7 = -4.79659413434248;
if ((a+b) >= (c+d)), G3C8 = 5.4368724631489;
result = c; G3C2 = -1.60008545182653;
else G3C0 = 7.06045716727195;
result = d; G3C9 = 1.27842036194952;
end G4C9 = 3.39152195806757;
G4C1 = 6.349330118717;
function result = gepGOE4D(a, b, c, d) G4C4 = 0.440334086519974;
if ((a+b) >= (c+d)), G4C0 = 5.64928128910184;
result = (a+b); G5C6 = 0.264541289155764;
else G6C3 = 9.32139347514267;
result = (c-d); G6C8 = 7.88597979674673;
end G6C1 = -7.0599475081637;
G6C6 = -5.60655537583544;
function result = gepLT4E(a, b, c, d) G7C3 = -1.31146977141636;
if ((a+b) < (c+d)), G7C8 = 4.20568468276009;
result = (a+b); G7C4 = 11.9301339762566;
else G7C0 = 7.90330848109378;
result = (c*d); G7C9 = 0.32624286629841;
end G7C2 = -2.57497756889554;
G7C7 = 4.83446760460219;
function result = gepLT4G(a, b, c, d) G8C1 = -0.992767113254189;
if ((a+b) < (c+d)), G8C8 = 4.82222968230232;
result = (a*b); G8C5 = -5.17723697778253;
else
result = (c+d); y = 0.0;
end
y =
function result = gepLOE4G(a, b, c, d) gepGT4L(gepLT4B(gepLT4L(G1C8,d(6),d(1),d(5)),
if ((a+b) <= (c+d)), gepOR5(d(3),G1C8),(d(5)-
result = (a*b); G1C8),d(1)),gepGOE2C(d(7),d(2)),(d(6)+G1C9),(G1
else C5+G1C5));
result = (c+d); y = y +
end (atan(gepLOE4G(gepGT2F(d(2),d(1)),gepGT4F(d(7)
,d(6),d(2),d(1)),gepLogi4(d(3),d(4),d(6),d(4)),d(4)))+
function result = gepNET4J(a, b, c, d) gep3Rt(reallog(d(7))));
if ((a+b) ~= (c+d)), y = y +
result = (a*b); gepET4L(gepGOE4D(gepET4K(d(4),d(4),d(2),G3C2
else ),gepLT4B(G3C0,G3C9,d(2),d(5)),d(7),d(2)),tanh(d(
result = (c/d); 4)),d(5),gepLT3E(d(4),G3C7,G3C8));
end y = y +
gepMap4B(gepAMax2(d(4),gepLT3I(G4C9,G4C1,G
function result = gepLT4K(a, b, c, d) 4C4)),d(3),gepGOE3J(gepLOE2B(G4C0,d(6)),gepN
if ((a+b) < (c+d)), ET3H(d(7),d(2),d(6)),d(7)));
result = sin(a*b); y = y +
else gepMap4B(gepLOE4D(d(7),(d(2)+d(2)+d(3)),gepA
result = sin(c*d); Max2(d(7),d(3)),gepNET3D(d(6),d(2),d(3))),gepGT2

123
 Engineering with Computers

Table 7 continued

A(d(2),d(6)),(1.0/(G5C6))); end
y = y +
sinh(gepAMax2(gepGOE4D(gepMap4C(d(5),d(1),d( function result = gepMap4C(a, b, c, d)
3),d(7)),gepGOE4E(d(7),d(5),G6C1,d(7)),gepLT4G( % evaluate minValue(a,b,c), maxValue(a,b,c) and
d(7),d(5),d(2),d(2)),gepGT3B(G6C3,G6C6,d(1))),gep midleValue(a,b,c)
ET4B(G6C3,G6C8,G6C1,d(6)))); %
y = y + % evaluate minValue(a,b,c) and argMin(a,b,c)
gepGOE4E(gepGau4(gepAMax2(G7C9,G7C9),gepL minValue = a;
T4G(d(6),d(7),d(6),d(2)),gepLT4G(G7C2,d(7),d(6),G argMin = 0;
7C7),d(6)),gepNET3J(G7C3,G7C3,d(7)),gepLT4J(d( if (minValue > b),
2),d(3),d(2),G7C8),gepLT2B(G7C4,G7C0)); minValue = b;
y = y + argMin = 1;
gep3Rt(asinh(acos(gepLOE2B(gepGOE3F(gepGT2F end
(d(7),G8C5),gepAMax2(d(3),d(6)),G8C1),(1.0/(G8C if (minValue > c),
8)))))); minValue = c;
argMin = 2;
SLOPE = 1.5646840064801; end
INTERCEPT = -9.45964999172027; % evaluate maxValue(a,b,c) and argMax(a,b,c)
maxValue = a;
probabilityOne = 1.0 / (1.0 + exp(-(SLOPE * y + argMax = 0;
INTERCEPT))); if (maxValue < b),
result = probabilityOne; maxValue = b;
argMax = 1;
function result = gep3Rt(x) end
if (x < 0.0), if (maxValue < c),
result = -((-x)^(1.0/3.0)); maxValue = c;
else argMax = 2;
result = x^(1.0/3.0); end
end % evaluate midleValue(a,b,c)
midleValue = c;
function result = gepLT2B(x, y) if (0 ~= argMin && 0 ~= argMax),
if (x < y), midleValue = a;
result = 1; end
else if (1 ~= argMin && 1 ~= argMax),
result = 0; midleValue = b;
end end

function result = gepGOE2C(x, y) if (d < minValue),

if (x >= y), result = 0.0;
result = (x+y); elseif (d >= minValue && d < midleValue),
else result = 1.0;
result = (x-y); elseif (d >= midleValue && d < maxValue),
end result = 2.0;
elseif (d >= maxValue),
function result = gepLogi4(a, b, c, d) result = 3.0;
if (abs(a+b+c+d) > 709.0), end
result = 1.0 / (1.0 + exp(abs(a+b+c+d) / (a+b+c+d)
* 709.0)); function result = gepAMax2(x, y)
else if (x >= y),
result = 1.0 / (1.0 + exp(-(a+b+c+d))); result = 0.0;
end else
result = 1.0;
function result = gepGau4(a, b, c, d) end
result = exp(-((a+b+c+d)^2));
function result = gepOR5(x, y)
function result = gepMap4B(x, y, z) if ((x >= 1) || (y >= 1)),
% evaluate minValue(x,y), maxValue(x,y) and result = 1;
midrange else
minValue = min(x,y); result = 0;
maxValue = max(x,y); end
midrange = (maxValue + minValue)/2.0;
function result = gepGT2A(x, y)
if (z < minValue), if (x > y),
result = 0.0; result = x;
elseif (z >= minValue && z < midrange), else
result = 1.0; result = y;
elseif (z >= midrange && z < maxValue), end
result = 2.0;
elseif (z >= maxValue), function result = gepLOE2B(x, y)
result = 3.0; if (x <= y),

123
Engineering with Computers

Table 7 continued

result = 1; result = c;
else else
result = 0; result = d;
end end

function result = gepGT2F(x, y) function result = gepET4B(a, b, c, d)

if (x > y), if ((a+b) == (c+d)),
result = (x+y); result = c;
else else
result = sin(x*y); result = d;
end end

function result = gepGT3B(x, y, z) function result = gepLOE4D(a, b, c, d)

if ((x+y) > z), if ((a+b) <= (c+d)),
result = (x+y); result = (a+b);
else else
result = z; result = (c-d);
end end

function result = gepNET3D(x, y, z) function result = gepGOE4D(a, b, c, d)

if ((x+y) ~= z), if ((a+b) >= (c+d)),
result = (x+y); result = (a+b);
else else
result = (x-z); result = (c-d);
end end

function result = gepLT3E(x, y, z) function result = gepGOE4E(a, b, c, d)

if ((x+y) < z), if ((a+b) >= (c+d)),
result = (x+y); result = (a+b);
else else
result = (x*z); result = (c*d);
end end

function result = gepGOE3F(x, y, z) function result = gepGT4F(a, b, c, d)

if ((x+y) >= z), if ((a+b) > (c+d)),
result = (x+y); result = (a+b);
else else
result = (x/z); result = (c/d);
end end

function result = gepNET3H(x, y, z) function result = gepLT4G(a, b, c, d)

if ((x+y) ~= z), if ((a+b) < (c+d)),
result = (x*y); result = (a*b);
else else
result = (x-z); result = (c+d);
end end

function result = gepLT3I(x, y, z) function result = gepLOE4G(a, b, c, d)

if ((x+y) < z), if ((a+b) <= (c+d)),
result = (x*y); result = (a*b);
else else
result = (x*z); result = (c+d);
end end

function result = gepGOE3J(x, y, z) function result = gepLT4J(a, b, c, d)

if ((x+y) >= z), if ((a+b) < (c+d)),
result = (x*y); result = (a*b);
else else
result = (x/z); result = (c/d);
end end

function result = gepNET3J(x, y, z) function result = gepET4K(a, b, c, d)

if ((x+y) ~= z), if ((a+b) == (c+d)),
result = (x*y); result = sin(a*b);
else else
result = (x/z); result = sin(c*d);
end end

function result = gepLT4B(a, b, c, d) function result = gepLT4L(a, b, c, d)

if ((a+b) < (c+d)), if ((a+b) < (c+d)),

123
 Engineering with Computers

Table 7 continued

result = atan(a*b); y=y+

else gepGT3L(gepLT3I(gep3Rt(G7C7),(G7C3*d(5)),(d(6)+d(5))),gep5
result = atan(c*d); Rt(gepET4K(d(6),d(5),d(7),d(7))),gepET4G(G7C4,G7C9,G7C2,d(
end 6)));
y=y+
function result = gepGT4L(a, b, c, d) gepGOE4G(gep5Rt(gepNET3B(d(7),G8C5,d(7))),gepLT4A(gepL
if ((a+b) > (c+d)), T3I(G8C9,d(2),G8C1),gepGT2D(G8C0,d(7)),d(1),d(6)),gepAND5
result = atan(a*b); (d(3),G8C2),gepCL3B(G8C6,G8C5));
else
result = atan(c*d); SLOPE = 0.18423802100867;
end INTERCEPT = 0.137558367523412;

function result = gepET4L(a, b, c, d) probabilityOne = 1.0 / (1.0 + exp(-(SLOPE * y + INTERCEPT)));

if ((a+b) == (c+d)), result = probabilityOne;
result = atan(a*b);
else function result = gep3Rt(x)
result = atan(c*d); if (x < 0.0),
end result = -((-x)^(1.0/3.0));
Case for Collapse else
%-------------------------------------------------------------- result = x^(1.0/3.0);
---------- end
function result = gepModel(d)
function result = gepAND1(x, y)
G1C3 = -1.27964110232856; if ((x < 0) && (y < 0)),
G1C1 = -5.47441088960842; result = 1;
G2C6 = 4.52391204379406; else
G2C1 = 4.3595690786462; result = 0;
G2C9 = 1.0908505508591; end
G4C9 = 1.54359874263741;
G4C6 = -8.91987975707266; function result = gep5Rt(x)
G4C8 = 3.24195602435377; if (x < 0.0),
G5C5 = -2.89971282082583; result = -((-x)^(1.0/5.0));
G6C6 = 7.89608478041932; else
G6C3 = -1.0422070986053; result = x^(1.0/5.0);
G6C9 = 0.390761531418805; end
G6C8 = 5.92943527665029;
G7C4 = -0.742515335551012; function result = gepCL3B(x, y)
G7C9 = 3.17300943021943; if (x >= 1.0 && y >= 1.0),
G7C2 = -9.23093356120487; result = 1.0;
G7C7 = -9.10747703482162; elseif (x <= -1.0 && y <= -1.0),
G7C3 = 8.65279671620838; result = -1.0;
G8C2 = -3.67970824304941; else
G8C6 = -8.9838962981048; result = 0.0;
G8C5 = -5.14389477217933; end
G8C9 = -3.27860347300638;
G8C1 = 1.01168858912931; function result = gepCL3C(x, y)
G8C0 = -4.24222540971099; if (x > 0.0 && y > 0.0),
result = 1.0;
y = 0.0; elseif (x < 0.0 && y < 0.0),
result = -1.0;
y= else
gepCL3C(gepMap5A(gepET2E(gepNET3J(G1C3,G1C1,d(2)),d(2) result = 0.0;
),gepGOE3F(d(5),d(7),d(7))),(gepNET3D(d(2),d(2),d(6))*gepLT4 end
D(d(3),d(5),d(6),d(7))));
y=y+ function result = gepMap5A(x, y)
gepGOE3G(gep5Rt(gepGT3I(d(6),G2C6,G2C6)),gep3Rt(gepGOE SLACK = 15.0;
3D(d(1),d(2),G2C1)),gepAMin4(gepGOE3D(d(1),G2C9,d(2)),d(3) if (y < (x - SLACK)),
,d(7),d(7))); result = 0.0;
y=y+ elseif (y >= (x - SLACK) && y < (x - SLACK/3.0)),
gepLT3J(gepNET3C(cosh(d(2)),gepET2F(d(3),d(5)),sinh(d(7))),fl result = 1.0;
oor(gep3Rt(d(6))),gepGOE2F(d(6),d(6))); elseif (y >= (x - SLACK/3.0) && y < (x + SLACK/3.0)),
y=y+ result = 2.0;
floor(gepAND6(gep5Rt(gepNET3J(d(7),d(7),G4C9)),gepGOE3E( elseif (y >= (x + SLACK/3.0) && y < (x + SLACK)),
gepLOE3I(d(1),G4C6,d(1)),sec(d(4)),gepET4K(G4C6,d(2),d(3),G result = 3.0;
4C8)))); elseif (y >= (x + SLACK)),
y = y + (G5C5*d(3)); result = 4.0;
y=y+ end
gepLT3I(gepET2A(gepAND1(G6C3,G6C9),(d(6)*d(3))),gepET4
K(gepET4G(d(2),d(6),d(1),G6C8),gepNET3L(d(3),d(7),d(2)),G6C function result = gepAMin4(a, b, c, d)
6,d(4)),gepET4K(d(5),d(5),d(7),G6C6)); temp = a;

123
Engineering with Computers

Table 7 continued

argMin = 0.0; function result = gepNET3C(x, y, z)

if (temp >= b), if ((x+y) ~= z),
temp = b; result = (x+y);
argMin = 1.0; else
end result = (x+z);
if (temp >= c), end
temp = c;
argMin = 2.0; function result = gepGOE3D(x, y, z)
end if ((x+y) >= z),
if (temp >= d), result = (x+y);
argMin = 3.0; else
end result = (x-z);
result = argMin; end

function result = gepAND5(x, y) function result = gepNET3D(x, y, z)

if ((x >= 1) && (y >= 1)), if ((x+y) ~= z),
result = 1; result = (x+y);
else else
result = 0; result = (x-z);
end end

function result = gepAND6(x, y) function result = gepGOE3E(x, y, z)

if ((x <= 1) && (y <= 1)), if ((x+y) >= z),
result = 1; result = (x+y);
else else
result = 0; result = (x*z);
end end

function result = gepET2A(x, y) function result = gepGOE3F(x, y, z)

if (x == y), if ((x+y) >= z),
result = x; result = (x+y);
else else
result = y; result = (x/z);
end end

function result = gepGT2D(x, y) function result = gepGOE3G(x, y, z)

if (x > y), if ((x+y) >= z),
result = (x*y); result = (x*y);
else else
result = (x/y); result = (x+z);
end end

function result = gepET2E(x, y) function result = gepLT3I(x, y, z)

if (x == y), if ((x+y) < z),
result = (x+y); result = (x*y);
else else
result = (x*y); result = (x*z);
end end

function result = gepGOE2F(x, y) function result = gepGT3I(x, y, z)

if (x >= y), if ((x+y) > z),
result = (x+y); result = (x*y);
else else
result = sin(x*y); result = (x*z);
end end

function result = gepET2F(x, y) function result = gepLOE3I(x, y, z)

if (x == y), if ((x+y) <= z),
result = (x+y); result = (x*y);
else else
result = sin(x*y); result = (x*z);
end end

function result = gepNET3B(x, y, z) function result = gepLT3J(x, y, z)

if ((x+y) ~= z), if ((x+y) < z),
result = (x+y); result = (x*y);
else else
result = z; result = (x/z);
end end

123
 Engineering with Computers

Table 7 continued References

function result = gepNET3J(x, y, z) 1. Garlock M, Paya-Zaforteza I, Kodur V, Gu L (2012) Fire hazard
if ((x+y) ~= z), in bridges: review, assessment and repair strategies. Eng Struct.
result = (x*y);
else
https://doi.org/10.1016/j.engstruct.2011.11.002
result = (x/z); 2. Lewis PMR, Reynolds K (2002) Forensic engineering: a reap-
end praisal of the Tay Bridge disaster. Interdiscip Sci Rev
27:287–298. https://doi.org/10.1179/030801802225005725
function result = gepGT3L(x, y, z) 3. Billah KY, Scanlan RH (1991) Resonance, Tacoma Narrows
if ((x+y) > z),
result = (x+y+z);
bridge failure, and undergraduate physics textbooks. Am J Phys
else 59:118–124. https://doi.org/10.1119/1.16590
result = atan(x*y*z); 4. Kodur VK, Aziz EM, Naser MZ (2017) Strategies for enhancing
end fire performance of steel bridges. Eng Struct. https://doi.org/10.
1016/j.engstruct.2016.10.040
function result = gepNET3L(x, y, z)
if ((x+y) ~= z),
5. Scheer J (2010) Failed bridges: case studies, causes and conse-
result = (x+y+z); quences. https://doi.org/10.1002/9783433600634
else 6. ASCE (2017) ASCE infrastructure report card. https://www.
result = atan(x*y*z); infrastructurereportcard.org/cat-item/bridges/. Accessed 23 May
end 2019
7. Wang L, Yang L, Huang D, Zhang Z, GC.-I.-J. of undefined
function result = gepLT4A(a, b, c, d)
if (a < b), 2008. An impact dynamics analysis on a new crashworthy device
result = c; against ship–bridge collision, Elsevier. (n.d.). https://www.scien
else cedirect.com/science/article/pii/S0734743X07001868. Accessed
result = d; 28 May 2019
end
8. Liu M, Frangopol DM (2004) Optimal bridge maintenance
function result = gepLT4D(a, b, c, d) planning based on probabilistic performance prediction. Eng
if ((a+b) < (c+d)), Struct. https://doi.org/10.1016/j.engstruct.2004.03.003
result = (a+b); 9. Robelin C, S.M.-J. of I. Systems, undefined 2007, History-de-
else pendent bridge deck maintenance and replacement optimization
result = (c-d); with Markov decision processes. Ascelibrary.Org. (n.d.). https://
end
ascelibrary.org/doi/abs/10.1061/(ASCE)1076-0342(2007)13:
function result = gepGOE4G(a, b, c, d) 3(195)?casa_token=NbdVyJnwlEsAAAAA:7HiaRCmIcgvXqp_
if ((a+b) >= (c+d)), FyVFITD2DC7tc7
result = (a*b); F4RasFg0Vt3xCVvQQm26gQ3HoQ3X8yif9qUQeIWnhQR.
else Accessed 21 May 2019
result = (c+d);
end 10. Naser MZ, Kodur VKR (2015) A probabilistic assessment for
classification of bridges against fire hazard. Fire Saf J 76:65–73.
function result = gepET4G(a, b, c, d) https://doi.org/10.1016/j.firesaf.2015.06.001
if ((a+b) == (c+d)), 11. Peris-Sayol G, Paya-Zaforteza I, Balasch-Parisi S, Alós-Moya J
result = (a*b); (2017) Detailed analysis of the causes of bridge fires and their
else
result = (c+d);
associated damage levels. J Perform Constr Facil. https://doi.org/
end 10.1061/(asce)cf.1943-5509.0000977
12. Kodur V, Gu L, Garlock MEM (2010) Review and assessment of
function result = gepET4K(a, b, c, d) fire hazard in bridges. Transp Res Rec J Transp Res Board.
if ((a+b) == (c+d)), https://doi.org/10.3141/2172-03
result = sin(a*b);
else
13. Cooper JD, Fiedland IM, Buckle IG, Nimis RB, McMullin Bobb
result = sin(c*d); N (1994) The Northridge earthquake: progress made, lessons
end learned in seismic-resistant bridge design. Public Roads 58(1)
14. Fujino Y, Yoshida Y (2002) Wind-induced vibration and control
of trans-Tokyo bay crossing bridge. J Struct Eng. https://doi.org/
10.1061/(asce)0733-9445(2002)128:8(1012)
15. AASHTO LRFD bridge design specifications, 8th edition (2017).
https://store.transportation.org/item/collectiondetail/152. Acces-
sed 10 Jun 2019
16. Kodur VKR, Garlock M, Iwankiw N (2012) Structures in fire:
state-of-the-art, research and training needs. Fire Technol
48:825–839. https://doi.org/10.1007/s10694-011-0247-4
17. Aziz E (2015) Response of fire exposed steel bridge girders.
Michigan State University, 2015. https://search.proquest.com/
openview/1f955351b06756dcd200eac6b337d266/1?pq-origsite=
gscholar&cbl=18750&diss=y. Accessed 21 May 2019
18. Kiremidjian A, Moore J, Fan YY, Yazlali O, Basoz N, Williams
M (2007) Seismic risk assessment of transportation network
systems. J Earthq Eng. https://doi.org/10.1080/
13632460701285277

123
Engineering with Computers

19. Kleindorfer PR, Saad GH (2009) Managing disruption risks in Constr Facil 30:04016030. https://doi.org/10.1061/(asce)cf.1943-
supply chains. Prod Oper Manag 14:53–68. https://doi.org/10. 5509.0000884
1111/j.1937-5956.2005.tb00009.x 38. BBC NEWS | South Asia | India train derails, killing 100. (n.d.).
20. Naser MZ (2019) Properties and material models for common http://news.bbc.co.uk/2/hi/south_asia/4387474.stm. Accessed 10
construction materials at elevated temperatures. Constr Build Jun 2019
Mater 10:192–206. https://doi.org/10.1016/j.conbuildmat.2019. 39. B. Åkesson, Understanding Bridge Collapse, CRC Press, 2008
04.182 40. Peris-Sayol G, Payá-Zaforteza I (2017) Bridge fires database.
21. Naser MZ (2019) Properties and material models for modern https://www.researchgate.net/publication/317561066_Bridge_
construction materials at elevated temperatures. Comput Mater Fires_Database. Accessed 3 Mar 2019
Sci 160:16–29. https://doi.org/10.1016/j.commatsci.2018.12.055 41. Fu Z, Ji B, Cheng M, Maeno H (2012) Statistical analysis of the
22. Abdalla JA, Hawileh R (2011) Modeling and simulation of low- causes of bridge collapse in China. In: Forensic Eng. 2012,
cycle fatigue life of steel reinforcing bars using artificial neural American society of civil engineers, Reston, VA, 2012, pp 75–83.
network. J Franklin Inst. https://doi.org/10.1016/j.jfranklin.2010. https://doi.org/10.1061/9780784412640.009
04.005 42. Smith D (1976) Bridge failures. Proc Inst Civ Eng 60:367–382.
23. Naser MZ (2019) AI-based cognitive framework for evaluating https://doi.org/10.1680/iicep.1976.3389
response of concrete structures in extreme conditions. Eng Appl 43. Biezma MV, Schanack F (n.d.) Collapse of steel bridges. https://
Artif Intell 81:437–449. https://doi.org/10.1016/j.engappai.2019. doi.org/10.1061/asce0887-3828200721:5398
03.004 44. Kodur VKR, Naser MZ (2013) Importance factor for design of
24. Kushida M, Miyamoto A, Kinoshita K (1997) Development of bridges against fire hazard. Eng Struct 54:207–220. https://doi.
concrete bridge rating prototype expert system with machine org/10.1016/j.engstruct.2013.03.048
learning. J Comput Civ Eng 11:238–247. https://doi.org/10.1061/ 45. Safavian SR, Landgrebe D (1991) A survey of decision tree
(asce)0887-3801(1997)11:4(238) classifier methodology. IEEE Trans Syst Man Cybern
25. Seitllari A (2014) Traffic flow simulation by neuro-fuzzy 21:660–674. https://doi.org/10.1109/21.97458
approach. In: Second international conference on traffic and 46. Chou J-SS, Tsai C-FF, Pham A-DD, Lu Y-HH (2014) Machine
transport engineering (ICTTE), Belgrade, pp 97–102. https://trid. learning in concrete strength simulations: multi-nation data ana-
trb.org/view/1408239. Accessed 28 Nov 2018 lytics. Constr Build Mater 73:771–780. https://doi.org/10.1016/j.
26. Jahangiri A, Rakha HA (2015) Applying machine learning conbuildmat.2014.09.054
techniques to transportation mode recognition using mobile 47. Che D, Liu Q, Rasheed K, Tao X (2011) Decision tree and
phone sensor data. IEEE Trans Intell Transp Syst 16:2406–2417. ensemble learning algorithms with their applications in bioin-
https://doi.org/10.1109/tits.2015.2405759 formatics. Springer, New York, pp 191–199. https://doi.org/10.
27. Levitt RE, Kartam NA, Kunz JC (2008) Artificial intelligence 1007/978-1-4419-7046-6_19
techniques for generating construction project plans. J Constr Eng 48. Goldberg DE, Holland JH (1988) Genetic algorithms and
Manag. https://doi.org/10.1061/(asce)0733-9364(1988) machine learning. Mach Learn. https://doi.org/10.1023/a:
28. Mohan S (1990) Expert systems applications in construction 1022602019183
management and engineering. J Constr Eng Manag 116:87–99. 49. Koza JR (1992) A genetic approach to finding a controller to back
https://doi.org/10.1061/(asce)0733-9364(1990)116:1(87) up a tractor-trailer truck. In: Proceedings of 1992 American
29. U.S.D. of Transportation, Highway Statistics, 1995–2010 (2010). control conference. IEEE, Chicago
https://www.fhwa.dot.gov/policyinformation/statistics.cfm 50. Ferreira C (2001) Gene expression programming: a new adaptive
30. Harik IE, Shaaban AM, Gesund H, Valli GYS, Wang ST (1990) algorithm for solving problems. Complex syst. 13. https://www.
United States bridge failures, 1951–1988. J Perform Constr Facil semanticscholar.org/paper/Gene-Expression-Programming%3A-
4:272–277. https://doi.org/10.1061/(asce)0887-3828(1990)4: a-New-Adaptive-for-Ferreira/3232b2a24c2584ca8e81cb5bf6
4(272) f55aef34f0aefe. Accessed 16 Mar 2019
31. Wardhana K, Hadipriono FC (2003) Analysis of recent bridge 51. Alavi AH, Gandomi AH, Sahab MG, Gandomi M (2010) Multi
failures in the United States. J Perform Constr Facil 17:144–150. expression programming: a new approach to formulation of soil
https://doi.org/10.1061/(asce)0887-3828(2003)17:3(144) classification. Eng Comput 26:111–118. https://doi.org/10.1007/
32. Cook W, Barr PJ, Halling MW (2015) Bridge failure rate. J Per- s00366-009-0140-7
form Constr Facil 29:04014080. https://doi.org/10.1061/(asce)cf. 52. Searson D (2009) GPTIPS genetic programming & symbolic
1943-5509.0000571 regression for MATLAB user guide. http://citeseerx.ist.psu.edu/
33. Year since Atlanta’s infamous I-85 bridge collapse | WSB-TV viewdoc/summary?doi=10.1.1.177.494. Accessed 22 Jan 2019
(n.d.). https://www.wsbtv.com/news/local/1-year-since-atlantas- 53. GMDH (2019) GMDH shell DS. https://gmdhsoftware.com/.
infamous-i-85-bridge-collapse/723090006. Accessed 10 Jun 2019 Accessed 3 Mar 2019
34. DesRoches R (2006) Hurricane Katrina: performance of trans- 54. Goldberg DE (2006) Genetic algorithms. Pearson Education
portation systems. American Society of Civil Engineers. https:// India. ISBN-13: 978-8177588293
books.google.com/books/about/Hurricane_Katrina.html?id= 55. Kodur VKR, Naser MZ (2019) Designing steel bridges for fire
6wxPOVVYTOUC. Accessed 10 Jun 2019 safety. J Constr Steel Res. https://doi.org/10.1016/j.jcsr.2019.01.
35. Wuttrich R, Wekezer J, Yazdani N, Wilson C (2002) Perfor- 020
mance evaluation of existing bridge fenders for ship impact. 56. Naser MZ (2019) Heuristic machine cognition to predict fire-
J Perform Constr Facil. https://doi.org/10.1061/(asce)0887- induced spalling and fire resistance of concrete structures. Autom
3828(2001)15:1(17) Constr 106:102916. https://doi.org/10.1016/j.autcon.2019.102916
36. Fire at historic bridge near Cold Lake deliberately set: investi- 57. Vdot (n.d.) Bridge inspection definitions. http://www.virginiadot.
gators | CTV News (n.d.). https://edmonton.ctvnews.ca/fire-at- org/info/resources/bridge_defs.pdf. Accessed 28 May 2019
historic-bridge-near-cold-lake-deliberately-set-investigators-1.
876660. Accessed 10 Jun 2019 Publisher’s Note Springer Nature remains neutral with regard to
37. Xu FY, Zhang MJ, Wang L, Zhang JR (2016) Recent highway jurisdictional claims in published maps and institutional affiliations.
bridge collapses in china: review and discussion. J Perform

123