Analyzing Taiwan Bridge Management System for Decision Making

in Bridge Maintenance

A Big Data Approach

Nie-Jia Yau and Yu-Han Chuang

Graduate Institute of Construction Engineering and Management, National Central University,

300 Jhongda Rd., Jhongli District, Taoyuan City, Taiwan

Keywords: Bridge Maintenance, Bridge Management, Big Data, Decision Making.

Abstract: The Taiwan Bridge Management System (TBMS) has been online for 15 years and has an inventory of

33,275 bridges, including all kinds of bridges and culverts in Taiwan. Currently, the number of fields in all

tables in the databases of TBMS is around 6,500 with more than 3 million data records in its databases.

Meanwhile, bridge inspection data and maintenance data are increasing at a speed of 15,000 records

annually. Thus, the TBMS databases are deemed as “Big Data.” There are more than 9,500 bridges that are

over 20 years old with another 7,200 bridge having unknown built years in the TBMS. The bridges in

Taiwan have stepped into the stage where maintenance is crucial and frequently required. Therefore, this

research aims at analysing the database in the TBMS using “Big Data” approach for determining

maintenance strategies for these bridges. This paper describes results of the first year’s research efforts.

Relevant literature in bridge maintenance, prioritization, and life-cycle bridge management were firstly

reviewed. Concepts, theories, techniques, and available software for analysing “Big Data” were also

intensively examined and summarized. In next year, functions will be programmed and applied to the

TBMS databases using appropriate “Big Data” software to obtain useful information in bridge deterioration,

repair methods, and maintenance costs.

1 INTRODUCTION

The Taiwan Bridge Management System (TBMS)

has been online since 2000 (TBMS, 2014). Its

inventory includes 33,275 bridges with 6,524 fields

in all tables, and the total amount of data records is

3,457,274 which increase 15,000 records annually.

Among these fields, there are 475 fields containing

kernel management information of a bridge such as

inventory data, inspection results, and repair records.

Thus, the databases in the TBMS have met the

definition of “Big Data.”

Table 1 shows the amount of bridge components

which are deemed necessary for maintenance

actions. It also shows that the number of seriously

deteriorated components still increase gradually,

even though maintenance activities have been

expedited by responsible agencies for many of such

components.

Due to limited budgets, especially for local

governments, prioritization of bridge maintenance is

Table 1: Amount of bridge components need maintenance

actions.

Road level 2010 2011 2012 2013 2014

City/

County

1,200 463 575 578 857

Freeway/

Highway

17 18 16 24 35

Railways 0 26 28 3 2

Total 1,217 507 619 605 894

always a tough task for the bridge management

agencies, in addition to determining which option is

better between maintaining and rebuilding of the

bridge. Life-cycle cost analysis is a feasible solution

for such problem; however, such technique requires

an appropriate deterioration prediction model which

does not yet implemented in the TBMS.

In order to effectively evaluate cost efficiency of

repairing work and rebuilding of bridge, this

research aims to analyze the TBMS databases to

obtain characteristics of bridge deterioration in

Taiwan that are useful for determining maintenance

Yau N. and Chuang Y..

Analyzing Taiwan Bridge Management System for Decision Making in Bridge Maintenance - A Big Data Approach.

DOI: 10.5220/0005554000730078

In Proceedings of the 10th International Conference on Software Engineering and Applications (ICSOFT-EA-2015), pages 73-78

ISBN: 978-989-758-114-4

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

strategies. This two-year research project has two

stages. For the first year, in addition to literature

review in bridge maintenance, techniques and

available software related to big data are thoroughly

investigated; and application of these techniques and

software to the TBMS databases is planned to be

performed in the second year.

For this research, it is anticipated to obtain

maintenance information such as repairing method,

repairing cost, maintenance period, progressive of

deterioration conditions, and factors that trigger the

repairing actions. Finally, a decision support and

evaluation model for rebuilding of deteriorated

bridges will be established from this research.

2 LITERATURE REVIEW

2.1 Factors caused Bridge

Deterioration in Taiwan

Su (Su, 2003) collected 935 bridge inspection data in

Taichung to analyze the relevancy between bridge

deterioration and its environment by a logistic

regression. The study discovered that the age of

bridge, the distance to sea, and using of I-type

girders are the major factors that caused

deterioration. In addition, Lin (Lin, 2007)

successfully established a service life prediction

model for expansion joint that obtained a 9%

difference between the predicted and the actual

service year, he also discovered that horizontal

acceleration, number of spans and traffic flow are

the most significant factors in determining the

service life of an expansion joint.

2.2 Prioritization of Bridge

Maintenance

For both central and local governments, distribution

of bridge maintenance budgets is always a difficult

task. Chen (Chen, 2007) established a model to

calculate a danger factor (DF) for a bridge by

assigning weights to its major components based on

their deterioration ratings multiplied by a traveller’s

factor determined by level of road that bridge was

on; then the component having the highest value was

normalized to represent the DF of the bridge. The

DF can be used for both prioritization of bridge

maintenance and distribution of maintenance

budgets. This model is currently incorporated by the

TBMS.

2.3 Effectiveness of Maintenance

Budget

Budget spent for bridge maintenance needs to be

effective. Feasible ways to check the effectiveness

of is to investigating results of maintenance within a

time period or under limited budgets. Weng (Weng,

2009) compared the same amount of cost spent

within a time period for fixing or replacing certain

components to find which way is more effective.

Lay (Lay, 2001) developed a maintenance cost

analysis model that allowed the user to input the

amount of budget for a given number of years, and

the model would allocate the budget to the bridges to

achieve the most effective result. Huang (Huang,

2007) proposed a concept of concurrently

maintaining multiple components on a bridge to

reduce the overall time spent for repairing various

components of the bridge.

2.4 Bridge Life-cycle Management

Many researchers have promoted life cycle cost

concept for bridge management for many years.

However, current practice in most bridge

construction bids still not yet considers maintenance

costs. Zhu (Zhu, J. and B. Liu, 2013) established an

optimal model for calculating bridge total life cycle

cost for RC beam bridges, considering travellers’

cost and social cost. Safi (Safi, M., H. Sundquist,

and R. Karoumi, 2014) analyzed the Sweden bridge

management system to find a total maintenance cost

for bridge components. The research results also

showed that the total maintenance cost is 15% to

25% of life cycle cost of a bridge, while different

types of bridges may have more than 50% difference

in construction cost.

2.5 Summary

Several studies in deterioration factors and

maintenance prioritization have obtained certain

valuable results for the bridges in Taiwan. However,

actual maintenance frequency, costs, and methods of

various bridge components could be used to generate

a life cycle cost model which is crucial to obtain a

more effective maintenance strategy. In addition,

decision making between continuing maintenance

actions and rebuilding of a new bridge still not yet

clarified. Thus, answers to these doubts by digging

into the actual inspection results and maintenance

records in the TBMS have become the major

objectives of this research.

ICSOFT-EA2015-10thInternationalConferenceonSoftwareEngineeringandApplications

3 TAIWAN BRIDGE

MANAGEMENT SYSTEM

(TBMS)

Supported by the Institute of transportation, Ministry

of Transportations and Communications, the TBMS

was developed by National Central University in

1999. TBMS is used by all the governmental

agencies which are responsible for bridge

management. There are 9 modules in TBMS, such as

Inventory, Inspection Data, Maintenance Records,

Statistic, Decision Support, etc., as shown in Figure

1. This research focuses on data in three of these

modules; they are Inventory, Inspection Data, and

Maintenance Records modules, as described below.

Figure 1: Major functional modules of the TBMS.

3.1 Inventory Module

There are 33,275 bridges in the inventory module,

among which only 28,000 bridges are still in use or

under maintenance, the rest were destroyed by

natural disasters, closed or demolished due to

serious deterioration. In this module, there are four

tables that describe the basic data of a bridge. Bridge

main inventory table is the top layer of data structure

in this module; below which are abutment, pier, and

span tables. The main inventory table consists of six

kinds of data such as management, geometry,

structure, particular structure, river, and design; the

total number of fields is 147 with roughly 33,000

records since year 2000.

The abutment, pier, and span tables have data

fields describing detailed geometry and design

information with 42, 58, and 39 fields and around

9,700, 24,000, and 90,000 records, respectively.

3.2 Inspection Data Module

The methodology of regular bridge inspections used

by the TBMS is called DER&U (MOTC, 2011). In

this methodology, four indices are used to evaluate

the condition of a bridge component: “D” represents

the degree of deterioration; “E” represents the extent

of the deterioration; “R” represents the

deterioration’s relevancy to bridge safety; and “U”

represents the urgency for repairing the

deterioration. All of these indices are numerically

rated on an integer scale from 0 to 4 to describe the

status of the deterioration, as exhibited in Table 2.

For a concrete bridge, 21 components need to be

inspected, for other types of bridges the number of

components may up to 25.

Table 2: The DER&U evaluation criteria.

0 1 2 3 4

Component

not existing

Good Fair Bad Serious

Unable to

inspect

Less than

10%

10~30% 30~60% Over 60%

Relevancy

uncertain

Minor Limited Major Large

Urgency

uncertain

Routine In 3 years In 1 year Immediately

This inspection data module stores visual inspection

results of all bridges. It has three layers of data

structure; they are main, overall, spans and piers

inspection sheets. These inspection sheets have 21,

69, and 51 fields to record the inspection results and

currently they have around 276,000, 277,000, and

2,000,000 records, respectively. Since current

regulation requires at least inspecting bridge once

per two years, these records increase roughly 15,000

annually. Notably, if deterioration is found during

inspection, it is required to input a suggested

repairing method by the inspector. Thus, at the

bottom of the data structure, the suggested repairing

method is also recorded by 34 fields; it has 521,000

records in the TBMS now.

3.3 Maintenance Records Module

In this module, there are seven tables used to record

a maintenance work such as maintenance contract,

contractor, and detail records of maintenance

activities, etc. Currently, 54,000 maintenance

records are stored in this module. The time for

maintenance, method used, costs and quantity of

repaired components of a bridge are deemed as

crucial information in this research.

4 BIG DATA ANALYSIS

TECHNIQUES

This research reviews current techniques and

software for big data analysis. These techniques are

normally referred to data mining techniques for

finding meaningful information from the big data;

such as supervised learning, unsupervised learning,

AnalyzingTaiwanBridgeManagementSystemforDecisionMakinginBridgeMaintenance-ABigDataApproach

affinity grouping, market basket analysis, clustering,

and description, etc. As for the definition of data

mining, it means data with particular relevancy

could be found by statistics, analysis, machine

learning or expert system (Wikipedia, 2014). These

data mining techniques are described below.

4.1 Big Data Analysis Techniques

4.1.1 Artificial Neural Network

Artificial neural network (ANN) is a mathematic

model simulating neurons connected as a network in

human brains. ANN is a tool in nonlinear statistics

used for investigating the relationships among data.

ANN consists of nodes, existing on a number of

layers; and links, connecting these nodes meanwhile

representing the weights of transmitted messages

between nodes. In sum, ANN is a learning machine

with a black box formed by these nodes and links,

after learning from a huge amount of input data sets

and their corresponding output answers, the trained

ANN can be used in many areas for prediction or

recognition.

4.1.2 Decision Tree

A decision tree consists of a decision diagram and

possible solutions. It can describe process and

procedures including random events and their

associated resources or costs. A decision can be used

as a prediction model in which nodes represent

issues while paths represent possible properties.

Normally, a decision tree has only one single output

as the answer after evaluation. If plural answers are

needed, multiple decision trees should be built

accordingly.

4.1.3 Genetic Algorithms

Genetic algorithm (GA) simulates biological

hereditary and evolution to solve the problems

through coding (Wikipedia, 2014). There are many

arithmetic operators simulate different

characteristics of evolution in various genetic

algorithms. In a genetic algorithm, the solution of a

problem is called individuality representing a

sequence of variable, and the function is called

chromosome. Individualities are generated by

inheritance or mutation, selection or crossover. Each

of the individuality’s suitability is evaluated and

prioritized by its evaluation result; individualities

with higher suitability values are then chosen to

produce a new generation. Optimum solutions can

be found after a number of generations.

4.1.4 Genetic Algorithms

Fuzzy logic was established by Prof L.A. Zadeh in

1965. While classical logic considers that true or

false be described by a binary and discrete variable;

i.e., either 0 or 1, fuzzy logic is able to represent a

linguistic description as partial true truth or false

using a decimal number between 0 and 1. Definition

of utility functions; i.e., determining the relationship

between the linguistic description and its

corresponding decimal number, is crucial in the

application of fuzzy logic.

4.1.5 Regression

Regression is a statistic method to display the

relationship, direction, and strength of multiple

variables. It’s also a model to predict the variation of

variables. There are seven commonly used

regression models such as simple-linear regression,

non-linear regression and multiple regression, etc.

4.2 Big Data Analysis Software

Big data analysis has become a popular issue

recently. After a thorough review of current

available software, 11 kinds of popular software

packages are found. They are Matlab, SAS, R,

Python, Julia, Java, Hadoop and Hive, Scala, Kafka

and Storm, Octave, and GO. Among which, Matlab

and SAS are widely used by academia, while R is

incorporated by many famous portals. Thus, this

research plans to utilize these three kinds of software

packages to perform the big data analysis; their

characteristics are depicted below.

4.2.1 Matlab

Matlab is commercially available software

developed by MathWorks. It can be used for

algorithm generation, data visualization, data mining,

data analysis and calculation. Its latest version is

R204b which allows the user to establish user

surfaces by its programing language or by calling

other programs written by C, C++, JAVA, Python or

FORTRAN.

Matlab also provides an easy-to-use tool box

established based on various techniques such as

generic algorithm, neural networks and ANN,

allowing the use to perform functions such as

optimal analysis, statistics, signal processing, image-

processing, vector analysis, and matrix calculation.

Notably, raw data preparation is crucial for Matlab

since that may affect calculation efficiency.

ICSOFT-EA2015-10thInternationalConferenceonSoftwareEngineeringandApplications

4.2.2 Statistics Analysis System (SAS)

Developed by SAS Institute Inc., SAS has been

commonly used in commercial areas for decades

(Wikipedia, 2014). The initial version of SAS was

written in language C, and now JAVA and C++ are

also included. Its latest version is 9.4, including 10

main modules for data mining, graphics and

presentation, econometrics and time series analysis,

clinical trial analysis, statistics analysis, interactive

matrix language, quality control, and database

transfer, etc.

4.2.3 R

R was developed by Professors Ihaka and

Gentleman at the University of Auckland in New

Zealand. R is written for statistic, drawing, and data

mining. R is capable of performing 25 kinds of

statistic and numerical analysis functions such as

obtaining mean value, standard deviation, plotting of

histogram, and executing regression process. Most

importantly, the source code of R is available freely.

Its famous users include Google, Facebook, Bank of

America, and New York Times.

In addition to the above functions, R can be used

for matrix calculation; its efficient performance can

be comparable to GNU Octave and Matlab.

Thousands of added software tools based on various

analysis techniques for economics and finance have

been established on R by various languages such as

LaTeX, JAVA, C, and FORTRAN.

4.3 Research Process and Anticipated

Results

The next step of this research taken is to formulate

single data records which consists of data of fields

from tables of bridge inventory, span, pier, abutment,

main inspection, detail inspection, suggested

maintenance method, and maintenance record. Data

records have missing data in any field or have logic

inconsistence will be eliminated. These records will

be input to the three software packages; Matlab,

SAS, and R as mentioned above. The anticipated

results will be a maintenance frequency for all the

bridge components, most maintained bridge

components, actual maintenance costs for bridge

components, and the relationship between

deterioration and bridge inventory data. Finally, an

evaluation model will be established for determining

continuation of maintenance or rebuilding of a

bridge based on these findings.

5 CONCLUSIONS

This research collected relevant literature in bridge

maintenance and life cycle costs analysis in Taiwan.

It was found that models for calculating bridge life

cycle costs still not yet established, nor the

effectiveness comparison between maintenance and

rebuilding of a deteriorated bridge. These have

become goals of this research and are intended to be

solved by digging into the big databases of the

TBMS which has already been used for 15 years.

This research also surveyed available software

packages for big data analysis and will soon apply

them to find relevant maintenance information for

decision making in bridge maintenance in Taiwan.

ACKNOWLEDGEMENTS

The authors appreciate the financial support

provided by Ministry of Science and Technology,

under project number MOST 103-2221-E-008-108.

REFERENCES

Institute of transportation, MOTC, 2014. Taiwan Bridge

Management System, < http://tbms.iot.gov.tw/bms2/>

Institute of transportation, MOTC, 2011. Establishment of

Bridge Visual Inspection and Evaluation Manual (Draft).

Huei-Jye Su, 2003. A Correlation Study of the Existing

Bridges for Failure Analysis

－

Case Study of Taichung

County, Taichung.

Jing-Jhan Lin, 2007. Modeling Prediction of Service Lives

of Bridge Expansion Joints, Yunlin.

Jyun-Jhong Chen, 2007. Developing a Maintenance

Decision Support Module for Taiwan Bridge

Management System – An Example for Directorate

General of Highways, Taoyuan.

Kai-hsiang Weng, 2009. Comparison of Economic

Efficiency of Rehabilitation and Replacement in

Bridge Maintenance, Taoyuan.

Zhu, J. and B. Liu, 2013. Performance Life Cost-Based

Maintenance Strategy Optimization for Reinforced

Concrete Girder Bridges, Journal of Bridge

Engineering, 18(2): p. 172-178.

Yuh-chiann Lay, 2001. A Maintenance Strategy Eva-

luation Model for Network Level Bridges, Taoyuan.

Hsun-Yi Huang, 2007. Establishment of Bridge Elements

Concurrent Maintenance Model, Yunlin.

Safi, M., H. Sundquist, and R. Karoumi, 2014. Cost-

Efficient Procurement of Bridge Infrastructures by

Incorporating Life-Cycle Cost Analysis with Bridge

Management Systems, Journal of Bridge Engineering.

0(0): p. 04014083.

AnalyzingTaiwanBridgeManagementSystemforDecisionMakinginBridgeMaintenance-ABigDataApproach

Wikipedia, 2014. datamining, <http://zh.wikipedia.org/

wiki/%E6%95%B0%E6%8D%AE%E6%8C%96%E6

%8E%98#.E6.96.B9.E6.B3.95>

Wikipedia, 2014. artificial neural network <http://

zh.wikipedia.org/wiki/%E4%BA%BA%E5%B7%A5

%E7%A5%9E%E7%BB%8F%E7%BD%91%E7%B

B%9C> (Dec. 15, 2014)

Wikipedia, 2014. artificial neural network <http://zh.

wikipedia.org/wiki/%E9%81%97%E4%BC%A0%E7

%AE%97%E6%B3%95>( Dec. 15, 2014).

Wikipedia, 2014. Statistics Analysis System <http://

zh.wikipedia.org/wiki/%E7%B5%B1%E8%A8%88%

E5%88%86%E6%9E%90%E7%B3%BB%E7%B5%

B1>( Dec. 15, 2014).

ICSOFT-EA2015-10thInternationalConferenceonSoftwareEngineeringandApplications