KSCE Journal of Civil Engineering (2013) 17(7):1551-1561 Construction Management

Copyright ⓒ2013 Korean Society of Civil Engineers

DOI 10.1007/s12205-013-0408-8 pISSN 1226-7988, eISSN 1976-3808
www.springer.com/12205

Infrastructure Asset Management System for Bridge Projects in South Korea

Taehoon Hong*, Myung Jin Chae**, Duyon Kim***, Choongwan Koo****,
Kyo Sun Lee*****, and Kyoung Ho Chin******
Received August 3, 2012/Accepted January 21, 2013

··································································································································································································································

Abstract

While there have been many studies on life cycle cost analysis and preventive maintenance planning, this study proposes an
innovative method of bridge asset management in South Korea. Two different levels of approaches were used in this study. First, in
the level of bridge elements, deterioration modeling and optimized maintenance repair and rehabilitation (MR&R) planning on
bridge assets are proposed, using the bridge historical data of Han River in the city of Seoul. Second, the network level of bridge asset
management is suggested, using historical MR&R cost and budget, overall-condition assessment results, and health index data.
These two levels of approaches were developed into an Internet-based application so that facility managers can use them to review
their past budgets and to plan their future budget based on historical data.
Keywords: bridge management system, asset management, markov chain, life cycle costs, deterioration modeling
··································································································································································································································

1. Introduction method, where the accuracy of the deterioration prediction model

depends on the reliability of the data regarding the deterioration
The purpose of infrastructure maintenance is to prevent casualties process of the infrastructure. Therefore, high-quality data are
and significant loss of Social Overhead Capital (SOC) and to required for this method (Durango-Cohen, 2004). Due to the
offer stability, thereby ensuring a balanced service to the users. In difficulty of obtaining such high-quality data, the researches that
other words, the core of infrastructure maintenance is preventive used the Markov method are limited, such as researches on the
maintenance through the performance of timely maintenance development of a deterioration model of an element (Hong and
actions. The basic procedure for preventive maintenance is, first, Prozzi, 2006; Morcous et al., 2003; Veshosky et al., 1994) and
to predict the deterioration of the elements of the infrastructure, researches focusing only on theoretical methods for the
and then, to establish a maintenance plan for the infrastructure development of the deterioration model (DeStefano and Grivas,
(Hong and Hastak, 2005; Hong and Hastak, 2007). 1998; Morcous et al., 2002; Ortiz-Garcia et al., 2006). In
Many studies have been conducted to develop a method for the addition, for the same reason, it is difficult to predict the detailed
deterioration prediction of infrastructure. Early studies focused maintenance cost via Life Cycle Cost Analysis (LCCA) (Cesare
mainly on deriving the deterioration rate of an element from the et al., 1992).
viewpoint of material research. Cady and Weyers (1984) suggested The Markov Decision Process (MDP) has been applied in the
a procedure for calculating the deterioration rate of a concrete popular software PONTIS and in existing bridge asset management
bridge deck based on the test data of the corrosion process. systems made by the IABMAS Bridge Management Committee
Veshosky et al. (1994) conducted a comparative research focusing and PIARC during the last decade. Generally, the MDP is used
on the deterioration rates of various materials of the upper for modeling deterioration in civil engineering facilities, such as
structure of a bridge, via regression analysis. These researches, bridges, pavements, and waste water systems. The MDP means
however, which were based on the deterioration rate, had the process of solving the problem with the concept of the zone
limitations in terms of the accuracy of their prediction. As an (infinite horizontal time) and a statistical dynamic programming
alternative, Cesare et al. (1992) suggested a deterioration prediction (Winston, 1994). In the MDP, the next state depends on the
framework based on real bridge data, through the Markov chain decisions made in the current state rather than the previous state.

*Associate Professor, Dept. of Architectural Engineering, Yonsei University, Seoul 120-749, Korea (Corresponding Author, E-mail: hong7@yonsei.ac.kr)
**Member, Senior Researcher, Korea Institute of Construction Technology, Ilsan 411-712, Korea (E-mail: chae@kict.re.kr)
***Member, Assistant Professor, School of Construction Engineering, Kyungil University, Kyungsan 712-701, Korea (E-mail: duyonkim@kiu.ac.kr)
****Member, Ph.D. Candidate, Dept. of Architectural Engineering, Yonsei University, Seoul 120-749, Korea (E-mail: cwkoo@yonsei.ac.kr)
*****Director, Construction Engineering Management Division, Korea Institute of Construction Technology, Ilsan 411-712, Korea (E-mail: kslee@kict.re.kr)
******Member, Senior Researcher, Korea Institute of Construction Technology 411-712, Korea (E-mail: khchin@kict.re.kr)

− 1551 −
 Taehoon Hong, Myung Jin Chae, Duyon Kim, Choongwan Koo, Kyo Sun Lee, and Kyoung Ho Chin

The MDP consists of four steps: (i) state space; (ii) decision set; process for bridge maintenance is to estimate the analysis period
(iii) transition probability; and (iv) expected costs. of LCCA based on the performance measurement of bridge
However, various problems need to be solved based on the five elements, to perform LCC analysis of the infrastructure during
perspectives enumerated below (Butt et al., 1987; Cesare et al., the predicted period, and to predict the cost of the required
1992; DeStefano and Grivas, 1998; Jiang and Sinha, 1989; Morcous annual maintenance.
et al., 2002; Morcous et al., 2003; Ortiz-Garcia et al., 2006; The estimation system of the bridge maintenance cost, which
Thompson and Johnson, 2005). is based on the deterioration prediction model, consists of four
1. A bridge as a single structure needs to be divided into sev- steps, as shown in Fig. 1.
eral elements to develop a transition probability matrix. It The first step is to construct a deterioration prediction model
should also be able to estimate the deterioration curve. for each bridge element. Cady and Weyers (1984) and Veshosky
2. The deterioration curve should be developed based on histori- et al. (1994) suggested a deterioration estimation method based
cal data of the bridge element, and its future performance on the manufacturers’ instructions. They suggested that the
needs to be accurately predicted using the curve. Therefore, longevities of each element be determined using the manufacturers’
the element breakdown structure for data collection needs to instructions, and that the MR&R actions be carried out according
be established properly. to the longevities of each element. This method has a limitation,
3. Since the environment affecting the deterioration of an however, with respect to the measurement of the practical and
infrastructure changes with time, the zoning concept needs realistic longevity of a bridge element (Durango-Cohen, 2004).
to be applied to address this problem. The entire life cycle of Therefore, in this study, a method that practically predicts the
the infrastructure can be divided into several periods, timing for the MR&R actions was adopted, which involves the
referred to as “zones.” In each zone, the transition period use of a deterioration prediction model of a bridge element based
and the transition probability are supposed to be homoge- on real deterioration data.
neous. The second step is to predict the performance improvement of
4. A variety of causes for deterioration in bridge elements need each bridge element based on the collected data with regard to
to be identified, and the degree of their impact analyzed. the type, timing, and cost of the bridge MR&R actions. The
Therefore, a monitoring system should be developed to col- optimum MR&R actions for each bridge element are determined
lect data on the cause of deterioration by bridge element. via dynamic programming. The deterioration prediction model
5. It is necessary to understand how deterioration of one ele- of each bridge element is calibrated based on the effects of each
ment impacts the others in the process of deterioration. The MR&R action through time.
transition probability matrix for the MDP should be devel- The third step is to predict the practical maintenance cost using
oped. the cost database of the MR&R actions of each bridge element.
In this step, the bridge maintenance cost estimation system
2. Bridge Asset Management Framework predicts the future state of each bridge element using the
deterioration prediction model, and estimates an appropriate
The recent Asset Management (AM) trend stresses preventive budget based on the cost of the MR&R actions.
or proactive management. In general, preventive management Finally, the system provides the procedure for the estimation of
consists of three steps: condition assessment, deterioration prediction, the Health Index (HI). Developed by the California Department
and intelligent maintenance (Chae and Abraham, 2001). To of Transportation, HI can be useful for a single bridge or a group
develop a preventive management system, the framework of the of bridges, thus providing an excellent performance measure and
asset estimation process for bridge asset maintenance is management tool for bridge AM. As shown in Eq. (1), HI is one
suggested in this study, as shown in Fig. 1. The suggested of the most efficient AM methods because it allows the
framework is for a bridge structure, but it can be applied to all prediction of the future health of the inventory based on various
types of infrastructure. The main purpose of a budget estimation funding levels. Namely, HI provides the damage status of bridge

Fig. 1. Framework of the Asset Estimation Process for Bridge Asset Maintenance

− 1552 − KSCE Journal of Civil Engineering

 Infrastructure Asset Management System for Bridge Projects in South Korea

elements and the corresponding maintenance cost (Shepard, the historic data. Then, the transition probability of the Markov
2005; Shepard and Johnson, 2001). In the case of South Korea, chain was predicted by finding the minimum total absolute value
there are five condition states from “A”, the highest condition from the differences between the developed regression analysis
state, to “E”, the lowest condition state. Accordingly, by using model and the value expected from the Markov chain model.
the following equation, WF i, the weight factors of A, B, C, D, Through regression analysis, the interrelationship between the
and E are calculated as 1.0, 0.75, 0.5, 0.25, and 0 respectively. performance of the elements by bridge type and time can be
defined. As shown in Eq. (2), the nonlinear-optimization method
⎛ ∑ CEV⎞
is used for estimating the transition probability of the Markov
HI = ⎜ ----------------
e
-⎟ × 100% (1)
⎜ TEV ⎟
⎝∑
chain.
e
⎠
te N
Z = min ∑ ∑ y ( t ) – E ( n, p ) (2)
TEV = TEQ × FCe, CEV = FC e × ∑ ( QCSi × WFi ), t = ts n = 1
i
i–1
WFi = 1 – --------------------------------------- where, 0 ≤ Pij ≤ 1 (i, j = 1, 2, ..., m),
State Count – 1
E(n, p) = Expected value of the bridge test level for the
where, transition period of the predicted n level using the
CEV = Current element value Markov chain model
FCe = Failure cost of the element m = Status (test level)
HI = Health index n = Amount of transition time (level)
TEQ = Total element quantity N = Total number of transition periods in each zone
TEV = Total element value t = Year (time)
QCSi = Quantity in the condition state i ts = Starting year (time) of each zone
WFi = Weight factor for the condition state i te = Ending year (time) of each zone
y(t) = Average test level at predicted time t from a
2.1. Step 1: Developing a Deterioration Model for Bridge regression analysis model
Elements by Bridge Type
In Eq. (2), however, it should be noted that the environment
Main bridge structures are categorized into decks, superstructures,
affecting an infrastructure’s deterioration changes with time; this
and substructures. To decide the deterioration level of a bridge, it
goes against the assumption that the transition period is fixed
is reasonable to consider the deterioration analysis results of each
during the life cycle of an infrastructure (Butt et al., 1987). The
element based on the main bridge structure. In the previous
“zoning” concept was applied to address this problem, which
literature reviews related to bridge maintenance, three main
pertains to grouping certain time periods. The entire life cycle
bridge structures were generally used to develop deterioration
of an infrastructure can be divided into several periods, referred
curves. The core structure of the PONTIS program, one of the
to as “zones.” In each zone, the transition period and the
most widely used bridge maintenance systems, is a performance
transition probability are supposed to be regular, and each zone
test module for each type of bridge element. Therefore, it is
is supposed to have a homogeneous Markov chain. The period
appropriate to develop the performance deterioration curves of
of this zone is determined by bridge maintenance experts or
each element (pavement, deck, girder, handrail/curb, expansion
based on the inspection interval of the bridge. For example, in
joint, abutment/bridge pier, drainage, second element, bridge
general, in the case of pavements and bridges, a six-year period
bearing, and foundation).
is considered one zone (Butt et al., 1987; Jiang and Sinha,
To develop a deterioration model for each type of bridge
1989). Therefore, as shown in Eq. (3), E(n, p) in Eq. (2) is
element, the historical data of bridge MR&R actions were
calculated by multiplying the condition vector of stage n by the
extracted from the data obtained regarding the bridge maintenance
condition rating vector S.
system of Han River since 1994. Then, the data were categorized
(n) T ( 0) (n) T
according to the three types of bridge structures: steel box E( n, p) = Q S = Q P S (3)
bridges, Pre-Stressed Concrete (PSC) bridges, and Reinforced- (n)
where, P = Probability matrix after n-step transition
concrete (RC) slab bridges. Each category had the historic data
Q(n) = Condition vector at stage n
of ten elements. The deterioration model was developed using
Q(0) = Initial condition vector at stage 0
the Markov chain model. A nonlinear-optimization model
ST = Transpose of condition rating vector S.
commonly used in the transition probability prediction of
bridges and other facilities (e.g., roads and drainages) was used In Eq. (3), the n-step transition probability matrix, P(n), is
in the development of the deterioration model using the Markov predicted using the nonlinear optimization from Eq. (2). For
chain (Bulusu and Sinha, 1997). A nonlinear-optimization instance, if a five-year “zone” is used, the optimization in the
method consists of two steps: regression analysis and nonlinear first zone begins with ts = 1 and ends with te = 5, and that in the
optimization. A regression analysis model was developed based on second zone begins with ts = 6 and ends with te = 10. Once the

Vol. 17, No. 7 / November 2013 − 1553 −

 Taehoon Hong, Myung Jin Chae, Duyon Kim, Choongwan Koo, Kyo Sun Lee, and Kyoung Ho Chin

transition probability for the first zone is estimated, Q(n) is : Q (7) =Q (6) × P2 = Q (0) × P15 × P22
calculated using Eq. (3). If the five-year zone is applied, P1 will … (5)
represent the transition probability matrix in the first zone, and Fifth transition (total tenth)
the condition vector of each stage will be as follows: : Q (10) =Q (9) × P2 = Q (0) × P15 × P25
Zone 1 The state vectors of the rest of the zones can be estimated in
First transition the same way as that stated above. Using this method and the
: Q (1) =Q (0) × P1 zoning concept, the deterioration model of each bridge element
Second transition was derived from the obtained data. For example, in the case of
: Q (2) =Q (1) × P1 = Q (0) × P12 (4) the deterioration model of abutments and piers for a PSC bridge,
…
Fifth transition
Table 1. Transition Probability of Abutments and Piers for PSC
: Q (5) =Q (4) × P1 = Q (0) × P15 Bridges
The condition vector for the fifth transition will be used as the Transition
initial condition vector in the second zone. The condition vector Year matrix of p1 p2 p3 p4 p5
the zone
in the second zone can be calculated using the transition
1-5 P1 0.7367 1 1 1 1
probability matrix in the second zone (P2). 6-10 P2 0.9008 0.9133 0.7864 0.8159 1
Zone 2 11-15 P3 0.9382 0.8924 0.7839 0.7777 1
First transition (total sixth) 16-20 P4 0.8277 0.7948 0.7171 0.7724 1
: Q (6) =Q (5) × P2 = Q (0) × P15 × P2 21-25 P5 0.4863 0.4924 0.4785 0.5908 1
Second transition (total seventh) 26-30 P6 0 0 0 0 1

Table 2. Cost Data of Each MR&R Action (Ex: A Steel Box Bridge)
Cost (USD)
Division MR&R action Standard Quantity Unit Remark
Unit Cost Total Cost
2
LMC method 40 mm 10,080 m 71.86 724,333.06
Pavement SMA method 80 mm 1,260 m2 36.83 46,411.68
Sidewalk pavement 5 mm 3,240 m2 70.25 227,609.00
Concrete 150 mm 4,050 m2 30.59 123,869.48
South of river
Asphalt 150 mm 900 m2 22.70 20,434.24 Concrete pavement
Pavement for
North of river Asphalt 150 mm 1,552.5 m2 22.70 35,249.06 (15-cm cutting,
water edge
Equip. trans. cost 1 Package 12,568.34 12,568.34 15-cm pavement)
Equip. rent 8 Day 2,094.72 16,757.79
0.2 mm and
3,721.03 m 40.85 151,993.31
Low-viscosity epoxy injection less
Bottom side Over 0.3 mm 11.28 m 72.27 815.18
of slab Section restoration 50 mm 409.46 m2 379.77 155,501.14
Section restoration + rebar
50 mm 159.81 m2 451.26 72,116.02
anticorrosion coating
2
50 mm 209.91 m 379.77 79,717.78
Section restoration 100 mm 2.18 m2 651.59 1,420.47
200 mm 0.19 m2 1,303.18 247.60
Pier
50 mm 15.14 m2 451.26 6,832.09
Section restoration + rebar
100 mm 2.66 m2 723.08 1,923.39
anticorrosion coating
200 mm 1.70 m2 1,446.16 2,458.47
Steel box Bolt change F10T-M20 1,459 EA 20.95 30,562.01
Barge 540 P 12 Month 5,341.54 64,098.54
Towboat 370 Hp 12 Month 4,775.97 57,311.63
Temporary Workbench
Hardware production 1 Package 10,473.62 10,473.62
equipment installation
Cargo crane 1 Month 5,236.81 5,236.81 Workbench moving
Ladder car 25 Day 523.68 13,092.02 Epoxy injection
Cutting under
Total cost 1,861,000.00
USD1,000
Note: Based on the exchange rate in fiscal year 2006 (USD1 = KRW954.78)

− 1554 − KSCE Journal of Civil Engineering

 Infrastructure Asset Management System for Bridge Projects in South Korea

Y(t) of Eq. (2) was predicted using Eq. (6) via regression Therefore, based on the transition probability of Eq. (7), the
analysis, and the transition probability is given in Table 1. The transition probability of when the MR&R actions are performed
transition probability for developing the deterioration model for is derived from statistical dynamic programming. Bridge MR&R
the other elements of a PSC bridge can be found in Tables A1 to actions can be categorized into five kinds: (i) routine, (ii) minor
A10 of Appendix. repair, (iii) major repair, (iv) rehabilitation, and (iv) replacement
2 (Hong and Hastak, 2005).
Y ( t ) = 1.369 × EXP ( 0.053t ), R = 0.363 (6)
The transition probabilities of these MR&R actions are derived
2.2 Step 2: Determining the Optimal Maintenance, Repair, from the judgment of bridge maintenance experts or from the
existing transition matrixes. For example, Eq. (8), the transition
and Rehabilitation (MR&R) Actions for Bridge Ele-
matrix of minor repair maintenance, implies that condition states
ments by Bridge Type
Various types of MR&R actions can be carried out, and their 2, 3, 4, and 5 of the transition probability matrix in Eq. (7) will
change into condition states 1, 2, 3, and 4 after minor repair.
effects vary with time. Therefore, the level of performance
improvement of a bridge element through MR&R actions
p11 1 – p11 0 0 0
depends on and varies with time. To identify the level of
performance improvement of a bridge element through MR&R p11 1 – p11 0 0 0
P= 0 p22 1 – p22 0 0 (8)
actions, the Markov chain transition was adopted in this study.
Towards this end, LCC analysis of the MR&R actions on each 0 0 p33 1 – p33 0
bridge element is required. First, the Cost Breakdown Structure 0 0 0 p44 1 – p44
(CBS) was organized, and the historical cost data of MR&R
actions were collected. Table 2 shows the MR&R action cost In the case of major repair maintenance, as shown in Eq. (9),
condition states 2 and 3 will be enhanced into condition state 1,
data of one steel box bridge of Han River. Using these data,
and condition states 4 and 5 will be enhanced into condition
dynamic programming was conducted to determine the optimal
MR&R action. Dynamic programming is an optimization states 2 and 3, after major repair maintenance.
technology used for finding an optimum proposal to make a
p11 1 – p11 0 0 0
decision. Other optimization methods, such as linear programming,
simultaneously find the best alternative, while dynamic p11 1 – p11 0 0 0
P= p11 1 – p11 0 0 0 (9)
programming disassembles the overall problems into subsets. In
other words, optimization through dynamic programming includes 0 p22 1 – p22 0 0
every subset to find the best alternative. This breakdown procedure 0 0 p33 1 – p33 0
is called “decomposition”, and the disassembled subsets are
called “stages.” Every stage has its own states and decisions. If rehabilitation maintenance is applied as shown in Eq. (10),
If the present stage does not figure out the cost or a state change condition state 2 will be upgraded by one step, and condition
in the next stage, it can be expressed in terms of probability states 3, 4, and 5 will be enhanced by two steps. The deterioration
calculated via probabilistic dynamic programming. This can be rate after rehabilitation action will follow the deterioration pattern
applied to the problem of finding an optimum MR&R action of condition state 1.
among various MR&R actions for each bridge element. In other
words, the method can figure out the maintenance cost in the p11 1 – p11 0 0 0
present stage but not in the next stage. Therefore, the uncertainty of p11 1 – p11 0 0 0
state change is described in a transition probability of the Markov P= p11 1 – p11 0 0 0 (10)
model. The transition probability drawn from step 1 is based on the
0 p11 1 – p11 0 0
assumption that there are no MR&R actions performed (the
naturally deteriorating status). When performing preventive or 0 0 p11 1 – p11 0
corrective MR&R actions, the deterioration transition probability of
the infrastructure changes depending on the performed MR&R In the case of the replacement and remodeling actions, the
transition probability will return to the initial state after
actions. For example, if the transition probability derived from the
replacement maintenance, as shown in Eq. (11) (Cesare et al.,
deterioration model of step 1 is described as Eq. (7), this transition
probability implies that there is no MR&R action. 1992).

p11 1 – p11 0 0 0 1 0 0 0 0
0 p22 1 – p22 0 0 1 0 0 0 0
P= (7) P= 1 0 0 0 0 (11)
0 0 p33 1 – p33 0
1 0 0 0 0
0 0 0 p44 1 – p44
1 0 0 0 0
0 0 0 0 1

Vol. 17, No. 7 / November 2013 − 1555 −

 Taehoon Hong, Myung Jin Chae, Duyon Kim, Choongwan Koo, Kyo Sun Lee, and Kyoung Ho Chin

According to the above four types of transition probabilities, Table 3. Feasible MR&R Actions of the Bearing Elements of a
the effectiveness of each MR&R action will be assessed. PSC Bridge
Towards this end, probabilistic dynamic programming is applied Alternative Method Type
as Eq. (12). A1 Bolt change Minor repair
A2 Painting Major repair
⎧ ⎫
fn ( i ) = min ⎨ Cn ( i, a) + α ∑ p( j i, a, n)fn – 1 ( i ) ⎬ (12) A3 Welding Major repair
⎩ j ⎭ A4 Reinforcement Rehabilitation management
A5 Bearing change Replacement management
where, a = Feasible MR&R actions if the state is “i” in “n”
year during the life cycle
Cn(i, a) = Expected cost generated in “n” year if the state is for a PSC bridge is described as S = {1 (best state), 2, …, 5
“i” and if feasible MR&R action “a” is selected (worst state)}, the decision set, which consists of all the feasible
fn(i) = Minimum expected cost required from “n” year MR&R actions of the bearing element for a PSC bridge, is as
during the life cycle given in Table 3.
α = Interest rate The transition probability of each MR&R action of each bridge
P(j|i, a, n) = The probability of next year’s state will be “j” if element is derived using Eqs. (7) to (11). For instance, if the
the present year’s state is “i” and if feasible transition matrix that there is no MR&R actions is as given in Eq.
MR&R action “a” is selected (13), the transition matrix of a minor repair like A1 (bolt change),
given in Table 3, can be drawn as that in Eq. (14) using Eq. (8).
Using the above probability model, the optimum MR&R
action of a bridge element is decided through the MDP, which
1 0 0 0 0
consists of the following four steps: (i) state space, (ii) decision
0 0.9078 0.0922 0 0
set, (iii) transition probability, and (iv) expected costs.
P= 0 0 0.8211 0.1789 0 (13)
(i) State space (S): Described in S = {1, 2, …, I}, where I is a
state level of elements 0 0 0 0.5004 0.4996
(ii) Decision set: Includes all the feasible MR&R actions for 0 0 0 0 1
bridge maintenance management (For example, in the
case of bridge abutment, the decision set includes the fol-
1 0 0 0 0
lowing alternatives: low-viscosity crack repair, reinforcing-
0 0 0 0 0
rod antifouling, anticorrosive technology, replenishment, etc.)
P= 0 0.9078 0.0922 0 0 (14)
(iii) Transition probability: Estimated in step 1
(iv) Expected costs: The required maintenance cost when the 0 0 0.8211 0.1789 0
state level is “i” and if the feasible MR&R actions are 0 0 0 0.5004 0.4996
chosen
The purpose of the MDP is to find the optimum action from The expected cost is the required maintenance cost when the
among the various MR&R actions. In the MDP, three methods given state level is “i” and if a feasible MR&R action is chosen.
(policy iteration, linear programming, and value iteration) are In the case of the PSC bridge bearing element, the expected cost
used to find the optimum action (Winston, 1994). The value of each MR&R action is as shown in Table 4, where the gray part
iteration method was used in this study because it is effective in means that the MR&R action is not applicable at the given
calculating the minimum discounted maintenance cost and condition state because it is assumed that minor repair is applied
because it is simpler than the other two methods in terms of to condition states 1 to 3.
calculation. For instance, if the state level of the bearing element As mentioned above, to decide the optimum MR&R action

Table 4. Expected Repair Costs for Each State Level and the Results of Value Iteration

⎧ 5
⎫
Condition f1( i ) = min ⎨ C1( i, a ) + α ∑ p ( j|i, a, 1 )f0 ( i ) ⎬
⎩ j=1 ⎭ f1(i) ($) a
State (i)
A1 ($) A2 ($) A3 ($) A4 ($) A5 ($)
1 28.97 106.31 139.31 28.97 A1
2 28.97 106.31 139.31 28.97 A1
3 28.97 106.31 139.31 502.58 28.97 A1
4 106.31 139.31 502.58 2,973.34 106.31 A2
5 502.58 2,973.34 502.58 A4
Note: Based on the exchange rate in fiscal year 2006 (USD1 = KRW954.78)

− 1556 − KSCE Journal of Civil Engineering

 Infrastructure Asset Management System for Bridge Projects in South Korea

Table 5. Optimum MR&R Actions for the Bearing Elements of a to the state level (1 to 5) of the bridge, and MR&R actions
PSC Bridge in Each Condition State indicate the feasible MR&R actions for the bearing elements of
MR&R alternative Method Type the PSC bridge. As described in Table 3, alternative 1 is bolt
A1 Low-viscosity crack repair Minor repair change, alternative 2 is painting, alternative 3 is welding,
A2 Reinforcing-rod antifouling Minor repair alternative 4 is reinforcement action, and alternative 5 is bearing
A3 Anticorrosive technology Minor repair change. “F × S” is the maintenance cost of the optimum MR&R
A4 Deck restoration Major repair action among the feasible MR&R actions (alternatives 1-5) for
A5 Replenishment technology Major repair every year and for every condition state. For instance, in the case
of condition state 2 in the first year of the analysis period, the
optimum MR&R action will be alternative 1 (bolt change), and
using dynamic programming, the value iteration method is the required MR&R cost will be 57.95 $/m. If the optimum
applied for the optimization, as described in Eq. (15). As the MR&R action is alternative 2 (painting) in the case of state level
initial minimum expected cost is zero, f0(i) will be zero in Eq. 4 in the same analysis period, the required MR&R cost will be
(15). The expected cost of a feasible MR&R action “a” in state 135.29 $/m.
level “i” will be the resulting iteration value, as shown in Table 4. The optimum MR&R costs for each bridge type for a 40-year
life cycle were estimated in this study. Table 6 shows only the
⎧ 5
⎫ two-year life cycle within 40 years. The system that was
fn ( i ) = min ⎨ Cn ( i, a) + α ∑ p ( j i, a, n )fn – 1 ( i ) ⎬
⎩ j=1 ⎭ (15) developed in the study was enhanced to more precisely and more
easily predict the annual optimum MR&R costs for a bridge
f0 ( i ) = 0 element given that the user collects more accurate unit cost data
where f0(i) denotes the minimum expected cost in “0” year in the of MR&R actions.
life cycle.
Likewise, through the above procedure, finally, the optimum 2.4 Step 4: Estimating the Annual HI of Different Bridge
MR&R action of each condition state for each bridge element is Types
derived. For example, in the case of the bearing element for a As a last step, HI is estimated, which shows the current
PSC bridge, the optimum MR&R action is as shown in Table 5. performance status for a particular bridge type for the efficient
management of the asset. HI is calculated using Eq. (1). In this
2.3 Step 3: Predicting the Annual Cost of MR&R Actions process, from the viewpoint of the asset manager, CEV is
for Bridge Elements by Bridge Type via LCCA considered the current value of the infrastructure, and TEV is the
Through steps 1 and 2, the annual cost of the MR&R actions value in the best state. TEV and CEV, however, can be different
for each bridge type was derived based on the deterioration from the asset value. As the value of the infrastructure is
model and MR&R actions of each bridge element. Table 6 separately calculated from the manager’s and user’s viewpoints,
shows the annual available optimum maintenance cost for the from the manager’s viewpoint, the estimation methods differ
PSC-bridge-bearing elements. In Table 6, condition state refers depending on the infrastructure type. Therefore, in this research,

Table 6. Prediction Results of the Maintenance Costs of the Bearing Elements of a PSC Bridge
Analysis Condition MR & R actions Optimum
F×S
Period State A1 ($) A2 ($) A3 ($) A4 ($) A5 ($) MR&R action
1 28.97 106.31 139.31 28.97 A1
2 28.97 106.31 139.31 28.97 A1
0 3 28.97 106.31 139.31 502.58 28.97 A1
4 106.31 139.31 502.58 2,973.34 106.31 A2
5 502.58 2,973.34 502.58 A4
1 57.95 135.29 168.28 57.95 A1
2 57.95 135.29 168.28 57.95 A1
1 3 57.95 135.29 168.28 531.55 57.95 A1
4 135.29 168.28 531.55 3,002.31 135.29 A2
5 552.76 3,002.31 552.76 A4
1 86.92 164.26 197.25 86.92 A1
2 86.92 164.26 197.25 86.92 A1
2 3 86.92 164.26 197.25 560.52 86.92 A1
4 164.26 197.25 560.52 3,031.29 164.26 A2
5 581.74 3,031.29 581.74 A4
Note: Based on the exchange rate in fiscal year 2006 (USD1 = KRW954.78)

Vol. 17, No. 7 / November 2013 − 1557 −

 Taehoon Hong, Myung Jin Chae, Duyon Kim, Choongwan Koo, Kyo Sun Lee, and Kyoung Ho Chin

TEV and CEV were considered tools for calculating HI. In other are, and then HI reaches 100%. If the status is bad, HI will have a
words, they are the means to estimate the maintenance costs lower value. For example, if the status level of a bridge element
required to maintain a certain performance level of the infrastructure, is C, HI will be 50%.
irrespective of the type.
The Weight Factor (WF) indicates the weight value for each 3. Development of a Prototype of the Bridge Asset
condition state. In the case of South Korea, there are five Management System
condition states (A, the highest, to E, the lowest). As a level for
HI estimation, A is first, B is second, and the rest follow in the A bridge asset management system prototype was developed
same order. For the calculation of CEV, the weight values of A, based on the server-client, using the bridge asset management
B, C, D, and E are 1.0, 0.75, 0.5, 0.25, and 0, respectively, framework (refer to Fig. 1). The asset management system provides
considering the infrastructure’s depreciation. Therefore, the a future maintenance cost prediction model through the
better the infrastructure status is, the more similar CEV and TEV deterioration model, HI, based on the actual cost data. This

Fig. 2. Overall Data Stream of the System

Fig. 3. Captured Screen of the System: (a) Front Page, (b) Registration of the Basic Bridge Data, (c) Asset Status of the Infrastructure,
(d) Automatic Estimation of the Proposed HI Decision

− 1558 − KSCE Journal of Civil Engineering

 Infrastructure Asset Management System for Bridge Projects in South Korea

system shows a practical application of asset management by actions using dynamic programming, (iii) prediction of the
defining the relationship between the maintenance cost and the annual maintenance cost, and (iv) estimation of the health index.
performance of the infrastructure. In this system, the asset value At the element level of MR&R action planning, the available
is estimated using HI, and then, based on the targeted HI level, deterioration data are very limited. Thus, it is not sufficient to
the system estimates the annual maintenance cost using cost make a realistic forecast for effective bridge management at the
database of MR&R actions for each bridge element. network level. To forecast the long-term financial planning of 20
As shown in Fig. 2, the system is divided into Modules 1 and large bridges at Han River in the city of Seoul, some broad
2. As Module 1 was designed to be applied immediately to the methods were needed, and the health index was used in this
present budget estimation duty, it raises the efficiency of the work. As a result, an Internet-based system was developed using
work process by analyzing the existing budget estimation both element-level maintenance planning and a network-level
process. Moreover, it enables the monitoring of the cost change budget forecast model.
for each facility, with the help of a tool, to collect and analyze the Several highlights in this research are as follows: (i) extensive
history data. Module 2 is the asset management system, which literature reviews were carried out, and the real data from bridges
estimates HI based on the state data, and which automatically located at Han River in Seoul, South Korea were collected, and
calculates the quantity of the MR&R actions required for each then using them, the MDP was developed, (ii) the result of this
element so that an optimum MR&R plan can be formulated study is not only applied to specific area but also expanded to all
based on the MR&R action cost data included in Module 1. If the types of infrastructure. As an Internet-based system was developed
appropriate management level is set by HI, a change in HI will based on the result, the foundation for continuous and systematic
be estimated given that the user selects the quantity as well as management was established, (iii) there are several limitations
calculates the optimum cost and quantity. Therefore, the such as low accuracy of the model due to insufficient data,
proposed system can check not only the maintenance cost history uncertainty of causes affecting a deterioration of the bridge
but also the HI history. elements and its interaction between each other. With regard to
The overall data stream of the system is shown in Fig. 2. The the limitations of this research, authors understood it is necessary
budget estimation of the asset begins with the result of the state to develop monitoring system for inspecting individual elements,
test for each element. The state level for an element is predicted and (iv) to overcome the limitations of regression model due to
using the mentioned deterioration prediction model. Module 1 insufficient data, new approach such as Support Vector Machine
focuses on collecting data in the database and on collecting cost (SVM) needs to be applied to develop the model.
data for an element because it follows the current work process This study, on the other hand, did not cover the level of service,
of the maintenance budget estimation duty. Using Module 1, which is an important aspect of infrastructure asset management.
users can calculate the budget following the current work To define the service levels, an in-depth review of the user and
process. In the case of Module 2, the present HI is automatically community needs should be performed. The performance
calculated according to the state test level. Module 2 has not only measurements should also be defined before the levels of service
a basic maintenance rule based on the HI criteria but also an could be defined. In this study, the focus was on element-level
immediate-repair-or-replacement rule when the state level of an bridge management and on a primitive form of asset
element is D or E under the Act of Infrastructure Safety management through the use of the health index and of historical
Management, and then automatically estimates the maintenance cost and condition assessment data. The follow-up research,
cost and HI changes. which is already being performed, covers the systematic methods
Figure 3 shows the captured screen of the system. Fig. 3(a) is of the service levels for infrastructure asset management and
the login page. The program runs on the Web, and multiple users defines the missing links between the service levels, performance
can access the system. The maintenance and management team measurements, and budget forecast.
members have access to the system. Fig. 3(b) is the registration
of bridge data for inputting new data into the system. Fig. 3(c) References
shows the historical HI data for each bridge. Based on the
historical data, users can review the previous budgets and health Bulusu, S. and Sinha, K. C. (1997). “Comparison of methodologies to
indices. In Fig. 3(d), by setting the HI for the next year, the predict bridge deterioration.” Transportation Research Record:
expected budget to meet the HI can be calculated based on the Journal of the Transportation Research Board, No. 1597, pp. 34-42.
historical cost and HI data. Butt, A. A., Shahin, M. Y., Feighan, K. J., and Carpenter, S. H. (1987).
“Pavement performance prediction model using the Markov
process.” Transportation Research Record, No. 1123, pp. 12-19.
4. Conclusions
Cady, P. D. and Weyers, R. E. (1984). “Deterioration rates of concrete
bridge decks.” Journal of Transportation Engineering, Vol. 110, No.
This study describes the development of a bridge asset 1, pp. 34-44.
management system. It includes the following: (i) a deterioration Cesare, M. A., Santamarina, C., Turkstra, C., and Vanmarcke, E. H.
prediction model at the element level of bridge management (1992). “Modeling bridge deterioration with markov chains.” Journal
using the Markov chain model, (ii) the optimization of MR&R of Transportation Engineering, Vol. 118, No. 6, pp. 820-833.

Vol. 17, No. 7 / November 2013 − 1559 −

 Taehoon Hong, Myung Jin Chae, Duyon Kim, Choongwan Koo, Kyo Sun Lee, and Kyoung Ho Chin

Chae, M. J. and Abraham, D. M. (2001). “Neuro-fuzzy approaches for environmental categories for markovian deterioration models of
sanitary sewer pipeline condition assessment.” Journal of Computing bridge decks.” Journal of Bridge Engineering, Vol. 8, No. 6, pp.
in Civil Engineering, Vol. 15, No. 1, pp. 4-14. 353-361.
DeStefano, P. D. and Grivas, D. A. (1998). “Method for Estimating transition Morcous, G., Rivard, H., and Hanna, A. M. (2002). “Modeling bridge
probability in bridge deterioration models.” Journal of Infrastructure deterioration using case-based reasoning.” Journal of Infrastructure
Systems, Vol. 4, No. 2, pp. 56-62. Systems, Vol. 8, No. 3, pp. 86-95.
Durango-Cohen, P. L. (2004). “Maintenance and repair decision making Ortiz-Garcia, J. J., Costello, S. B., and Snaith, M. S. (2006). “Derivation
for infrastructure facilities without a deterioration model.” Journal of transition probability matrices for pavement deterioration modeling.”
of Infrastructure Systems, Vol. 10, No. 1, pp. 1-8. J. Transp. Eng., Vol. 132, No. 2, pp. 141-161.
Hong, T. and Hastak, M. (2005). “MEMRRES: Model for evaluating Shepard, R. W. (2005). “Bridge management issues in a large agency,
maintenance, repair and rehabilitation strategies in concrete bridge structure & infrastructure engineering: Maintenance, management.”
decks.” Civil Engineering & Environmental Systems, Vol. 22, No. 4, Life-Cycle Design & Performance, Vol. 1, No. 2, pp. 159-164.
pp. 233-248. Shepard, R. W. and Johnson, M. B. (2001). California bridge health
Hong, T. and Hastak, M. (2007). “Evaluation and determination of index: A diagnostic tool to maximize bridge longevity investment,
optimal MR&R strategies in concrete bridge decks.” Automation in TR News, TRB, pp. 6-11.
Construction, Vol. 16, No. 2, pp. 165-175. Thompson, P. D. and Johnson, M. B. (2005). “Markovian bridge
Hong, F. and Prozzi, J. A. (2006). “Estimation of pavement performance deterioration: Developing models from historical data.” Structure &
deterioration using Bayesian approach.” Journal of Infrastructure Infrastructure Engineering, Vol. 1, No. 1, pp. 85-91.
Systems, Vol. 12, No. 2, pp. 77-86. Veshosky, D., Beidleman, C. R., Buetow, G. W., and Demir, M. (1994).
Jiang, Y. and Sinha, K. C. (1989). “Bridge service life prediction model “Comparative analysis of bridge superstructure deterioration.” Journal
using the Markov chain.” Transportation Research Record, No. of Structural Engineering, Vol. 120, No. 7, pp. 2123-2136.
1223, pp. 24-30. Winston, W. L. (1994). Operations research: Applications and algorithms
Morcous, G., Lounis, Z., and Mirza, M. S. (2003). “Identification of 3rd edition, Duxbury Press, ISBN 0534520200, Philadelphia.

Appendix. Dataset used for Calculating Transition Probability for PSC Bridge

Table 7. Transition Probability of Pavement for PSC Bridges Table 9. Transition Probability of Girder for PSC Bridges
Transition Transition
Year matrix p1 p2 p3 p4 p5 Year matrix of p1 p2 p3 p4 p5
of the zone the zone
1-5 P1 0.7358 1 1 1 1 1-5 P1 0.7588 1 1 1 1
6-10 P2 1 0.9435 0.1614 0.7969 1 6-10 P2 1 0.9477 0.1741 0.7044 1
11-15 P3 0.9439 0.9035 0.7170 0.8607 1 11-15 P3 0.9517 0.9097 0.7286 0.8014 1
16-20 P4 0.8956 0.8539 0.7490 0.8593 1 16-20 P4 0.9114 0.8642 0.7578 0.8433 1
21-25 P5 0.7867 0.7289 0.6937 0.7704 1 21-25 P5 0.8093 0.7647 0.7055 0.8050 1
26-30 P6 0.2043 0.2441 0.3002 0.4568 1 26-30 P6 0.4864 0.4754 0.4766 0.6245 1
31-35 P7 0.0998 0.0873 0 0 1 31-35 P7 0 0 0 0 1

Table 8. Transition Probability of Deck for PSC Bridges Table 10. Transition Probability of Handrail/curb for PSC Bridges
Transition Transition
Year matrix of p1 p2 p3 p4 p5 Year matrix of p1 p2 p3 p4 p5
the zone the zone
1-5 P1 0.7288 1 1 1 1 1-5 P1 0.7446 1 1 1 1
6-10 P2 0.9456 0.9218 0.7044 0.9134 1 6-10 P2 1 0.9502 0 0.6798 1
11-15 P3 0.9430 0.8988 0.7765 0.8444 1 11-15 P3 0.9547 0.9111 0.7231 0.7444 1
16-20 P4 0 0.3110 0.8921 0.7585 1 16-20 P4 0.7942 0.8810 0.8535 0.7981 1
21-25 P5 0.1000 0.1790 0.7950 0.7338 1 21-25 P5 0.8000 0.7802 0.7348 0.7125 1
26-30 P6 0.1000 0 0 0 1 26-30 P6 0.4492 0.4693 0.4858 0.5483 1

− 1560 − KSCE Journal of Civil Engineering

 Infrastructure Asset Management System for Bridge Projects in South Korea

Table 11. Transition Probability of Expansion Joint for PSC Bridges Table 14. Transition Probability of Second Element for PSC Bridges
Transition Transition
Year matrix of p1 p2 p3 p4 p5 Year matrix of p1 p2 p3 p4 p5
the zone the zone
1-5 P1 0.7237 1 1 1 1 1-5 P1 0.7525 1 1 1 1
6-10 P2 1 0.9058 0.8575 0 1 6-10 P2 1 0.9246 0.5861 0.8601 1
11-15 P3 0.9583 0.9039 0.8211 0.6060 1 11-15 P3 0.9460 0.8970 0.7521 0.8747 1
16-20 P4 0.8731 0.8281 0.7662 0.7434 1 16-20 P4 0.8750 0.8335 0.7608 0.8585 1
21-25 P5 0.6754 0.6224 0.6128 0.6667 1 21-25 P5 0.7353 0.6920 0.6581 0.7681 1
26-30 P6 0 0 0 0.0379 1 26-30 P6 0.0151 0.1091 0.1884 0.3777 1
31-35 P7 0 0 0 0 1

Table 12. Transition Probability of Abutment/bridge Pier for PSC Table 15. Transition Probability of Bridge Bearing for PSC Bridges
Bridges Transition
Transition Year matrix of p1 p2 p3 p4 p5
Year matrix of p1 p2 p3 p4 p5 the zone
the zone 1-5 P1 0.7257 1 1 1 1
1-5 P1 0.7367 1 1 1 1 6-10 P2 1 0.9078 0.8211 0.5004 1
6-10 P2 0.9008 0.9133 0.7864 0.8159 1 11-15 P3 0.9510 0.9009 0.8147 0.7278 1
11-15 P3 0.9382 0.8924 0.7839 0.7777 1 16-20 P4 0.8710 0.8249 0.7656 0.7851 1
16-20 P4 0.8277 0.7948 0.7171 0.7724 1 21-25 P5 0.6670 0.6284 0.6114 0.6919 1
21-25 P5 0.4863 0.4924 0.4785 0.5908 1 26-30 P6 0 0 0 0.0659 1
26-30 P6 0 0 0 0 1

Table 13. Transition Probability of Drainage for PSC Bridges Table 16. Transition Probability of Foundation for PSC Bridges
Transition Transition
Year matrix of p1 p2 p3 p4 p5 Year matrix of p1 p2 p3 p4 p5
the zone the zone
1-5 P1 0.5198 1 1 1 1 1-5 P1 0.7359 1 1 1 1
6-10 P2 0.6535 0.9177 0.8999 0.2941 1 6-10 P2 0.9908 0.9161 0.8392 0 1
11-15 P3 0.9608 0.9187 0.8434 0.6897 1 11-15 P3 0.9528 0.9031 0.8262 0.6471 1
16-20 P4 0.8884 0.8623 0.7893 0.7646 1 16-20 P4 0.8823 0.8282 0.8100 0.7440 1
21-25 P5 0.1651 0.6755 0.6270 0.6696 1 21-25 P5 0.7276 0.6797 0.6802 0.6821 1
26-30 P6 0.1000 0 0 0.0205 1 26-30 P6 0 0 0 0 1

Vol. 17, No. 7 / November 2013 − 1561 −