RECENT ADVANCES in COMPUTATIONAL INTELLIGENCE, MAN-MACHINE SYSTEMS and CYBERNETICS

Fuzzy Diagnostic System for Railway Bridges

PETR RUDOLF1,2, JAROSLAV MENČÍK2, JIŘÍ KŘUPKA3
1
Railway Infrastructure Administration, Directorate, Dlážděná 1003/7, 110 00 Prague
2
Jan Perner Transport Faculty, Department of Mechanics, Materials and Machine Parts
3
Faculty of Economics and Administration, Institute of System Engineering and Informatics
University of Pardubice, Studentská 95, 532 10 Pardubice
CZECH REPUBLIC
[Link]@[Link], [Link]@[Link], [Link]@[Link]

Abstract: - This paper deals with the use of fuzzy logic for building of classification models of the technical condition
evaluation of bridge objects, both their superstructure and substructure. The proposed models have hierarchical
architecture, built of the Mamdani’s fuzzy inference systems. The models were validated on a data set of real bridges in
operation. In the modelling process, the analysis of bridge rating methods in the Czech Republic and abroad was
applied. The analysis of the amounts and shapes of input and output membership functions of given fuzzy sets was
carried out, the numbers of fuzzy inference rules were determined. On the basis of the achieved results, the utility of
the presented method of soft computing in the evaluation of the bridge technical conditions was proved.

Key-Words: - Bridge condition evaluations; soft computing; fuzzy logic; hierarchical architecture

1 Introduction According to the current level of knowledge and degree

Large and long-life structures, such as bridges, suffer by of processing of parameters entering into the process of
gradual deterioration due to corrosion, fatigue and other structure evaluation, its reliability is quantified using
processes, and after some time they must be repaired or reliability conditions [24]. These conditions are defined
replaced by a new object. This is very expensive, and in relation to the applied method of reliability theory.
such decision must be based on the good knowledge of According to the way of expressing of random character
actual condition. Regular inspections of bridges belong of reliability parameters, deterministic, semiprobabilistic
to common practice. However, it is impossible to and fullyprobabilistic methods can be distinguished.
characterise the overall condition of a bridge by a simply If classical mathematical statistics come from the law
measurable quantity. It is influenced by many factors, of empirical probability, using the knowledge of
and many of them can be characterised only verbally distribution of probability of random events, the methods
(e.g. many short cracks, water seeping into the for work with uncertainty come from the so-called law
construction, etc.). As a consequence, the result of of distribution of possibility [18]. The quality of human
evaluation depends less or more on the subjective judgement is characterized by the ability of effective
opinion of the inspector. With respect to the tens of processing of not very precise information. This
various criteria, it can happen that the evaluation of the capability, together with the other qualities of human
same object by two inspectors is less or more different. reasoning, becomes the interest centre of an artificial
Therefore, a method is needed, which would be more intelligence (AI) [21]. AI methods seem to be very
objective. The probabilistic methods cannot be applied promising for the description and control of complicated
simply in this case, just because of the lack of data and systems. The most important of them are the
vagueness of the characteristic criteria and way of their characteristics of processing, above all, non-numerical,
evaluation. Fortunately, it appears that the situation can linguistic information. Approaches of modelling, where
be improved by the application of methods based on this integration of knowledge is enabled, the ability of
modern tools of artificial and computational intelligence, self-learning, robustness and easy implementation are
such as fuzzy logic (FL). supported at the expense of preciseness, and they are
The objective of this paper is the reliability and ranged into the framework of the so called “soft
service life of existing bridge objects and their computing“ or “computational intelligence”
assessment using modern computational tools. General [9,13,15,16,18].
principles on reliability for various structures are in [4]. Goal of this paper is a verification of FL utilization
Bases for design of structures and assessment of existing for the evaluation of the technical conditions of existing
structures are in [2,5]. From the point of design of new bridge objects, thereby also their reliability and service
structures, and the assessment of existing ones, we are life, on the bases of models of their defects (damages).
interested in quantification of their reliability level.

ISSN: 1790-5117 200 ISBN: 978-960-474-144-1

 RECENT ADVANCES in COMPUTATIONAL INTELLIGENCE, MAN-MACHINE SYSTEMS and CYBERNETICS

2 Problem Formulation 4th “discontinuity”, 5th “displacement” and the last is

Within the 6th Framework programme of the European “loss of material”. In the lower level, each defect type
Union, an international research project has been solved has more defect kinds, e.g. there are 6.1 “loss of
[25]. The principles for the carrying out of inspections concrete” and 6.2 “loss of steel” for “loss of material”. In
and condition assessments of existing railway bridges another of the remaining two levels, defect kinds can
are presented in [22] and for load determining and have categories and these then can have defect classes
resistance assessment of railway bridges in [23]. [10,19,20,24].
Available results of [22,23,25] have been adopted Furthermore, according to bridge defects di are
into this research, it means methodology of hierarchical described as triple by their parameters defect extent ei,
classification of railway bridge defects, application of defect intensity ii and defect location the following way:
non-dimensional geometrical bridge model, the way of di = { defect location, ei, ii }, (1)
quantitative defects description and principles for initial
assessment (rating) level of bridge condition. where: defect location means superstructure or
The basic quantification of reliability of an existing substructure of the bridge, ei and ii are defined for defect
bridge is load-carrying capacity or capability, regarding types of the bridge.
its actual technical condition and representing also the
basic quantitative parameter. Data obtained during
inspection and condition assessment are crucial to 3 Fuzzy diagnostic model
estimate the current state of a bridge structures The part is focused on the design of diagnostic model of
reliability. So that the basis of reliability assessment of bridge defect evaluation. This model is possible
the bridge is the evaluation of its condition (rating), perceived classification problem. Classification deals
which in practical judging data is, however, often with knowledge and data characterized by uncertainty.
incomplete, numerically imprecise and also linguistic. This was realized by means of a fuzzy inference system
A supervising activity in [26] consists of general (FIS) [8,16]. The heuristic approach for the creation of
(annual) and detailed (three yearly) inspections namely. FIS (it means the shape and number of membership
Protocol about a detailed bridge inspection quotes found function of inputs and output variables, and the fuzzy
faults and the proposal of total condition assessment of rule base (FRB)) was used because an exact general
the railway bridge object using three degrees [26]. method for definition of their number does not exist [9].
Degree 1 - condition state “good” means that bridge A definition of the number of fuzzy rules (FRs) is
object requires only general maintenance. Degree 2 - described in [8,9] or the method in [12,27] can be used.
“satisfactory” means that bridge object requires repair The number of FRs can be also optimized by genetic
extending the general maintenance framework, and if algorithms and evolution strategies [1,11].
necessary replacement of some parts, however the defects The FIS is (Fig.1) represented by a block with inputs
do not immediately threaten the safety of operation. xn and output y and can be defined as MISO (Multiple
Degree 3 - “unsatisfactory” means that bridge object Inputs and Single Output) system. It is more described in
requires full reconstruction, reconstruction of supports or [9,17]. A disadvantage of this approach to the design of
the replacement of superstructure, a nd if necessary, even FIS [3,15,17] is an exponential growth of the number of
only the repair or replacement of some parts, whose FRs in FRB and the FIS can be realized ineffectively.
condition do not immediately threaten the safety of
operation. The condition evaluation of bridge
superstructure and substructure is always recorded
separately.
In this paper we chose twelve real bridge objects (of
the given construction type) with various proposed
condition evaluations of their superstructure and
substructure, evaluated by their inspectors. Then, data
could be evaluated about defects found from protocols Fig.1 MISO fuzzy inference system
about detailed inspections of these bridges. Afterwards,
we described these defects quantitatively according to This problem can be removed by a hierarchical
the principles, given in [22]. The condition of bridge structure (Fig.2) of FIS [3,7,15,17]. In the hierarchical
superstructure and substructure is always evaluated on structure of FIS (HSFIS) it is necessary to determine the
the basis of their found defects [26]. Bridge defects are number of FRs for the first and other levels, see more in
hierarchically classified [22] in four levels. In the highest [17].
level, there are defects classified into six types: 1st is
“contamination”, 2nd “deformation”, 3rd “deterioration”,

ISSN: 1790-5117 201 ISBN: 978-960-474-144-1

 RECENT ADVANCES in COMPUTATIONAL INTELLIGENCE, MAN-MACHINE SYSTEMS and CYBERNETICS

INPUTS:
Extent ei and Intensity ii Data Values:
Loss of concrete e1, i1 and steel e2, i2
Discontinuity of concrete e3, i3 and steel e4, i4
Deterioration of concrete e5, i5 and steel e6, i6
Contamination e7, i7
Deformation e8, i8
Deterioration e9, i9

…
First level
Indicator of loss of concrete in1
Indicator of loss of steel in2
Indicator of discontinuity of concrete in3
Indicator of discontinuity of steel in4
Indicator of deterioration of concrete in5 …
Fig.2 Types of FIS hierarchical structure
Indicator of deterioration of steel in6
The part is focused on design of hierarchical fuzzy
diagnostic models (HFDMs). This problem is composed …
of two phases: first one is a synthesis and analysis of Second level
HFDM1 for evaluation of bridge superstructure defects Index of loss of material ix1
(Fig.3) and second one is a synthesis and analysis of Index of discontinuity ix2
HFDM2 for evaluation of bridge substructure defects. Index of displacement ix3
Index of contamination ix4
Parameters for HFDMs can be expressed by Index of deformation ix5
incompleteness and disproportion. HFDMs are created in Index of deterioration ix6
MATLAB.
Proposed HFDM1 represent three-level hierarchical …
structure. Real inputs are e1, i1, e2, …, e9, i9. The first OUTPUT / Third level
level is created by 6 Mamdani’s FISs and represents the Index of technical condition ix7
evaluation of indicators of (some) defect kinds of bridge
superstructure and has inputs e1, i1, e2, …, e6, i6. The Fig.3 Fuzzy diagnostic model for evaluation of
second level is created by 6 FISs and represents the bridge superstructure defects
evaluation of indices of (all) defect types of bridge
Table 1 Value ranges of inputs for FISs in the first level
superstructure and has inputs in1, in2, …, in6 and e7, i7,
Extent of Intensity of
…, e9, i9. The third level is created by 4 FISs, has inputs Kind of
defect kind defect kind
ix1, ix2, …, ix6 and represents the evaluation of the (one) defect
Min. Max. Min. Max.
index of the technical condition of bridge superstructure Deterioration
ix7 - the output (real data) of HFDM1. In a choice of the 0 100 0 40
of concrete
input and output membership functions, a comparison of Deterioration
Gaussian MFs of the first type with (starting) triangular 0 100 0 20
of steel
MFs was carried out. FRB has 9 rules in all FISs for the Discontinuity
evaluation of kinds and types of bridge defects and of 0 100 0 20
of concrete
729 rules in all 4 FISs for the evaluation of index of Discontinuity
bridge technical condition. For defuzzification the 0 100 0 20
of steel
method of the middle of maximum was chosen in the Loss of
0 100 0 40
end. concrete
In the following Tables 1 to 3 the value ranges of Loss of
0 100 0 20
inputs for FISs in all three HFDM1 levels are presented. steel
Verification of HFDM1 of the condition evaluation of Note: All values are in percents, just only value of intensity of
bridge superstructure was carried out by means of testing discontinuity of concrete is in millimeters.
values in each of the sixth of the range of values
(universum, scale). In the following Tables 4 to 6 the
testing values of the input and output parameters for
FISs in all three HFDM1 levels are presented.

ISSN: 1790-5117 202 ISBN: 978-960-474-144-1

 RECENT ADVANCES in COMPUTATIONAL INTELLIGENCE, MAN-MACHINE SYSTEMS and CYBERNETICS

Table 2 Value ranges of inputs for FISs in the second level for the condition evaluation of the bridge substructure
Indicator of defect Indicator of defect (HFDM2).
Type of defect kind of concrete kind of steel
Min. Max. Min. Max. Table 5 Testing values of the input and output defect type
Deterioration 0.00 1.00 0.00 1.00 parameters
Discontinuity 0.00 1.00 0.00 1.00 Input/ Part of scale
Loss of output 0/6 1/6 2/6 3/6 4/6 5/6 6/6
0.00 1.00 0.00 1.00
material e7 0 17 33 50 67 83 100
Note: All values are non-dimensional. i7 0 17 33 50 67 83 100
Extent of Intensity of ix4 0.025 0.085 0.495 0.5 0.495 0.915 0.975
Type of defect defect type defect type e8 0 17 33 50 67 83 100
Min. Max. Min. Max. i8 0 17 33 50 67 83 100
Contamination 0 100 0 100 ix5 0.025 0.085 0.495 0.5 0.495 0.915 0.975
Deformation 0 100 0 100 in5 0.025 0.085 0.5 0.5 0.5 0.915 0.975
Displacement 0 100 0 100 in6 0.025 0.085 0.495 0.5 0.495 0.915 0.975
Note: All values are in percents. ix3 0.025 0.04 0.5 0.5 0.5 0.96 0.975
in3 0.025 0.085 0.495 0.5 0.495 0.915 0.975
Table 3 Value ranges of inputs for FISs in the third level in4 0.025 0.085 0.495 0.5 0.495 0.915 0.975
Index of ix2 0.025 0.04 0.5 0.5 0.5 0.96 0.975
Type of defect defect type e9 0 17 33 50 67 83 100
Min. Max. i9 0 17 33 50 67 83 100
Contamination 0.00 1.00 ix6 0.025 0.085 0.495 0.5 0.495 0.915 0.975
Deformation 0.00 1.00 in1 0.025 0.085 0.5 0.5 0.5 0.915 0.975
Deterioration 0.00 1.00 in2 0 3 7 10 13 17 20
Discontinuity 0.00 1.00 ix1 0.025 0.04 0.5 0.5 0.5 0.96 0.975
Displacement 0.00 1.00 Note: All values ei and ii are in percents, all other values ini
Loss of and ixi are non-dimensional.
0.00 1.00
material
Note: All values are non-dimensional. Table 6 Testing values of the input and output technical
condition parameters
Table 4 Testing values of the input and output defect kind Input/ Part of scale
parameters output 0/6 1/6 2/6 3/6 4/6 5/6 6/6
Input/ Part of scale ix4 0.025 0.085 0.495 0.5 0.495 0.915 0.975
output 0/6 1/6 2/6 3/6 4/6 5/6 6/6 ix5 0.025 0.085 0.495 0.5 0.495 0.915 0.975
e5 0 17 33 50 67 83 100 ix3 0.025 0.04 0.5 0.5 0.5 0.96 0.975
i5 0 7 13 20 27 33 40 ix2 0.025 0.04 0.5 0.5 0.5 0.96 0.975
in5 0.025 0.085 0.5 0.5 0.5 0.915 0.975 ix6 0.025 0.085 0.495 0.5 0.495 0.915 0.975
e6 0 17 33 50 67 83 100 ix1 0.025 0.04 0.5 0.5 0.5 0.96 0.975
i6 0 3 7 10 13 17 20 ix7
in6 0.025 0.085 0.495 0.5 0.495 0.915 0.975 1.05 1.08 2 2 2 2.92 2.95
(3MFs)
e3 0 17 33 50 67 83 100 ix7
i3 0 3 7 10 13 17 20 1.02 1.04 2 2 2 2.96 2.98
(5MFs)
in3 0.025 0.085 0.495 0.5 0.495 0.915 0.975 ix7
e4 0 17 33 50 67 83 100 1.02 1.02 2 2 2 2.98 2.98
(7MFs)
i4 0 3 7 10 13 17 20 ix7
1.02 1.02 2 2 2 2.98 2.98
in4 0.025 0.085 0.495 0.5 0.495 0.915 0.975 (9MFs)
e1 0 17 33 50 67 83 100 Note: All values ixi are non-dimensional.
i1 0 7 13 20 27 33 40
in4 0.025 0.085 0.5 0.5 0.5 0.915 0.975 4 Conclusion
e2 0 17 33 50 67 83 100
We have presented the synthesis and analysis of models
i2 0 3 7 10 13 17 20
for the technical conditon evaluation of railway bridges
in2 0.025 0.085 0.495 0.5 0.495 0.915 0.975
Note: All values ei and ii are in percents, values i4 are in
by one of the methods of soft computing. The proposed
millimeters, and values ini are non-dimensional. method of evaluation of technical condition of existing
bridges using fuzzy logic is interesting and effective. At
In the similar way like the fuzzy diagnostic model for the same time the paper concentrates on practical
the condition evaluation of the bridge superstructure application and indicates the way this system can
(HFDM1), we also created the fuzzy diagnostic model proceed in its development.

ISSN: 1790-5117 203 ISBN: 978-960-474-144-1

 RECENT ADVANCES in COMPUTATIONAL INTELLIGENCE, MAN-MACHINE SYSTEMS and CYBERNETICS

We have contributed to a science and bridge practice [6] Jirava, P., Křupka, J. Modelling of Rough-Fuzzy
development and we have proven the utility of fuzzy Classifier. WSEAS Transactions on Systems. Vol.7,
logic for the diagnostics of bridge technical conditions. No.3, 2008, pp. 251-262.
To further utilise this method, we propose that the data [7] Křupka, J., Kašparová, M., Modelling of Internal
of bridge technical condition should be collected in a Human Population Migration Classifiers by Fuzzy
more appropriate manner by means of the proposed Inference System and Its Hierarchical Structure.
inspection forms. WSEAS Transaction on Systems, Vol.6, WSEAS
It is possible to state, as shown above, that the best Press, Athens New York, 2007, pp. 461-467.
result is achieved by the simulation model, created using [8] Kuncheva, L.I., Fuzzy Classifier Design. Physica-
a fuzzy inference system, with nine output membership Verl., Heidelberg New York, 2000.
functions ix7 (9MFs). It is because of the fact that the [9] Lee, [Link]., Fuzzy Logic in Control Systems:
bigger amount of membership functions of the output Fuzzy Logic Controller- Part I and II. IEEE
variable expresses the resulting assessment of the bridge Transaction on Systems, Man, and Cybernetics,
technical condition more exactly and with more details. Vol.20, 1990, pp. 404-433.
Using the resulting values concerning the technical [10] Menčík, J., Rudolf, P. Optimisation of Maintenance
condition of individual bridges with values 1, 2, 3, Strategy in a Bridge Network. In. Proceedings 5.
realistic approximate values (e.g. 2.25) are obtained. Pardubice: University of Pardubice, 2005, pp. 224-
Managerial decision-makers would then be able to make 231.
use of this technical condition index data and as such [11] Pal, S.K., Wang, P.P., Genetic Algorithm for
they would be able to prioritise funding regarding the Pattern Recognition. CRC Press Inc., Boston, 1996.
repair of bridge structures (e.g. 2.75 being more [12] Nozaki, K., Ishibuchi, H., Tanaka, H., Adaptive
necessary than 2.25). Fuzzy Rules Based Classification Systems. IEEE
The obtained knowledge is the way forward for the Transaction on Fuzzy Systems, Vol.4, No.3, 1996,
authors’ further research activities in the given branch. pp. 238-250.
On the basis of analysis of the simulation model it would [13] Olej, V., Křupka, J., Analysis of Decision Processes
be possible to optimise the number and shapes of input of an Automation Control System with Uncertainty.
and output membership functions by means of genetic [Scientific Monograph, Series: Technical
algorithms [11,18], and also to optimise the number of Cybernetics]. Košice (Slovakia): University Press
rules in the FIS by means of the so-called theory of Elfa Ltd., 1996.
rough sets [6,14]. [14] Pawlak, Z. Rough Sets: Theoretical Aspects of
Reasoning About Data. 1st Ed. Dordrecht: Kluwer
Academic Publisher, 1991.
Acknowledgement [15] Pedrycz, W., Fuzzy Control and Fuzzy Systems. 2nd
The work was supported by the Czech Science edn. Research Studies Press Ltd., London, 1993.
Foundation - project GAČR 103/08/1340 “Fatigue [16] Ross, T.J., Fuzzy Logic with Engineering
endurance of steel orthotropic bridge decks’. Applications. 2nd edn. John Wiley and Sons, Ltd.,
New York, 2004.
[17] Raju, G.V.S., Zhou, J., Adaptive Hierarchical
References: Fuzzy Controler. IEEE Transactions on Systems,
[1] Buckley, J.J., Hayaashi, Y., Genetic Algorithm and Man, and Cybernetics, Vol.23, No.4, 1993, pp. 973-
Applications. Fuzzy Sets and Systems, Vol.61, 1994, 980.
pp. 129-136. [18] Pokorný, M. Artificial inteligence in modelling and
[2] ČSN ISO 13822 Bases for design of structures – managamenent. 1. vyd. Praha: Nakladatelství BEN,
Assessment of existing structures. Praha: Český 1996. (in Czech)
normalizační institut, 2005. (in Czech) [19] Rudolf, P. Reliability and the Service Life of
[3] Gegov, A.E., Frank, P.M., Hierarchical Fuzzy Existing Bridges and Their Assessment Using
Control Multivariable Systems. Fuzzy Sets and Modern Computational Tools. Dissertation thesis.
Systems, Vol.72, 1995, pp. 299-310. Pardubice: University of Pardubice, Jan Perner
[4] [4r] ISO 2394 General principles on reliability for Transport Faculty, 2009. (submitted for defence)
structures. The International Standard. Genève: [20] Rudolf, P. Bridge Condition Evaluation Using
International Organization for Standardization, 1998. Fuzzy Logic. In. Proc. of Technical Computing
[5] ISO 13822 Bases for design of structures – Prague 2009, 10 pp. URL:
Assessment of existing structures. The International [Link] matlab/
Standard. Genève: International Organization for MATLAB09/prispevky/088_rudolf.pdf.
Standardization, 2001.

ISSN: 1790-5117 204 ISBN: 978-960-474-144-1

 RECENT ADVANCES in COMPUTATIONAL INTELLIGENCE, MAN-MACHINE SYSTEMS and CYBERNETICS

[21] Russell, S., Norvig, P., Artificial Intelligence: A

Modern Approach. 2nd Ed. New Jersey: Prentice
Hall. 2003.
[22] SB-ICA Guideline for Inspection and Condition
Assessment of Existing European Railway Bridges.
Guideline of the 6th FP project Sustainable Bridges
– Assessment for Future Traffic Demands and
Longer Lives. European Commission DG VII.
Göteborg: Skanska Teknik AB, 2007. URL:
[Link]
[23] SB-LRA Guideline for Load and Resistance
Assessment of Existing European Railway Bridges.
Guideline of the 6th FP project Sustainable Bridges
– Assessment for Future Traffic Demands and
Longer Lives. European Commission DG VII.
Göteborg: Skanska Teknik AB, 2007. URL:
[Link]
[24] Šertler, H. Determination of reliability of existing
railway bridges Vědeckotechnický sborník Českých
drah, No.7, 1999, pp. 29-37. (in Czech).
[25] Sustainable Bridges – Assessment for Future
Traffic Demands and Longer Lives. The 6th EU
Framework Programme project. Wrocław:
Dolnośląskie Wydawnictwo Edukacyjne. 2007.
[26] SŽDC (ČD) S 5 Administration of bridge objects.
Service instructions (republicated). Praha:
Ministerstvo dopravy ČR, 1995. (in Czech)
[27] Wang, L.X., Mendel, J.M., Generating Fuzzy Rules
by Learning from Examples. IEEE Transaction on
Systems, Man, and Cybernetics, Vol.22, 1992, pp.
1414-1427.

ISSN: 1790-5117 205 ISBN: 978-960-474-144-1