Computer-Aided Civil and Infrastructure Engineering 16 (2001) 126–142

REVIEW ARTICLE

Neural Networks in Civil Engineering: 1989−2000

Hojjat Adeli*
Department of Civil and Environmental Engineering and Geodetic Science,
The Ohio State University, Columbus, Ohio 43210, USA

Abstract: The first journal article on neural network Yeh (1989) in this journal. Since then, a large number of arti-
application in civil/structural engineering was published in cles have been published on civil engineering applications
this journal in 1989. This article reviews neural network of neural networks. Most of these articles deal with some
articles published in archival research journals since then. type of pattern-recognition or learning problem. A neural
The emphasis of the review is on the two fields of structural network can be trained to learn to perform a particular task.
engineering and construction engineering and management. The approach is particularly attractive for hard-to-learn
Neural networks articles published in other civil engi- problems and when there is no formal underlying theory for
neering areas are also reviewed, including environmental the solution of the problem. Engineering design and image
and water resources engineering, traffic engineering, high- recognition are two such problems (Adeli and Hung, 1995).
way engineering, and geotechnical engineering. The great One of the reasons for popularity of the neural network is
majority of civil engineering applications of neural net- the development of the simple error backpropagation (BP)
works are based on the simple backpropagation algorithm. training algorithm (Rumelhart et al., 1986), which is based
Applications of other recent, more powerful and efficient neu- on a gradient-descent optimization technique. The BP algo-
ral networks models are also reviewed. Recent works on inte- rithm is now described in many textbooks (Adeli and Hung,
gration of neural networks with other computing paradigms 1995; Mehrotra et al., 1997; Topping and Bahreininejad,
such as genetic algorithm, fuzzy logic, and wavelet to enhance 1997; Haykin, 1999), and unfamiliar readers can refer to
the performance of neural network models are presented. any one of them. A review of the BP algorithm with sug-
gestions on how to develop practical neural network appli-
cations is presented by Hegazy et al. (1994). The great
1 INTRODUCTION majority of the civil engineering application of neural net-
works is based on use of the BP algorithm primarily
Artificial neural networks (ANNs) are a functional abstrac- because of its simplicity. Training of a neural network with
tion of the biologic neural structures of the central nervous a supervised learning algorithm such as BP means finding
system (Aleksander and Morton, 1993; Rudomin et al., the weights of the links connecting the nodes using a set
1993; Arbib, 1995; Anderson, 1995). They are powerful of training examples. An error function in the form of the
pattern recognizers and classifiers. They operate as black- sum of the squares of the errors between the actual out-
box, model-free, and adaptive tools to capture and learn puts from the training set and the computed outputs is min-
significant structures in data. Their computing abilities have imized iteratively. The learning or training rule specifies
been proven in the fields of prediction and estimation,
how the weights are modified in each iteration.
pattern recognition, and optimization (Adeli and Hung,
1995; Golden, 1996; Mehrotra et al., 1997; Adeli and Park,
1998; Haykin, 1999). They are suitable particularly for 2 STRUCTURAL ENGINEERING
problems too complex to be modeled and solved by classi-
cal mathematics and traditional procedures. 2.1 Pattern recognition and machine learning in
The first journal article on civil/structural engineering structural analysis and design
applications of neural networks was published by Adeli and
Adeli and Yeh (1989) present a model of machine learn-
* E-mail: Adeli.1@osu.edu. ing in engineering design based on the concept of internal

© 2001 Computer-Aided Civil and Infrastructure Engineering. Published by Blackwell Publishers, 350 Main Street, Malden, MA 02148, USA,
and 108 Cowley Road, Oxford OX4 1JF, UK.
 Neural networks in civil engineering: 1989–2000 127

control parameters and perceptron (Rosenblatt, 1962). A The BP algorithm is used by Yeh et al. (1993) as a
perceptron is defined as a four-tuple entity (sensors to knowledge-acquisition tool for a knowledge-based system
receive inputs, weights to be multiplied by the sensors, for diagnosing damage to prestressed concrete piles (such
a function collecting all the weighted data to produce a as spalling of concrete and transverse cracking or break-
proper measurement on the impact of the observed phe- ing of the pile); by Kang and Yoon (1994) for design of
nomenon, and a constant threshold), and the structural simple trusses; by Hoit et al. (1994) for equation renum-
design problem is formulated as a perceptron without hid- bering in finite element analysis of structures to improve
den units. Adeli and Yeh apply the model to the design of profile and wavefront characteristics; by Rogers (1994) for
steel beams. fast approximate structural analysis in a structural opti-
Vanluchene and Sun (1990) demonstrate potential appli- mization program; by Mukherjee and Deshpande (1995a,
cations of the BP algorithm (Rumelhart et al., 1986) 1995b) for the preliminary design of structures; by Abdalla
in structural engineering by presenting its application and Stavroulakis (1995) to predict the behavior of semirigid
to three problems—a simple beam load location prob- connections in steel structures from experimental moment
lem involving pattern recognition, the cross-section selec- rotation curves for single-angle and single-plate beam-
tion of reinforced-concrete beams involving typical design column connections; by Turkkan and Srivastava (1995)
decisions, and analysis of a simply supported plate— to predict the steady-state wind pressure profile for air-
showing how numerically complex solutions can be esti- supported cylindrical and hemispherical membrane struc-
mated quickly with the neural network approach. tures; by Mukherjee et al. (1996) to predict the buckling
Hajela and Berke (1991) demonstrate that neural net- load of axially loaded columns based on experimental data;
works can be used for rapid reanalysis for structural opti- by Papadrakakis et al. (1996) for structural reliability anal-
mization. Hung and Adeli (1991a) present a model of ysis in connection with the Monte Carlo simulation; by
machine learning in engineering design, called PERHID, Anderson et al. (1997) to predict the bilinear moment-
based on the concept of the perceptron learning algorithm rotation characteristics of the minor-axis beam-to-column
(Rosenblatt, 1962; Adeli and Yeh, 1989) with a two-layer connections based on experimental results; by Szewczyk
neural network. PERHID has been constructed by com- and Noor (1996, 1997) for sensitivity and nonlinear anal-
bining a perceptron with a single-layer AND neural net. ysis of structures; by Kushida et al. (1997) to develop a
Extending this research, Hung and Adeli (1994a) present a concrete bridge rating system; by Hegazy et al. (1998) to
neural network machine learning development environment model the load-deflection behavior, concrete strain distribu-
using the object-oriented programming paradigm (Yu and tion at failure, reinforcing steel strain distribution at failure,
Adeli, 1991, 1993). and crack-pattern formation of concrete slabs; by Chuang
Adeli and Zhang (1993) present an improved perceptron et al. (1998) to predict the ultimate load capacity of pin
learning algorithm by introducing an adjustment factor in ended reinforced concrete columns; by Stavroulakis and
each self-modification iteration of the original perceptron Antes (1998) for crack identification in steady-state elas-
learning model. The adjustment factor in each iteration is todynamics; by Cao et al. (1998) to identify loads on air-
determined such that the domain error is reduced in the craft wings modeled approximately as a cantilever beam
subsequent iterations. This leads to global improvement in subjected to a set of concentrated loads; by Mathew et al.
the iterative process toward finding the final weight vector. (1999) for analysis of masonry panels under biaxial bend-
The application of the new algorithm to the steel beam ing; and by Jenkins (1999) for structural re-analysis of two-
design problem demonstrates that the number of iterations dimensional trusses.
needed for convergence of the vector is substantially fewer Biedermann (1997) investigates the use of the BP neural
than that using the original perceptron algorithm. networks to represent heuristic design knowledge such as
Theocaris and Panagiotopoulos (1993) describe the how to classify the members of a multistory frame into
parameter identification problem in fracture mechanics a limited number of groups for practical purposes (design
as a neural network learning problem. Gunaratnam and fabrication groups). Cattan and Mohammadi (1997) use the
Gero (1994) study the effect of representation on the BP algorithm to relate the subjective rating of bridges based
performance of neural networks in structural engineering on visual inspection of experienced bridge inspectors to
applications using the the BP algorithm. They suggest the analytical rating based on detailed structural analyses
that dimensional analysis provides a suitable representa- under standard live loads as well as the bridge parameters.
tion framework for training the input-output pattern pairs. They conclude that “neural networks can be trained and
Messner et al. (1994) describe a neural network system used successfully in estimating a rating based on bridge
for preliminary selection of the most appropriate struc- parameters.”
tural members (beams, columns, and slabs) given a build- Adeli and Park (1995c) present application of counter-
ing project’s attributes such as available site space, budget, propagation neural networks (CPNs) with competition and
and height. interpolation layers (Hecht-Nielsen, 1987a, 1987b, 1988) in
128 Adeli

structural engineering. A problem with the CPN algorithm the efficiency of the algorithm by (1) providing a good
is the arbitrary trial-and-error selection of the learning coef- initial solution and (2) playing the role of the precondi-
ficients encountered in the algorithm. The authors propose tioner in the PCG algorithm. Consolazio applies the method
a simple formula for the learning coefficients as a function to finite element analysis of flat-slab highway bridges and
of the iteration number and report excellent convergence concludes the neural network to be an effective method for
results. The CPN algorithm is used to predict elastic criti- accelerating the convergence of iterative methods. Use of
cal lateral torsional buckling moment of wide-flange steel neural networks in finite element analysis is also discussed
beams (W shapes) and the moment-gradient coefficient for by Li (2000).
doubly and singly symmetric steel beams subjected to end
moments. The latter is a complex stability analysis problem 2.2 Design automation and optimization
requiring a large neural network with 4224 links, exten-
sive numerical analysis, and management of a large amount Automation of design of large one-of-a-kind civil engineer-
of data. It took less than 30 iterations to train the large ing systems is a challenging problem due partly to the
CPN network in both competition and interpolation lay- open-ended nature of the problem and partly to the highly
ers using 528 training instances. Compared with the BP nonlinear constraints that can baffle optimization algo-
algorithm, the authors found superior convergence property rithms (Adeli, 1994). Optimization of large and complex
and a substantial decrease in the processing time for the engineering systems is particularly challenging in terms
CPN algorithm with the proposed formula for the learning of convergence, stability, and efficiency. Most of the neu-
coefficients. ral network research has been done in the area of pattern
The computation of an effective length factor, K, is com- recognition and machine learning (Adeli and Hung, 1995).
plicated but essential for design of members in compres- Neural network computing also can be used for optimiza-
sion in steel-frame structures. The present AISC codes for tion (Berke et al., 1993).
design of steel structures (AISC, 1995, 1998) present sim- Adeli and Park (1995a) present a neural dynamics
plified alignment charts for determining the effective length model for optimal design of structures by integrating the
factor. Duan and Chen (1989) and Kishi et al. (1997) have penalty function method, the Lyapunov stability theorem,
pointed out the gross underestimation (leading to an unsafe Kuhn-Tucker conditions, and the neural dynamics concept.
design) and overestimation (leading to an overly conserva- A pseudo-objective function in the form of a Lyapunov
tive design) of the alignment charts for different boundary energy functional is defined using the exterior penalty func-
conditions. Hung and Jan (1999a) describe a variation of tion method. The Lyapunov stability theorem guarantees
the cerebellar model articulation controller (CMAC), used that solutions of the corresponding dynamic system (tra-
mostly in the control domain, for predicting the effective jectories) for arbitrarily given starting points approach an
length factor, K for columns in unbraced frames. They equilibrium point without increasing the value of the objec-
conclude that the results obtained from the neural network tive function. The robustness of the model was first verified
model are more accurate than those obtained from the AISC by application to a linear structural optimization problem,
alignment charts. the minimum-weight plastic design of low-rise planar steel
In the finite element analysis of structures, the relation- frames (Park and Adeli, 1995). Optimization algorithms are
ship between the loads and displacements is represented known to deteriorate with increases in size and complexity
by the structure or global stiffness matrix. A neural net- of the problem. The significance of the new optimization
work can be trained to perform the same task. Solution model is that it provides the optimal design of large struc-
of the simultaneous linear equations including the stiff- tures with thousands of members subjected to complicated
ness matrix is the most time-consuming part of any large- and discontinuous constraints with excellent convergence
scale finite element analysis. To speed up this step of the results.
finite element analysis, neural networks can be used to cre- In order to achieve automated optimal design of realis-
ate domain-specific equation solvers using the knowledge tic structures subjected to actual constraints of commonly
of a particular domain such as highway bridges. However, used design codes such as the American Institute of Steel
neural networks can provide only an approximate solution Construction (AISC) allowable stress design (ASD) and
where an “exact” solution is usually required. Consolazio load and resistance factor design (LRFD) specifications
(2000) proposes combining neural networks with iterative (AISC, 1995, 1998), Adeli and Park (1995b, 1996) devel-
equation-solving techniques such as a preconditioned con- oped a hybrid CPN–neural dynamics model for discrete
jugate gradient algorithm (PCG) (Adeli and Kumar, 1999). optimization of structures consisting of commercially avail-
In particular, he uses the BP neural network algorithm able sections such as the wide-flange (W) shapes used in
to compute approximate displacements at each iteration, steel structures. The computational models are shown to
whereas the overall PCG steers convergence to the exact be highly stable and robust and particularly suitable for
solution. The neural network part of the algorithm improves design automation and optimization of large structures no
 Neural networks in civil engineering: 1989–2000 129

matter how large the size of the problem is, how irregular promise in structural dynamic model identification by using
the structure is, or how complicated the constraints are. For neural networks” based on simulation results for a real mul-
their innovative work, the authors were awarded a patent tistory building subjected to earthquake ground motions.
by the U.S. Patent and Trademark Office on September 29, The BP algorithm is used by Yun and Bahng (2000) for sub-
1998 (United States Patent Number 5,815,394). structural identification and estimating the stiffness param-
An important advantage of cold-formed steel is the eters of two-dimensional trusses and frames. Huang and
greater flexibility of cross-sectional shapes and sizes avail- Loh (2001) propose a neural network–based model for
able to the structural steel designer. The lack of standard modeling and identification of a discrete-time nonlinear
optimized shapes, however, makes selection of the most hysteretic system during strong earthquakes. They use two-
economical shape very difficult, if not impossible. This task dimensional models of a three-story frame and a real bridge
is further complicated by the complex and highly nonlin- in Taiwan subjected to several earthquake accelerograms to
ear nature of the rules that govern their design. Adeli and validate the feasibility and reliability of the method for esti-
Karim (1997a) present a general mathematical formulation mating the changes in structural response under different
and computational model for optimization of cold-formed earthquake events.
steel beams. The nonlinear optimization problem is solved
by adapting the robust neural dynamics model of Adeli and 2.4 Structural condition assessment and
Park (1996). The basis of design can be the AISI ASD monitoring
or LRFD specifications (AISI, 1996, 1997). The computa-
tional model is applied to three different commonly used Wu et al. (1992) discuss use of the BP algorithm for detec-
types of cross-sectional shapes: hat, I, and Z shapes. The tion of structural damage in a three-story frame with rigid
computational model was used to perform extensive para- floors. The damage is defined as a reduction in the mem-
metric studies to obtain the global optimal design curves ber stiffness. Elkordy et al. (1994) question the reliability
for cold-formed hat- , I- , and Z-shaped steel beams based of the traditional methods for structural damage diagno-
on the AISI code to be used directly by practicing design sis and monitoring that rely primarily on visual inspection
engineers (Karim and Adeli, 1999a, 1999b, 2000). and simple on-site tests. They propose a structural damage
Optimization of space structures made of cold-formed monitoring system for identifying the damage associated
steel is complicated because an effective reduced area with changes in structural signatures using the BP algo-
must be calculated for members in compression to take rithm. For training, they used experimental results from a
into account the nonuniform distribution of stresses in shaking table as well as numerical results from a finite ele-
thin cold-formed members due to torsional/flexural buck- ment analysis of the structure for strain-mode shapes as
ling. The effective area varies not only with the level of the vibrational signatures. They point out that “analyzing
the applied compressive stress but also with its width-to- the data obtained from different types of sensors to detect
thickness ratio. Tashakori and Adeli (2001) present optimal damage is a very complex problem, particularly because
(minimum weight) design of space trusses made of cold- of the noise associated with the signals,” and suggest that
formed steel shapes in accordance with the AISI specifica- neural networks can diagnose complicated damage patterns
tions (AISI, 1996, 1997) using the neural dynamics model and “can handle noisy and partially incomplete data sets.”
of Adeli and Park (1996). The model has been used to Stephens and Vanluchene (1994) describe an approach for
find the minimum-weight design for several space trusses assessing the safety condition of structures after the occur-
commonly used as roof structures in long-span commer- rence of a damaging earthquake using multiple quantitative
cial buildings and canopies, including a large structure with indices and the BP algorithm. They conclude that the neural
1548 members, with excellent convergence results. network model “generated more reliable assessments than
Arslan and Hajela (1997) discuss counterpropagation could be obtained using any single indicator or from a lin-
neural networks in decomposition-based optimal design. ear regression model that utilized all indicators.”
Parvin and Serpen (1999) discuss a procedure to solve an Defining damage as a reduction in the stiffness of struc-
optimization problem with a single-layer, relaxation-type tural members, Szewczyk and Hajela (1994) use a CPN
recurrent neural network but do not present a solution to for damage detection in truss and frame structures. They
any significant structural design problem. describe the problem as an inverse static analysis prob-
lem where the elements of the structure stiffness matrix
are found based on experimentally observed response data.
2.3 Structural system identification
Pandey and Barai (1995) describe use of the BP algorithm
Masri et al. (1993) describe neural networks as a power- for damage detection of steel-truss bridge structures. A sim-
ful tool for identification of structural dynamic systems. ilar study for vibration signature analysis of steel trusses is
Chen et al. (1995b) use the BP algorithm for identification discussed in Barai and Pandey (1995). Masri et al. (1996)
of structural dynamic models. The authors indicate “great explore the use of neural networks to detect changes in
130 Adeli

structural parameters during vibrations. Masri et al. (2000) using a two-degree-of-freedom dynamic system and report
describe application of neural networks to a nonparametric the latter is “far more effective.” Most control algorithms
structural damage detection methodology based on nonlin- are based on the availability of a complete state vector
ear system identification approaches. from measurement. Tang (1996b) uses the BP algorithm
High-strength bolts in spliced joints of steel bridges may as the state-vector estimator when only a limited number
become loose gradually during their lifetime. This problem of sensors are installed in the structure, and consequently,
has to be detected and corrected during periodic inspec- a complete state vector is not available. Bani-Hani and
tion and maintenance of the bridge. Mikami et al. (1998) Ghaboussi (1998) discuss nonlinear structural control using
present a system based on the BP algorithm to estimate the neural networks through numerical simulations on a two-
residual axial forces of high-strength bolts in steel bridges dimensional three-story steel frame considering its inelastic
using the reaction and acceleration waveforms collected material behavior.
by an automatic hammer or looseness detector. An impor- Ankireddi and Yang (1999) investigate the use of neu-
tant issue in structural health monitoring is selection of the ral networks for failure detection and accommodation in
members and locations of the structure to be monitored. structural control problems. They propose a failure detec-
Feng and Bahng (1999) use the BP algorithm to estimate tion neural network for monitoring structural responses and
the change in stiffness based on the measured vibration detecting performance-reducing sensor failures and a fail-
characteristics for damage assessment of reinforced con- ure accommodation neural network to account for the failed
crete columns retrofitted by advanced composite jackets. sensors using the Widrow-Hoff (Widrow and Lehr, 1995)
Kim et al. (2000b) describe a two-stage procedure where in training rule. Kim et al. (2000a) propose an optimal con-
the first stage traditional sensitivity analysis is used to rank trol algorithm using neural networks through minimization
and select critical members. In the second stage, the results of the instantaneous cost function for a single-degree-of-
of the sensitivity analysis and a trained neural network are freedom system. Hung et al. (2000) describe an active pulse
used to identify the optimal numbers and locations of mon- structural control using neural networks with a training
itoring sensors. The method is applied to two-dimensional algorithm that does not require the trial-and-error selection
trusses and multistory frames. of the learning ratio needed in the BP algorithm and present
simulation results for a small frame.
2.5 Structural control
Active control of structures has been an active area of 2.6 Finite element mesh generation
research in recent years (Adeli and Saleh, 1999). Ghaboussi In finite element analysis of structures, creating the right
and Joghataie (1995) present application of neural networks mesh is a tedious trial-and-error process often requiring a
in structural control. A neural network training algorithm, high level of human expertise. The accuracy and efficiency
a modified BP algorithm in this case, performs the role of of the method rely heavily on the selected mesh. Auto-
the control algorithm. The structure’s response, measured matic creation of an effective finite element mesh for a
at a selected number of points by sensors, and the actua- given problem has been an active area of research. Different
tor signals are the input to the neurocontroller. Its output is approaches have been explored in the literature, including
the subsequent value of the actuator signal to produce the neural networks. For a given number of nodes and mesh
desired actuator forces. The neurocontroller learns to con- topology, Manevitz et al. (1997) use the self-organizing
trol the structure after being trained by an emulator neu- algorithm of Kohonen (1988) to create a near-optimal finite
ral network. The authors suggest that neurocontrollers are element mesh for a two-dimensional domain using a com-
a potentially powerful tool in structural control problems bination of different types of elements. Bahreininejad et al.
based on simulation results for a three-story frame with one (1996) explore application of the BP and Hopfield neural
actuator. networks for finite element mesh partitioning. Pain et al.
Chen et al. (1995a) also describe use of the BP algo- (1999) present a neural network graph-partitioning algo-
rithm in structural control and present simulation results rithm for partitioning unstructured finite element meshes.
based on the model of an actual multistory apartment build- First, an automatic graph coarsening method is used to cre-
ing subjected to recorded earthquake ground motions. The ate a coarse mesh, followed by a mean field theorem neural
BP algorithm is also used by Tang (1996a) for active network to perform partitioning optimization.
control of a single-degree-of-freedom system and by Yen
(1996) for vibration control in flexible multibody dynamics.
2.7 Structural material characterization and
Nikzad et al. (1996) compare the performances of a conven-
modeling
tional feedforward controller and a neurocontroller based
on a modified BP algorithm in compensating the effects Ghaboussi et al. (1991) describe use of the BP neural
of the actuator dynamics and computational phase delay network for modeling behavior of conventional materials
 Neural networks in civil engineering: 1989–2000 131

such as concrete in the state of plane stress under mono- literature. Park and Adeli (1997b) present distributed neu-
tonic biaxial loading. Brown et al. (1991) demonstrate ral dynamics algorithms on the Cray T3D multiprocessor
the applicability of neural networks to composite mate- employing the work-sharing programming paradigm.
rial characterization. They use the BP algorithm to predict
hygral, thermal, and mechanical properties of composite-
ply materials. The BP algorithm also has been used for 3 CONSTRUCTION ENGINEERING
constitutive modeling of concrete (Sankarasubramanian and
Rajasekaran, 1996) and viscoplastic materials (Furukawa 3.1 Construction scheduling and management
and Yagawa, 1998).
Adeli and Karim (1997b) present a general mathemati-
Ghaboussi et al. (1998) present autoprogressive train-
cal formulation for scheduling of construction projects and
ing of neural network constitutive models using the global
load-deflection response measured in a structural test with apply it to the problem of highway construction scheduling.
application to laminated composites. In their approach, a Repetitive and nonrepetitive tasks, work-continuity consid-
partially trained neural network generates its own training erations, multiple-crew strategies, and the effects of vary-
cases through an iterative nonlinear finite element analy- ing job conditions on the performance of a crew can be
sis of the test specimen. Yeh (1999) uses the BP algorithm modeled. An optimization formulation is presented for the
to model the concrete workability in the design of a high- construction project scheduling problem with the goal of
performance concrete mixture. Neural networks are also minimizing the direct construction cost. The nonlinear opti-
used to model generalized hardening plasticity (Theocaris mization is then solved by the neural dynamics model of
and Panagiotopoulos, 1995), the alkali-silica reaction of Adeli and Park (1996). For any given construction duration,
concrete with admixtures (Li et al., 2000), and elastoplas- the model yields the optimal construction schedule for min-
ticity (Daoheng et al., 2000). imum construction cost automatically. By varying the con-
struction duration, one can solve the cost-duration tradeoff
problem and obtain the global optimal schedule and the cor-
2.8 Parallel neural network algorithms for
large-scale problems responding minimum construction cost. Karim and Adeli
(1999c) present an object-oriented information model for
The convergence speed of neural network learning mod- construction scheduling, cost optimization, and change-
els is slow. For large networks, several hours or even order management based on the new neural network–based
days of computer time may be required using the con- construction scheduling model of Adeli and Karim (1997b).
ventional serial workstations. A parallel BP learning algo- The model can be used by the owner/client who has to
rithm has been developed by Hung and Adeli (1993) and approve any change-order requests made by the contractor,
implemented on the Cray YMP supercomputer. A parallel- as well as by the contractor. The model provides support for
processing implementation of the BP algorithm on a schedule generation and review, cost estimation, and cost-
Transputer network with application to finite element mesh time tradeoff analysis. The model has been implemented
generation is also presented by Topping et al. (1997). in a prototype software system called CONSCOM (CON-
Optimization of large structures with thousands of mem- struction Scheduling, Cost Optimization, and Change-Order
bers subjected to actual constraints of commonly used Management) using Microsoft Foundation Classes under
codes requires an inordinate amount of computer pro- the Windows environment (Karim and Adeli, 1999d).
cessing time and high-performance computing resources
(Adeli and Kamal, 1993; Adeli, 1992a, 1992b; Adeli and 3.2 Construction cost estimation
Soegiarso, 1999). Park and Adeli (1997a) present a data
parallel neural dynamics model for discrete optimization Williams (1994) attempts to use the BP algorithm for pre-
of large steel structures implemented on a distributed- dicting changes in construction cost indexes for 1 and 6
memory multiprocessor, the massively parallel Connection months ahead but concludes that “the movement of the
Machine CM-5 system. The parallel algorithm has been cost indexes is a complex problem that cannot be predicted
applied to optimization of several high-rise and super- accurately by a BP neural network model.” Automating the
high-rise building structures, including a 144-story steel process of construction cost estimation based on objective
super-high-rise building structure with 20,096 members in data is highly desirable not only for improving the effi-
accordance with the AISC ASD and LRFD codes (AISC, ciency but also for removing the subjective questionable
1995, 1998) and subjected to multiple loading conditions human factors as much as possible. The costs of construc-
including wind loading according to the Uniform Build- tion materials, equipment, and labor depend on numerous
ing Code (UBC, 1997). This is by far the largest structural factors with no explicit mathematical model or rule for
optimization problem subjected to actual constraints of a price prediction. Adeli and Wu (1998) point out that “high-
widely used design code ever solved and reported in the way construction costs are very noisy, and the noise is the
132 Adeli

result of many unpredictable factors such as human judg- 3.5 Other applications of BP and other neural
ment factors, random market fluctuations, and weather con- network models in construction engineering
ditions.” They also discuss the problem of overfitting data, and management
noting that “because of the noise in the data, a perfect fit
Moselhi et al. (1991) were among the first to realize the
usually is not the best fit,” and underfitting results in poor
potential applications of neural networks in construction
generalization. Adeli and Wu (1998) present a regulariza-
engineering. They present an application of the BP algo-
tion neural network model and architecture for estimating
rithm for optimal markup estimation under different bid
the cost of construction projects. The model is applied to
conditions. They use a small set of 10 bid situations to
estimate the cost of reinforced concrete pavements as an
train the system but report up to 30,000 iterations for the
example. The new computational model is based on a solid
BP algorithm to converge with a small error. The BP algo-
mathematical foundation, making the cost estimation con-
rithm also has been used for selection of vertical concrete
sistently more reliable and predictable. Moreover, the prob-
formwork supporting walls and columns for a building site
lem of noise in the data is taken into account in a rational
(Kamarthi et al., 1992), for estimating construction produc-
manner.
tivity (Chao and Skibniewski, 1994; Sonmez and Rowings,
1998), for markup estimation using knowledge acquired
3.3 Resource allocation and scheduling
from contractors in Canada and the United States (Hegazy
Mohammad et al. (1995) formulate the problem of opti- and Moselhi, 1994), for evaluation of new construction
mally allocating available yearly budget to bridge reha- technology acceptability (Chao and Skibniewski, 1995), for
bilitation and replacement projects among a number of selection of horizontal concrete formwork to support slabs
alternatives as an optimization problem using the Hopfield and roofs (Hanna and Senouci, 1995), and for measur-
network (Hopfield, 1982, 1984). Savin et al. (1996, 1998) ing the level of organization effectiveness in a construc-
also discuss the use of a discrete-time Hopfield net tion firm.
in conjunction with an augmented Lagrangian multiplier Murtaza and Fisher (1994) describe the use of neu-
optimization algorithm for construction resource leveling. ral networks for decision making about construction
Elazouni et al. (1997) use the BP algorithm to estimate the modularization. Yeh (1995) uses a combination of simu-
construction resource requirements at the conceptual design lated annealing (Kirkpatrick et al., 1983) and a Hopfield
stage and apply the model to the construction of concrete neural network (Hopfield, 1982, 1984) to solve the
silo walls. construction-site layout problem. Kartam (1996) uses neu-
Senouci and Adeli (2001) present a mathematical model ral networks to determine optimal equipment combinations
for resource scheduling considering project scheduling for earthmoving operations. Pompe and Feelders (1997)
characteristics generally ignored in prior research, includ- use neural networks to predict corporate bankruptcy. Li
ing precedence relationships, multiple-crew strategies, and et al. (1999) discuss rule extractions from a neural network
the time-cost tradeoff. Previous resource scheduling formu- trained by the BP algorithm for construction markup esti-
lations traditionally have focused on project-duration min- mation in order to explain how a particular recommenda-
imization. The new model considers the total project cost tion is made.
minimization. Furthermore, resource leveling and resource-
constrained scheduling are performed simultaneously. The
model is solved using the neural dynamics optimization 4 NEURAL NETWORK APPLICATIONS IN
model of Adeli and Park (1996). OTHER CIVIL ENGINEERING FIELDS

3.4 Construction litigation 4.1 Environmental and water resources

engineering
Disputes and disagreements between the contractor and the
owner for reasons such as misinterpretation of the contract, Karunanithi et al. (1994) demonstrate the use of neu-
changes made by the owner or the contractor, differing site ral networks for river flow prediction using the cascade-
and weather conditions, labor problems, and unexpected correlation algorithm. The BP algorithm is used by Du
delays can lead to litigation. Arditi et al. (1998) use neu- et al. (1994) to predict the level of solubilization of six
ral networks to predict the outcome of construction litiga- heavy metals from sewage sludge using the bioleaching
tion. They use the outcomes of circuit and appellate court process, by Grubert (1995) to predict the flow conditions
decisions to train the network and report a successful pre- at the interface of stratified estuaries and fjords, by Kao
diction rate of 67 percent for the “extremely complex data and Liao (1996) to facilitate the selection of an appropri-
structure of court proceedings.” A comparison of the neural ate facility combination for municipal solid-waste incin-
network approach with case-based reasoning (CBR) for the eration, by Tawfik et al. (1997) to model stage-discharge
same problem is presented by Arditi and Tokdemir (1999). relationships at stream gauging locations at the Nile River,
 Neural networks in civil engineering: 1989–2000 133

by Deo et al. (1997) to interpolate the ocean wave heights feature map, and adaptive resonance theory (ART) model
over short intervals (weekly mean wave heights) from the two (ART2)—for the identification of incident patterns in
values obtained by remote sensing techniques and satellites traffic data. Faghri and Hua (1995) use ART model one
over long durations (a month), and by Liong et al. (2000) (ART1) to estimate the average annual daily traffic (AADT)
for water-level forecasting in Dhaka, Bangladesh. including the seasonal factors and compare its performance
Crespo and Mora (1995) describe neural network learn- with clustering and regression methods. They conclude that
ing for river streamflow estimation, prediction of carbon the neural network model yields better results than the other
dioxide concentration from a gas furnace, and a feed- two approaches. Dia and Rose (1997) use field data to
water control system in a boiling water reactor. Basheer test a multilayer perceptron neural network as an incident-
and Najjar (1996) use neural networks to model fixed- detection classifier. Eskandarian and Thiriez (1998) use
bed adsorber dynamics. Rodriguez and Serodes (1996) use neural networks to simulate a driver’s function of steering
the BP neural network to estimate the disinfectant dose and braking and develop a controller on a moving platform
adjustments required during water rechlorination in stor- (vehicle) encountering obstacles of various shapes. The sys-
age tanks based on representative operational and water- tem can generalize its learned patterns to avoid obstacles
quality historical data and conclude that the model “can and collisions. The BP neural network is used by Lingras
adequately mimic an operator’s know-how in the control of and Adamo (1996) to estimate average and peak hourly
the water quality within distribution systems.” Maier and traffic volumes, by Ivan and Sethi (1998) for traffic incident
Dandy (1997) discuss the use of neural networks for mul- detection, by Sayed and Abdelwahab (1998) for classifica-
tivariate forecasting problems encountered in the field of tion of road accidents for road improvements, and by Park
water resources engineering, including estimation of salin- and Rilett (1999) to predict the freeway link travel times
ity in a river. Thirumalaiah and Deo (1998) present neural for one through five time periods into the future.
networks for real-time forecasting of stream flows. Flood Saito and Fan (2000) present an optimal traffic signal
values during storms are forecast with a lead time of 1 hour timing model that uses the BP algorithm to conduct an
or more using the data from past flood values at a specific analysis of the level of service at a signalized intersection
location. Deo and Chaudhari (1998) use neural networks to by learning the complicated relationship between the traffic
predict tides at a station located in the interior of an estuary delay and traffic environment at signalized intersections.
or bay.
Gangopadhyay et al. (1999) integrate the BP algorithm 4.3 Highway engineering
with a Geographic Information System (GIS) for generation
of subsurface profiles and for identification of the distribu- Gagarin et al. (1994) discuss the use of a radial-Gaussian-
tion of subsurface materials. The model is applied to find based neural network for determining truck attributes such
the aquifer extent and its parameters for the multiaquifer as axle loads, axle spacing, and velocity from strain-
system under the city of Bangkok, Thailand. Coulibaly response readings taken from the bridges over which the
et al. (2000) use feedforward and recurrent neural networks truck is traveling. Eldin and Senouci (1995) describe the
for long-term forecasting of potential energy inflows for use of a BP algorithm for condition rating of roadway
hydropower operations planning. This is one of the few arti- pavements. They report very low average error when com-
cles addressing the problem of overfitting in neural network pared with a human expert determination. Cal (1995) uses
pattern recognition. The authors conclude that “the neu- the BP algorithm for soil classification based on three pri-
ral network–based models provide more accurate forecasts mary factors: plastic index, liquid limit water capacity,
than traditional stochastic models.” Liu and James (2000) and clay content. Razaqpur et al. (1996) present a com-
use the BP algorithm to estimate the discharge capacity in bined dynamic programming and Hopfield neural network
meandering compound (or two-stage) channels consisting (Hopfield, 1982, 1984) bridge-management model for effi-
of a main channel flanked by floodplains on one or both cient allocation of a limited budget to bridge projects over
sides. Guo (2001) presents a semivirtual watershed model a given period of time. The time dimension is modeled by
for small urban watersheds with a drainage area of less than dynamic programming, and the bridge network is simulated
150 acres using neural networks where the network train- by the neural network. Roberts and Attoh-Okine (1998) use
ing and the determination of the matrix of time-dependent a combination of supervised and self-organizing neural net-
weights to rainfall and runoff vectors is guided by the kine- works to predict the performance of pavements as defined
matic wave theory. by the International Roughness Index. The BP algorithm
is used by Owusu-Ababio (1998) for predicting flexible
pavement cracking and by Alsugair and Al-Qudrah (1998)
4.2 Traffic engineering
to develop a pavement-management decision support sys-
Cheu and Ritchie (1995) use three different neural net- tem for selecting an appropriate maintenance and repair
work architectures—multilayer perceptron, self-organizing action for a damaged pavement. Attoh-Okine (2001) uses
134 Adeli

the self-organizing map or competitive unsupervised learn- Yeh, 1998). As such, a number of other neural network
ing model of Kohonen (1988) for grouping of pavement- learning models have been proposed in recent years. Some
condition variables (such as the thickness and age of of them with applications in civil engineering are reviewed
pavement, average annual daily traffic, alligator cracking, briefly in this section.
wide cracking, potholing, and rut depth) to develop a model
for evaluation of pavement conditions. 5.2 Adaptive conjugate gradient neural
network algorithm
4.4 Geotechnical engineering
In an attempt to overcome the shortcomings of the BP algo-
A common method for evaluation of elastic moduli and rithm, Adeli and Hung (1994) have developed an adaptive
layer thicknesses of soils and pavements is the seismic conjugate gradient learning algorithm for training of multi-
spectral analysis of surface waves (SASW). Williams and layer feedforward neural networks. Powell’s modified con-
Gucunski (1995) use the BP algorithm to perform the inver- jugate gradient algorithm has been used with an approxi-
sion of SASW test results. Core penetration test (CPT) mate line search for minimizing the system error. The prob-
measurements are frequently used to find soil strength and lem of arbitrary trial-and-error selection of the learning and
stiffness parameters needed in design of foundations. Goh momentum ratios encountered in the momentum backprop-
(1995) demonstrates application of the BP algorithm for agation algorithm is circumvented in the new adaptive algo-
correlating various experimental parameters and evaluating rithm. Instead of constant learning and momentum ratios,
the CPT calibration chamber test data. The BP algorithm the step length in the inexact line search is adapted dur-
is used by Chikata et al. (1998) to develop a system for ing the learning process through a mathematical approach.
aesthetic evaluation of concrete retaining walls and by Teh Thus the new adaptive algorithm provides a more solid
et al. (1997) to estimate static capacity of precast reinforced mathematical foundation for neural network learning. The
concrete piles from dynamic stress wave data. Juang and algorithm has been applied to the domain of image recog-
Chen (1999) present neural network models for evaluating nition. It is shown that the adaptive neural networks algo-
the liquefaction potential of sandy soils. Use of neural net- rithm has a superior convergence property compared with
works to predict the collapse potential of soils is discussed the momentum BP algorithm.
by Juang et al. (1999).
After pointing out that “classical constitutive model- 5.3 Radial basis function neural networks
ing of geomaterials based on the elasticity and plasticity
The radial basis function neural network (RBFNN) learns
theories suffers from limitations pertaining to formulation
complexity, idealization of behavior, and excessive empiri- an input-output mapping by covering the input space with
cal parameters,” Basheer (2000) proposes neural networks basis functions that transform a vector from the input space
as an alternative for modeling the constitutive hysteresis to the output space (Moody and Darken, 1989; Poggio and
behavior of soils. He examines several mapping techniques Girosi, 1990). Conceptually, the RBFNN is an abstraction
to be used as frameworks for creating neural network mod- of the observation that biologic neurons exhibit a receptive
els for constitutive response of soils, including a hybrid field of activation such that the output is large when the
approach that provides high accuracy. input is closer to the center of the field and small when the
input moves away from the center. Structurally, the RBFNN
has a simple topology with a hidden layer of nodes hav-
5 SHORTCOMINGS OF THE BP ALGORITHMS ing nonlinear basis transfer functions and an output layer
AND OTHER RECENT APPROACHES of nodes with linear transfer functions (Adeli and Karim,
2000). The most common type of the basis function is
5.1 Shortcomings of the BP algorithm Gaussian. Yen (1994) proposes the use of radial basis func-
tion networks as a neurocontroller for vibration suppres-
The momentum BP learning algorithm (Rumelhart et al.,
sion. Amin et al. (1998) use the RBFNN to predict the flow
1986, Adeli and Hung, 1995) is widely used for training
of traffic. Jayawardena and Fernando (1998) present appli-
multilayer neural networks for classification problems. This
cation of the RBFNN for hydrologic modeling and runoff
algorithm, however, has a slow rate of learning. The num-
simulation in a small catchment and report that it is more
ber of iterations for learning an example is often in the
efficient computationally than the BP algorithm.
order of thousands and sometimes more than one hundred
thousands (Carpenter and Barthelemy, 1994). Moreover, the
5.4 Other approaches
convergence rate is highly dependent on the choice of the
values of learning and momentum ratios encountered in Masri et al. (1999) propose a stochastic optimization algo-
this algorithm. The proper values of these two parameters rithm based on adaptive random search techniques for
depend on the type of the problem (Adeli and Hung, 1994; training neural networks in applied mechanics applications.
 Neural networks in civil engineering: 1989–2000 135

Castillo et al. (2000a) present functional networks where unsupervised fuzzy neural network classification algorithm.
neural functions are learned instead of weights but apply The second stage is a supervised neural network learning
the concept to simple problems such as predicting the model using the classified clusters as training instances.
behavior of a cantilever beam and approximating the dif- The genetic algorithm is used in this stage to accelerate the
ferential equation for vibration of a simple single-degree- whole learning process in the hybrid learning algorithm.
of-freedom system with spring and viscous damping. Some The third stage is the process of defuzzification. The hybrid
learning methods in functional networks are presented in fuzzy neural network learning model has been applied to
Castillo et al. (2000b). the domain of image recognition. The performance of the
model has been evaluated by applying it to a large-scale
6 INTEGRATING NEURAL NETWORKS WITH training example with 2304 training instances.
OTHER COMPUTING PARADIGMS Hurson et al. (1994) discuss the use of fuzzy logic in
automating knowledge acquisition in a neural network–
6.1 Genetic algorithms based decision support system. Anantha Ramu and Johnson
(1995) present a fuzzy logic–BP neural network approach
Hung and Adeli (1991b) present a hybrid learning to detect, classify, and estimate the extent of damage from
algorithm by integrating a genetic algorithm with error the measured vibration response of composite laminates.
backpropagation multilayer neural networks. The algorithm Kasperkiewicz et al. (1995) use a fuzzy ART neural net-
consists of two learning stages. The first learning stage is work (Carpenter et al., 1991) to predict strength properties
to accelerate the learning process by using a genetic algo- of high-performance concrete mixes as a factor of six com-
rithm with the feedforward step of the BP algorithm. In ponents: cement, silica, superplasticizer, water, fine aggre-
this stage, the weights of the neural network are encoded gate, and coarse aggregate.
on chromosomes as decision variables. The objective func- Furuta et al. (1996) describe a fuzzy expert system for
tion for the genetic algorithm is defined as the average damage assessment of reinforced concrete bridge decks
squared system error. After performing several iterations using genetic algorithms and neural networks. The goal is
and meeting the stopping criterion, the first learning stage to automatically acquire fuzzy production rules through use
is terminated, and the chromosome returning the minimum of the genetic algorithm and the BP neural networks. The
objective function is considered as the initial weights of weights of the links obtained from the neural networks are
the neural network in the second stage. Next, the BP algo- used in the genetic algorithm evaluation function to obtain
rithm performs the second learning process until the termi- the optimal combination of rules to be used in the knowl-
nal condition is satisfied. edge base of the expert system (Adeli, 1988; Adeli and
Moselhi et al. (1993) use the BP neural networks and Balasubramanyam, 1988; Adeli, 1990a, 1990b).
the genetic algorithm (Adeli and Hung, 1995) to develop Ni et al. (1996) present a fuzzy neural network approach
a decision support system to aid contractors in preparing for evaluating the stability of natural slopes considering
bids. A parallel genetic–neural network algorithm is also the geologic, topographic, meteorologic, and environmental
presented by Hung and Adeli (1994b). Jinghui et al. (1996), conditions that can be described mostly in linguistic terms.
Hajela and Lee (1997), and Papadrakakis et al. (1998) use Parameters of the neural networks are represented by fuzzy
the BP algorithm to improve the efficiency of genetic algo- sets (Zadeh, 1970, 1978). Faravelli and Yao (1996) discuss
rithms for structural optimization problems. Topping et al. the use of neural networks in fuzzy control of structures.
(1998) present parallel finite element analysis on a MIMD Rajasekaran et al. (1996) describe the integration of fuzzy
distributed computer. They describe a mesh partitioning logic and neural networks for a prestressed concrete pile
technique for planar finite element meshes where a BP diagnosis problem and concrete mix design. Hung and Jan
neural network is used to find the approximate number of (1999b) present a fuzzy neural network learning model con-
elements within a coarse mesh. The coarse mesh is then sisting of both supervised and unsupervised learning and
divided into several subdomains using a genetic algorithm apply it to simply supported concrete and steel beam design
optimization approach. problems. Sayed and Razavi (2000) combine fuzzy logic
with an adaptive B-spline network to model the behavioral
6.2 Fuzzy logic mode choice in the area of transportation planning. They
apply the model to a bimodal example for shipment of
Adeli and Hung (1993) present a fuzzy neural network
commodities (rail and Interstate Commerce Commission–
learning model by integrating an unsupervised fuzzy neural
regulated motor carriers for shipments over 500 lb).
network classification algorithm with a genetic algorithm
and the adaptive conjugate gradient neural network learn-
6.3 Wavelets
ing algorithm. The learning model consists of three major
stages. The first stage is used to classify the given train- Neural network models can lose their effectiveness when
ing instances into a small number of clusters using the the patterns are very complicated or noisy. Traffic data
136 Adeli

collected from loop detectors installed in a freeway system • Journal of Construction Engineering and Management,
and transmitted to a central station present such patterns. ASCE (1991, 10)
Neural networks have been used to detect incident pat- • Journal of Structural Engineering, ASCE (1995, 9)
terns from nonincident patterns with limited success. The • Canadian Journal of Civil Engineering (1994, 8)
dimensionality of the training input data is high, and the • Computer Methods in Applied Mechanics and
embedded incident characteristics are not easily detectable. Engineering (1993, 7)
Adeli and Samant (2000) present a computational model
for automatic traffic incident detection using the discrete
ACKNOWLEDGMENTS
wavelet transform (Samant and Adeli, 2000) and neural net-
works. The wavelet transform is used for feature extraction, The author’s research in recent years has been sponsored
denoising, and effective preprocessing of data before the by the National Science Foundation, American Iron and
adaptive conjugate gradient neural network model of Adeli Steel Institute, American Institute of Steel Construction, the
and Hung (1994) is used to make the traffic incident detec- Ohio Department of Transportation, the Ohio Supercom-
tion. The authors show that for incidents with a duration of puter Center, and the Federal Highway Administration.
more than 5 minutes, the incident-detection model yields a
detection rate of nearly 100 percent and false-alarm rate of
about 1 percent for two- or three-lane freeways. REFERENCES
Adeli and Karim (2000) present a new multiparadigm
intelligent system approach to the traffic incident-detection Abdalla, K. M. & Stavroulakis, G. E. (1995), A backpropagation
problem through integration of fuzzy, wavelet, and neural neural network model for semi-rigid steel connections, Micro-
computing techniques to improve reliability and robustness. computers in Civil Engineering, 10 (2), 77–87.
Adeli, H., ed. (1988), Expert Systems in Construction and Struc-
A wavelet-based denoising technique is employed to elim-
tural Engineering, Chapman and Hall, London.
inate undesirable fluctuations in observed data from traffic
Adeli, H., ed. (1990a), Knowledge Engineering, Vol. 1: Funda-
sensors. Fuzzy c-mean clustering is used to extract signif- mentals, McGraw-Hill, New York.
icant information from the observed data and to reduce its Adeli, H., ed. (1990b), Knowledge Engineering, Vol. 2: Applica-
dimensionality. A radial basis function neural network is tions, McGraw-Hill, New York.
developed to classify the denoised and clustered observed Adeli, H., ed. (1992a), Parallel Processing in Computational
data. The authors report excellent incident-detection rates Mechanics, Marcel Dekker, New York.
with no false alarms when tested using both real and sim- Adeli, H., ed. (1992b), Supercomputing in Engineering Analysis,
ulated data. Marcel Dekker, New York.
Liew and Wang (1998) describe application of wavelets Adeli, H., ed. (1994), Advances in Design Optimization, E. & F.
N. Spon, London.
for crack identification in structures. Marwala (2000) uses
Adeli, H. & Balasubramanyam, K. V. (1988), Expert Sys-
the wavelet transform and neural networks for damage tems for Structural Design: A New Generation, Prentice-Hall,
identification in structures. Englewood Cliffs, NJ.
Adeli, H. & Hung, S.-L. (1993), A fuzzy neural network learn-
ing model for image recognition, Integrated Computer-Aided
7 FINAL COMMENTS Engineering, 1 (1), 43–55.
Adeli, H. & Hung, S.-L. (1994), An adaptive conjugate gradi-
The neural networks articles reviewed in this article have ent learning algorithm for effective training of multilayer neu-
been published mostly in the following journals (the first ral networks, Applied Mathematics and Computation, 62 (1),
number in the parentheses refers to the year when the first 81–102.
Adeli, H. & Hung, S.-L. (1995), Machine Learning: Neural
article on neural networks was published in that particular
Networks, Genetic Algorithms, and Fuzzy Systems, Wiley,
journal, and the second number refers to the number of New York.
articles published and reviewed in this article). This list will Adeli, H. & Kamal, O. (1993), Parallel Processing in Structural
guide readers on where to find additional articles on the Engineering, Elsevier, London.
subject in the future. Adeli, H. & Karim, A. (1997a), Neural dynamics model for opti-
mization of cold-formed steel beams, Journal of Structural
• Computer-Aided Civil and Infrastructure Engineer-
Engineering, ASCE, 123 (11), 1535–43.
ing (formerly Microcomputers in Civil Engineering, Adeli, H. & Karim, A. (1997b), Scheduling/cost optimization and
1986–1997) (1989, 52) neural dynamics model for construction, Journal of Construc-
• Journal of Computing in Civil Engineering, ASCE tion Management and Engineering, ASCE, 123 (4), 450–8.
(1992, 36) Adeli, H. & Karim, A. (2000), Fuzzy-wavelet RBFNN model
• Computers and Structures (1991, 24) for freeway incident detection, Journal of Transportation Engi-
• Journal of Engineering Mechanics, ASCE (1991, 13) neering, ASCE, 126 (6), 464–71.
 Neural networks in civil engineering: 1989–2000 137

Adeli, H. & Kumar, S. (1999), Distributed Computer-Aided Engi- minor axis steel connections, Computers and Structures, 63 (4),
neering for Analysis, Design, and Visualization, CRC Press, 685–92.
Boca Raton, FL. Anderson, J. A. (1995), An Introduction to Neural Networks, A
Adeli, H. & Park, H. S. (1995a), A neural dynamics model for Bradford Book, MIT Press, Cambridge, MA.
structural optimization: Theory, Computers and Structures, 57 Ankireddi, S. & Yang, H. T. Y. (1999), Neural networks for sen-
(3), 391–9. sor fault correction in structural control, Journal of Structural
Adeli, H. & Park, H. S. (1995b), Optimization of space structures Engineering, 125 (9), 1056–64.
by neural dynamics model, Neural Networks, 8 (5), 769–81. Arbib, M. A. (1995), The Handbook of Brain Theory and Neural
Adeli, H. & Park, H. S. (1995c), Counterpropagation neural net- Networks, MIT Press, Cambridge, MA.
works in structural engineering, Journal of Structural Engi- Arditi, D., Oksay, F. E. & Tokdemir, O. B. (1998), Predicting
neering, ASCE, 121 (8), 1205–12. the outcome of construction litigation using neural networks,
Adeli, H. & Park, H. S. (1996), Hybrid CPN-neural dynamics Computer-Aided Civil and Infrastructure Engineering, 13 (2),
model for discrete optimization of steel structures, Microcom- 75–81.
puters in Civil Engineering, 11 (5), 355–66. Arditi, D. & Tokdemir, O. B. (1999), Comparison of case-based
Adeli, H. & Park, H. S. (1998), Neurocomputing in Design reasoning and artificial neural networks, Journal of Computing
Automation, CRC Press, Boca Raton, FL. in Civil Engineering, ASCE, 13 (3), 162–9.
Adeli, H. & Saleh, A. (1999), Control, Optimization, and Smart Arslan, M. A. & Hajela, P. (1997), Counterpropagation neural net-
Structures: High-Performance Bridges and Buildings of the works in decomposition based optimal design, Computers and
Future, Wiley, New York. Structures, 65 (5), 641–50.
Adeli, H. & Samant, A. (2000), An adaptive conjugate gradi- Attoh-Okine, N. O. (2001), Grouping pavement condition vari-
ent neural network-wavelet model for traffic incident detection, ables for performance modeling using self-organizing maps,
Computer-Aided Civil and Infrastructure Engineering, 15 (4), Computer-Aided Civil and Infrastructure Engineering, 16 (2),
251–60. 112–25.
Adeli, H. & Soegiarso, R. (1999), High-Performance Computing
Bahreininejad, A., Topping, B. H. V. & Khan, A. I. (1996), Finite
in Structural Engineering, CRC Press, Boca Raton, FL.
element mesh partitioning using neural networks, Advances in
Adeli, H. & Wu, M. (1998), Regularization neural network for
Engineering Software, 27, 103–15.
construction cost estimation, Journal of Construction Engineer-
Bani-Hani, K. & Ghaboussi, J. (1998), Nonlinear structural con-
ing and Management, ASCE, 124 (1), 18–24.
trol using neural networks, Journal of Engineering Mechanics,
Adeli, H. & Yeh, C. (1989), Perceptron learning in engineering
124 (3), 319–27.
design, Microcomputers in Civil Engineering, 4 (4), 247–56.
Barai, S. V. & Pandey, P. C. (1995), Vibration signature analysis
Adeli, H. & Zhang, J. (1993), An improved perceptron learning
using artificial neural networks, Journal of Computing in Civil
algorithm, Neural, Parallel, and Scientific Computations, 1 (2),
Engineering, ASCE, 9 (4), 259–65.
141–52.
Basheer, I. A. (2000), Selection of methodology for neural net-
AISC (1995), Manual of Steel Construction: Allowable Stress
work modeling of constitutive hysteresis behavior of soils,
Design, 9th ed., 2d rev., American Institute of Steel Construc-
tion, Chicago. Computer-Aided Civil and Infrastructure Engineering, 15 (6),
AISI (1996), Specification for the Design of Cold-Formed 445–63.
Steel Structural Members, American Iron and Steel Institute, Basheer, I. A. & Najjar, Y. M. (1996), Predicting dynamic
Washington, DC. response of adsorption columns with neural nets, Journal of
AISI (1997), Cold-Formed Steel Design Manual, American Iron Computing in Civil Engineering, ASCE, 10 (1), 31–9.
and Steel Institute, Washington, DC. Berke, L., Patnaik, S. N. & Murthy, P. L. N. (1993), Optimum
AISC (1998), Manual of Steel Construction: Load and Resistance design of aerospace structural components using neural net-
Factor Design, Vol. 1, 2d ed., 2d rev., American Institute of works, Computers and Structures, 48 (6), 1001–10.
Steel Construction, Chicago, Illinois. Biedermann, J. D. (1997), Representing design knowledge with
Aleksander, I. & Morton, H. (1993), Neurons and Symbols: The neural networks, Computer-Aided Civil and Infrastructure
Stuff That Mind Is Made of, Chapman and Hall, London. Engineering, 12 (4), 277–85.
Alsugair, A. M. & Al-Qudrah, A. A. (1998), Artificial neural net- Brown, D. A., Murthy, P. L. N. & Berke, L. (1991), Computa-
work approach for pavement maintenance, Journal of Comput- tional simulation of composite ply micromechanics using arti-
ing in Civil Engineering, ASCE, 12 (4), 249–55. ficial neural networks, Microcomputers in Civil Engineering,
Amin, S. M., Rodin, E. Y. & Garcia-Ortiz, A. (1998), Traffic pre- 6 (2), 87–97.
diction and management via RBF neural nets and semantic Cal, Y. (1995), Soil classification by neural network, Advances in
control, Computer-Aided Civil and Infrastructure Engineering, Engineering Software, 22, 95–7.
13 (5), 315–27. Cao, X., Sugiyama, Y. & Mitsui, Y. (1998), Application of arti-
Anantha Ramu, S. & Johnson, V. T. (1995), Damage assessment of ficial neural networks to load identification, Computers and
composite structures: A fuzzy logic integrated neural network Structures, 69, 63–78.
approach, Computers and Structures, 57 (3), 491–502. Carpenter, G. A., Grossberg, S. & Rosen, D. B. (1991), Fuzzy
Anderson, D., Hines, E. L., Arthur, S. J. & Eiap, E. L. (1997), ART: Fast stable learning of analog patterns by an adaptive
Application of artificial neural networks to the prediction of resonance system, Neural Networks, 4, 759–71.
138 Adeli

Carpenter, W. C. & Barthelemy, J.-F. (1994), Common miscon- sludge: A neural-net-based approach, Canadian Journal of
ceptions about neural networks as approximators, Journal of Civil Engineering, 21, 728–35.
Computing in Civil Engineering, ASCE, 8 (3), 345–58. Duan, L. & Chen, W. F. (1989), Effective length factor for
Castillo, E., Cobo, A., Gutierrez, J. M. & Pruneda, E. (2000a), columns in unbraced frames, Journal of Structural Engineer-
Functional networks: A new network–based methodology, ing, ASCE, 115 (1), 149–65.
Computer-Aided Civil and Infrastructure Engineering, 15 (2), Elazouni, A. M., Nosair, I. A., Mohieldin, Y. A. & Mohamed,
90–106. A. G. (1997), Estimating resource requirements at conceptual
Castillo, E., Gutierrez, J. M., Cobo, A. & Castillo, C. (2000b), design stage using neural networks, Journal of Computing in
Some learning methods in functional networks, Computer- Civil Engineering, ASCE, 11 (4), 217–23.
Aided Civil and Infrastructure Engineering, 15 (6), 426–38. Eldin, N. N. & Senouci, A. B. (1995), A pavement condition rating
Cattan, J. & Mohammadi, J. (1997), Analysis of bridge condi- model using backpropagation neural network, Microcomputers
tion rating using neural networks, Computer-Aided Civil and in Civil Engineering, 10 (6), 433–41.
Infrastructure Engineering, 12 (6), 419–29. Elkordy, M. F., Chang, K. C. & Lee, G. C. (1994), A structural
Chao, L. & Skibniewski, M. J. (1994), Estimating construc- damage neural network monitoring system, Microcomputers in
tion productivity: Neural network–based approach, Journal of Civil Engineering, 9 (2), 83–96.
Computing in Civil Engineering, 8 (2), 234–51. Eskandarian, A. & S. Thiriez, S. (1998), Collision avoidance using
Chao, L. & Skibniewski, M. J. (1995), Neural network method cerebral model arithmetic computer neural network, Computer-
of estimating construction technology acceptability, Journal of Aided Civil and Infrastructure Engineering, 13 (5), 303–14.
Construction Engineering and Management, 121 (1), 130–42. Faghri, A. & Hua, J. (1995), Roadway seasonal classification
Chen, H. M., Qi, G. Z., Yang, J. C. S. & Amini, F. (1995b), Neural using neural networks, Journal of Computing in Civil Engi-
network for structural dynamic model identification, Journal of neering, ASCE, 9 (3), 209–15.
Engineering Mechanics, 121 (12), 1377–81. Faravelli, L. T. & Yao, T. (1996), Use of adaptive networks in
Chen, H. M., Tsai, K. H., Qi, G. Z., Yang, J. C. S. & Amini, F. fuzzy control of civil structures, Microcomputers in Civil Engi-
(1995a), Neural network for structure control, Journal of Com- neering, 11 (1), 67–76.
puting in Civil Engineering, ASCE, 9 (2), 168–76.
Feng, M. Q. & Bahng, E. Y. (1999), Damage assessment of jack-
Cheu, R. L. & Ritchie, S. G. (1995), Automated detection of lane-
eted RC columns using vibration tests, Journal of Structural
blocking freeway incidents using artificial neural networks,
Engineering, ASCE, 125 (3), 265–71.
Transportation Research C, 3 (6), 371–88.
Furukawa, T. & Yagawa, G. (1998), Implicit constitutive modeling
Chikata, Y., Yasuda, N., Matsushima, M. & Kobori, T. (1998),
for viscoplasticity using neural networks, International Journal
Inverse analysis of aesthetic evaluation of planted concrete
for Numerical Methods in Engineering, 43, 195–219.
structures by neural networks, Computer-Aided Civil and
Furuta, H., He, J. & Watanabe, E. (1996), A fuzzy expert sys-
Infrastructure Engineering, 13 (4), 255–64.
tem for damage assessment using genetic algorithms and neural
Chuang, P. H., Goh, A. T. C. & Wu, X. (1998), Modeling
networks, Microcomputers in Civil Engineering, 11 (1), 37–45.
the capacity of pin-ended slender reinforced concrete columns
Gagarin, N., Flood, I. & Albrecht, P. (1994), Computing truck
using neural networks, Journal of Structural Engineering,
ASCE, 124 (7), 830–8. attributes with artificial neural networks, Journal of Computing
Consolazio, G. R. (2000), Iterative equation solver for bridge anal- in Civil Engineering, ASCE, 8 (2), 179–200.
ysis using neural networks, Computer-Aided Civil and Infras- Gangopadhyay, S., Gautam, T. R. & Gupta, A. D. (1999), Subsur-
tructure Engineering, 15 (2), 107–19. face characterization using artificial neural network and GIS,
Coulibaly, P., Anctil, F. & Bobee, B. (2000), Neural network– Journal of Computing in Civil Engineering, ASCE, 13 (3),
based long term hydropower forecasting system, Computer- 153–61.
Aided Civil and Infrastructure Engineering, 15 (5), 355–64. Ghaboussi, J., Garrett, J. H. & Wu, X. (1991), Knowledge-based
Crespo, J. L. & Mora, E. (1995), Neural networks as inference and modeling of material behavior with neural networks, Journal
learning engines, Microcomputers in Civil Engineering, 10 (2), of Engineering Mechanics, ASCE, 117 (1), 132–53.
89–96. Ghaboussi, J. & Joghataie, A. (1995), Active control of structure
Daoheng, S., Qiao, H. & Hao, X. (2000), A neurocomputing using neural networks, Journal of Engineering Mechanics, 121
model for the elastoplasticity, Computer Methods in Applied (4), 555–67.
Mechanics and Engineering, 182, 177–86. Ghaboussi, J., Pecknold, D. A., Zhang, M. & Haj-Ali, R. M.
Deo, M. C. & Chaudhari, G. (1998), Tide prediction using neural (1998), Autoprogressive training of neural network constitutive
networks, Computer-Aided Civil and Infrastructure Engineer- models, International Journal for Numerical Methods in Engi-
ing, 13 (2), 113–20. neering, 42, 105–26.
Deo, M. C., Venkat Rao, V. & Sakar, A. (1997), Neural net- Goh, A. T. C. (1995), Neural networks for evaluating CPT calibra-
works for wave height interpolation, Microcomputers in Civil tion chamber test data, Microcomputers in Civil Engineering,
Engineering, 12 (3), 217–25. 10 (2), 147–51.
Dia, H. & Rose, G. (1997), Development and evaluation of neural Golden, R. M. (1996), Mathematical Methods for Neural Network
network freeway incident detection models using field data, Analysis and Design, MIT Press, Cambridge, MA.
Transportation Research C, 5 (5), 313–31. Grubert, J. P. (1995), Prediction of estuarine instabilities with arti-
Du, Y. G., Sreekrishnan, T. R., Tyagi, R. D. & Campbell, P. G. C. ficial neural networks, Journal of Computing in Civil Engineer-
(1994), Effect of pH on metal solubilization from sewage ing, ASCE, 9 (4), 266–74.
 Neural networks in civil engineering: 1989–2000 139

Gunaratnam, D. J. & Gero, J. S. (1994), Effect of representation Hung, S. L. & Adeli, H. (1994b), A parallel genetic/neural net-
on the performance of neural networks in structural engineering work algorithm for MIMD shared memory machines, IEEE
applications, Microcomputers in Civil Engineering, 9 (2), 97– Transactions on Neural Networks, 5 (6), 900–9.
108. Hung, S. L. & Jan, J. C. (1999a), MS CMAC neural network
Guo, J. C. Y. (2001), A semivirtual watershed model by neural learning model in structural engineering, Journal of Computing
networks, Computer-Aided Civil and Infrastructure Engineer- in Civil Engineering, ASCE, 13 (1), 1–11.
ing, 16 (2), 106–11. Hung, S. L. & Jan, J. C. (1999b), Machine learning in engineer-
Hajela, P. & Berke, L. (1991), Neurobiological computational ing analysis and design: An integrated fuzzy neural network
models in structural analysis and design, Computers and Struc- learning model, Computer-Aided Civil and Infrastructure Engi-
tures, 41 (4), 657–67. neering, 14 (3), 207–19.
Hajela, P. & Lee, E. (1997), Topological optimization of rotorcraft Hung, S. L., Kao, C. Y. & Lee, J. C. (2000), Active pulse structural
subfloor structures for crashworthiness considerations, Comput- control using artificial neural networks, Journal of Engineering
ers and Structures, 64 (1–4), 65–76. Mechanics, 126 (8), 839–49.
Hanna, A. S. & Senouci, A. B. (1995), NEUROSLAB: Neural Hurson, A. R., Pakzad, S. & Jin, B. (1994), Automated knowledge
network system for horizontal formwork selection, Canadian acquisition in a neural network–based decision support sys-
Journal of Civil Engineering, 22, 785–92. tem for incomplete database systems, Microcomputers in Civil
Haykin, S. (1999), Neural Networks: A Comprehensive Founda- Engineering, 9 (2), 129–43.
tion, 2d ed., Prentice-Hall, Englewood Cliffs, NJ. Ivan, J. N. & Sethi, V. (1998), Data fusion of fixed detector and
Hecht-Nielsen, R. (1987a), Counterpropagation networks, in Pro- probe vehicle data for incident detection, Computer-Aided Civil
ceedings of the IEEE 1st International Conference on Neural and Infrastructure Engineering, 13 (5), 329–37.
Networks, Vol. II, IEEE Press, New York, pp. 19–32. Jayawardena, A. W. & Fernando, D. A. K. (1998), Use of radial
Hecht-Nielsen, R. (1987b), Counterpropagation networks, Applied basis function type artificial neural networks for runoff simula-
Optics, 26 (23), 4979–85. tion, Computer-Aided Civil and Infrastructure Engineering, 13
Hecht-Nielsen, R. (1988), Application of counterpropagation net- (2), 91–9.
Jenkins, W. M. (1999), A neural network for structural re-analysis,
works, Neural Networks, 1 (2), 131–9.
Computers and Structures, 72, 687–98.
Hegazy, T. & Moselhi, O. (1994), Analogy-based solution to
Jingui, L., Yunliang, D., Bin, W. & Shide, X. (1996), An improved
markup estimation problem, Journal of Computing in Civil
strategy for GAs in structural optimization, Computers and
Engineering, ASCE, 8 (1), 72–87.
Structures, 61 (6), 1185–91.
Hegazy, T., Fazio, P. & Moselhi, O. (1994), Developing practical
Juang, C. H. & Chen, C. J. (1999), CPT-based liquefaction eval-
neural network applications using backpropagation, Microcom-
uation using artificial neural networks, Computer-Aided Civil
puters in Civil Engineering, 9 (2), 145–59.
and Infrastructure Engineering, 14 (3), 221–9.
Hegazy, T., Tully, S. & Marzouk, H. (1998), A neural network
Juang, C. H., Ni, S. H. & Lu, P. C. (1999), Training artificial
approach for predicting the structural behavior of concrete
neural networks with the aid of fuzzy sets, Computer-Aided
slabs, Canadian Journal of Civil Engineering, 25, 668–77.
Civil and Infrastructure Engineering, 14 (6), 407–15.
Hoit, M., Stoker, D. & Consolazio, G. (1994), Neural networks Kamarthi, S. V., Sanvido, V. E. & Kumara, S. R. T. (1992), Neuro-
for equation renumbering, Computers and Structures, 52 (5), form: Neural network system for vertical formwork selection,
1011–21. Journal of Computing in Civil Engineering, ASCE, 6 (2), 178–
Hopfield, J. J. (1982), Neural networks and physical systems with 9.
emergent collective computational abilities, Proceedings of the Kang, H.-T. & Yoon, C. J. (1994), Neural network approaches
National Academy of Sciences, 79, 2554–8. to aid simple truss design problems, Microcomputers in Civil
Hopfield, J. J. (1984), Neurons with graded response have collec- Engineering, 9 (3), 211–18.
tive computational properties like those of two-state neurons, Kao, J.-J. & Liao, Y. Y. (1996), IAC network for composition
Proceedings of the National Academy of Sciences, 81, 3088–92. of waste incineration facility, Journal of Computing in Civil
Huang, C.-C. & Loh, C. H. (2001), Nonlinear identification of Engineering, ASCE, 10 (2), 168–71.
dynamic systems using neural networks, Computer-Aided Civil Karim, A. & Adeli, H. (1999a), Global optimum design of cold-
and Infrastructure Engineering, 16 (1), 28–41. formed steel Z-shape beams, Practice Periodicals on Structural
Hung, S. L. & Adeli, H. (1991a), A model of perceptron learning Design and Construction, ASCE, 4 (1), 17–20.
with a hidden layer for engineering design, Neurocomputing, Karim, A. & Adeli, H. (1999b), Global optimum design of cold-
3, 3–14. formed steel hat-shape beams, Thin-Walled Structures, 35 (4),
Hung, S. L. & Adeli, H. (1991b), A hybrid learning algorithm for 275–88.
distributed memory multicomputers, Heuristics: The Journal of Karim, A. & Adeli, H. (1999c), Object-oriented information
Knowledge Engineering, 4 (4), 58–68. model for construction project management, Journal of Con-
Hung, S. L. & Adeli, H. (1993), Parallel backpropagation learning struction Engineering and Management, ASCE, 125 (5),
algorithm on cray YMP8/864 supercomputer, Neurocomputing, 361–7.
5, 287–302. Karim, A. & Adeli, H. (1999d), CONSCOM: An OO con-
Hung, S. L. & Adeli, H. (1994a), Object-oriented backpropagation struction scheduling and change management system, Journal
and its application to structural design, Neurocomputing, 6 (1), of Construction Engineering and Management, ASCE, 125 (5),
45–55. 368–76.
140 Adeli

Karim, A. & Adeli, H. (2000), Global optimum design of cold- Marwala, T. (2000), Damage identification using committee of
formed steel I-shape beams, Practice Periodicals on Structural neural networks, Journal of Engineering Mechanics, 126 (1),
Design and Construction, ASCE, 5 (2), 78–81. 43–50.
Kartam, N. (1996), Neural network: Spreadsheet integration for Masri, S. F., Chassiakos, A. G. & Caughey, T. K. (1993), Identi-
earthmoving operations, Microcomputers in Civil Engineering, fication of nonlinear dynamic systems using neural networks,
11 (4), 283–8. Journal of Applied Mechanics, ASME, 60, 123–33.
Karunanithi, N., Grenney, W. J., Whitley, D. & Bovee, K. (1994), Masri, S. F., Nakamura, M., Chassiakos, A. G. & Caughey, T.
Neural networks for river flow prediction, Journal of Comput- K. (1996), Neural network approach to detection of changes
ing in Civil Engineering, ASCE, 8 (2), 201–20. in structural parameters, Journal of Engineering Mechanics,
Kasperkiewicz, J., Racz, J. & Dubrawski, A. (1995), HPC strength 122 (4), 350–60.
prediction using artificial neural networks, Journal of Comput- Masri, S. F., Smyth, A. W., Chassiakos, A. G., Caughey, T. K. &
ing in Civil Engineering, ASCE, 9 (4), 279–84. Hunter, N. F. (2000), Application of neural networks for detec-
Kim, J. T., Jung, H. J. & Lee, I. W. (2000a), Optimal structural tion of changes in nonlinear systems, Journal of Engineering
control using neural networks, Journal of Engineering Mechan- Mechanics, 126 (7), 666–76.
ics, 126 (2), 201–5. Masri, S. F., Smyth, A. W., Chassiakos, A. G., Nakamura, M. &
Kim, S. H., Yoon, C. & Kim, B. J. (2000b), Structural monitor- Caughey, T. K. (1999), Training neural networks by adaptive
ing system based on sensitivity analysis and a neural network, random search techniques, Journal of Engineering Mechanics,
Computer-Aided Civil and Infrastructure Engineering, 15 (4), 125 (2), 123–32.
309–18. Mathew, A., Kumar, B., Sinha, B. P. & Pedreschi, R. F. (1999),
Kirkpatrick, S., Gelatt, C. & Vecchi, M. (1983), Optimization by Analysis of masonry panel under biaxial bending using ANNs
simulated annealing, Science, 220, 671–80. and CBR, Journal of Computing in Civil Engineering, ASCE,
Kishi, N., Chen, W. F. & Goto, Y. (1997), Effective length fac- 13 (3), 170–7.
tor of columns in semirigid and unbraced frames, Journal of Mehrotra, K., Mohan, C. K. & Ranka, S. (1997), Elements of
Structural Engineering, ASCE, 123 (3), 313–20. Artificial Neural Networks, MIT Press, Cambridge, MA.
Kohonen, T. 1988, Self-Organization and Associative Memory, Messner, J. I., Sanvido, V. E. & Kumara, S. R. T. (1994), Struct-
Springer-Verlag, Berlin. Net: A neural network for structural system selection, Micro-
Kushida, M., Miyamoto, A. & Kinoshita, K. (1997), Development computers in Civil Engineering, 9 (2), 109–18.
of concrete bridge rating prototype expert system with machine Mikami, I., Tanaka, S. & Hiwatashi, T. (1998), Neural network
learning, Journal of Computing in Civil Engineering, ASCE, system for reasoning residual axial forces of high-strength bolts
11 (4), 238–47. in steel bridges, Computer-Aided Civil and Infrastructure Engi-
Li, H., Shen, L. Y. & Love, P. D. E. (1999), ANN-based mark-up neering, 13 (4), 237–46.
estimation system with self-explanatory capacities, Journal of Mohammed, H. A., Abd El Halim, A. O. & Razaqpur, A. G.
Construction Engineering and Management, 125 (3), 185–9. (1995), Use of neural networks in bridge management sys-
Li, S. (2000), Global flexibility simulation and element stiff- tems, Transportation Research Record No. 1490, Washington,
ness simulation in finite element analysis with neural network, pp. 1–8.
Computer Methods in Applied Mechanics and Engineering, Moody, J. & Darken, C. J. (1989), Fast learning in networks of
186, 101–8. locally-tuned processing units, Neural Computation, 1, 281–94.
Li, Z., Mu, B. & Peng, J. (2000), Alkali-silica reaction of concrete Moselhi, O., Hegazy, T. & Fazio, P. (1991), Neural networks as
with admixtures, Journal of Engineering Mechanics, 126 (3), tools in construction, Journal of Construction Engineering and
243–49. Management, 117 (4), 606–25.
Liew, K. M. & Wang, Q. (1998), Application of wavelet theory Moselhi, O., Hegazy, T. & Fazio, P. (1993), DBID: Analogy-based
for crack identification in structures, Journal of Engineering DSS for bidding in construction, Journal of Construction Engi-
Mechanics, ASCE, 124 (2), 152–7. neering and Management, 119 (3), 466–79.
Lingras, P. & Adamo, M. (1996), Average and peak traffic Mukherjee, A. & Deshpande, J. M. (1995a), Application of artifi-
volumes: Neural nets, regression, and factor approaches, Jour- cial neural networks in structural design expert systems, Com-
nal of Computing in Civil Engineering, ASCE, 10 (4), 300–6. puters and Structures, 54 (3), 367–75.
Liong, S.-Y., Lim, W.-H. & Paudyal, G. N. (2000), River stage Mukherjee, A. & Deshpande, J. M. (1995b), Modeling initial
forecasting in Bangladesh: Neural network approach, Journal design process using artificial neural networks, Journal of Com-
of Computing in Civil Engineering, ASCE, 14 (1), 1–8. puting in Civil Engineering, ASCE, 9 (3), 194–200.
Liu, W. & James, C. S. (2000), Estimation of discharge capacity Mukherjee, A., Deshpande, J. M. & Anmala, J. (1996), Prediction
in meandering compound channels using artificial neural net- of buckling load of columns using artificial neural networks,
works, Canadian Journal of Civil Engineering, 27, 297–308. Journal of Structural Engineering, 122 (11), 1385–7.
Maier, H. R. & Dandy, G. C. (1997), Determining inputs for Murtaza, M. B. & Fisher, D. J. (1994), Neuromodex: Neural
neural networks models of multivariate time series, Computer- network system for modular construction decision, Journal of
Aided Civil and Infrastructure Engineering, 12 (5), 353–68. Computing in Civil Engineering, ASCE, 8 (2), 221–3.
Manevitz, L., Yousef, M. & Givoli, D. (1997), Finite element mesh Ni, S. H., Lu, P. C. & Juang, C. H. (1996), A fuzzy neural net-
generation using self-organizing neural networks, Computer- work approach to evaluation of slope failure potential, Micro-
Aided Civil and Infrastructure Engineering, 12 (4), 233–50. computers in Civil Engineering, 11 (1), 59–66.
 Neural networks in civil engineering: 1989–2000 141

Nikzad, K., Ghaboussi, J. & Paul, S. L. (1996), Actuator dynam- Rosenblatt, F. (1962), Principle of Neurodynamics, Spartan
ics and delay compensation using neurocontrollers, Journal of Books, New York.
Engineering Mechanics, 122 (10), 966–75. Rudomin, P., Arbib, M. A., Cervantes-Perez, F. & Romo, R.
Owusu-Ababia, S. (1998), Effect of neural network topology on (1993), Neuroscience: From Neural Networks to Artificial Intel-
flexible pavement cracking prediction, Computer-Aided Civil ligence, Springer-Verlag, Berlin.
and Infrastructure Engineering, 13 (5), 349–55. Rumelhart, D. E., Hinton, G. E. & Williams, R. J. (1986), Learn-
Pain, C. C., De Oliveira, C. R. E. & Goddard, A. J. H. (1999), ing internal representation by error propagation, in Rumelhart,
A neural network graph partitioning procedure for grid-based D. E. et al., eds., Parallel Distributed Processing, MIT Press,
domain decomposition, International Journal for Numerical Cambridge, MA, pp. 318–62.
Methods in Engineering, 44, 593–613. Saito, M. & Fan, J. (2000), Artificial neural network–based heuris-
Pandey, P. C. & Barai, S. V. (1995), Multilayer perceptron in dam- tic optimal traffic signal timing, Computer-Aided Civil and
age detection of bridge structures, Computers and Structures, Infrastructure Engineering, 15 (4), 281–91.
54 (4), 597–608. Samant, A. & Adeli, H. (2000), Feature extraction for traffic inci-
Papadrakakis, M., Lagaros, N. D. & Tsompanakis, Y. (1998), dent detection using wavelet transform and linear discriminant
Structural optimization using evolution strategies and neural analysis, Computer-Aided Civil and Infrastructure Engineering,
networks, Computer Methods in Applied Mechanics and Engi- 15 (4), 241–50.
neering, 156, 309–33. Sankarasubramanian, G. & Rajasekaran, S. (1996), Constitutive
Papadrakakis, M., Papadopoulos, V. & Lagaros, N. D. (1996), modeling of concrete using a new failure criterion, Computers
Structural reliability analysis of elastic-plastic strutures using and Structures, 58 (5), 1003–14.
neural networks and Monte Carlo simulation, Computer Meth- Savin, D., Alkass, S. & Fazio, P. (1996), Construction resource
ods in Applied Mechanics and Engineering, 136, 145–63. leveling using neural networks, Canadian Journal of Civil
Park, H. S. & Adeli, H. (1995), A neural dynamics model for Engineering, 23, 917–25.
structural optimization: Application to plastic design of struc- Savin, D., Alkass, S. & Fazio, P. (1998), Calculating weight
tures, Computers and Structures, 57 (3), 391–9. matrix of neural network for resource leveling, Journal of Com-
Park, H. S. & Adeli, H. (1997a), Data parallel neural dynamics puting in Civil Engineering, 12 (4), 241–8.
model for integrated design of steel structures, Microcomputers Sayed, T. & Abdelwahab, W. (1998), Comparison of fuzzy and
in Civil Engineering, 12 (5), 311–26. neural classifiers for road accident analysis, Journal of Com-
Park, H. S. & Adeli, H. (1997b), Distributed neural dynamics puting in Civil Engineering, ASCE, 12 (1), 42–6.
algorithms for optimization of large steel structures, Journal of Sayed, T. & Razavi, A. (2000), Comparison of neural and conven-
Structural Engineering, ASCE, 123 (7), 880–8. tional approaches to mode choice analysis, Journal of Comput-
Park, D. & Rilett, L. R. (1999), Forecasting freeway link travel ing in Civil Engineering, ASCE, 14 (1), 23–30.
times with a multilayer feedforward neural network, Computer- Senouci, A. & Adeli, H. (2001), Resource scheduling using neural
Aided Civil and Infrastructure Engineering, 14 (5), 357–67. dynamics model of Adeli and Park, Journal of Construction
Parvin, A. & Serpen, G. (1999), Recurrent neural networks for Engineering and Management, ASCE, 127 (1).
structural optimization, Computer-Aided Civil and Infrastruc- Sinha, S. K. & McKim, R. A. (2000), Artificial neural network for
ture Engineering, 14 (6), 445–51. measuring organization effectiveness, Journal of Computing in
Poggio, T. & Girosi, F. (1990), Networks for approximation and Civil Engineering, ASCE, 14 (1), 9–14.
learning, Proceedings of the IEEE, 78, 1481–97. Sonmez, R. & Rowings, J. E. (1998), Construction labor produc-
Pompe, P. P. M. & Feelders, A. J. (1997), Using machine learning, tivity modeling with neural networks, Journal of Construction
neural networks, and statistics to predict corporate bankruptcy, Engineering and Management, 124 (6), 498–504.
Computer-Aided Civil and Infrastructure Engineering, 12 (4), Stavroulakis, G. E. & Antes, H. (1998), Neural crack identification
267–76. in steady state elastodynamics, Computer Methods in Applied
Rajasekaran, S., Febin, M. F. & Ramasamy, J. V. (1996), Artifi- Mechanics and Engineering, 165, 129–46.
cial fuzzy neural networks in civil engineering, Computers and Stephens, J. E. & Vanluchene, R. D. (1994), Integrated assess-
Structures, 61 (2), 291–302. ment of seismic damage in structures, Microcomputers in Civil
Razaqpur, A. G., Abd El Halim, A. O. & Mohamed, H. A. (1996), Engineering, 9 (2), 119–28.
Bridge management by dynamic programming and neural net- Szewczyk, Z. P. & Hajela, P. (1994), Damage detection in struc-
works, Canadian Journal of Civil Engineering, 23, 1064–9. tures based on feature-sensitive neural networks, Journal of
Roberts, C. A. & Attoh-Okine, N. O. (1998), A comparative Computing in Civil Engineering, ASCE, 8 (2), 163–78.
analysis of two artificial neural networks using pavement per- Szewczyk, Z. P. & Noor, A. K. (1996), A hybrid neurocomputing/
formance prediction, Computer-Aided Civil and Infrastructure numerical strategy for nonlinear structural analysis, Computers
Engineering, 13 (5), 339–48. and Structures, 58 (4), 661–77.
Rodriguez, M. J. & Sérodes, J. (1996), Neural network–based Szewczyk, Z. P. & Noor, A. K. (1997), A hybrid numerical/ neu-
modelling of the adequate chlorine dosage for drinking water rocomputing strategy for sensitivity analysis of nonlinear struc-
disinfection, Canadian Journal of Civil Engineering, 23, tures, Computers and Structures, 65 (6), 869–80.
621–31. Tang, Y. (1996a), Active control of SDF systems using artificial
Rogers, J. L. (1994), Simulating structural analysis with neural neural networks, Computers and Structures, 60 (5), 695–703.
network, Journal of Computing in Civil Engineering, ASCE, Tang, Y. (1996b), New algorithm for active structural control,
8 (2), 252–65. Journal of Structural Engineering, ASCE, 122 (9), 1081–8.
142 Adeli

Tashakori, A. & Adeli, H. (2001), Optimum design of cold-formed Theory and Neural Networks, MIT Press, Cambridge, MA,
steel space structures using neural dynamics model, Journal of pp. 719–24.
Structural Engineering, ASCE, 127, (in press). Williams, T. P. (1994), Predicting changes in construction cost
Tawfik, M., Ibrahim, A. & Fahmy, H. (1997), Hysteresis sensitive indexes using neural networks, Journal of Construction Engi-
neural network for modeling rating curves, Journal of Comput- neering and Management, ASCE, 120 (2), 306–20.
ing in Civil Engineering, ASCE, 11 (3), 206–11. Williams, T. P. & Gucunski, N. (1995), Neural networks for back-
Teh, C. I., Wong, K. S., Goh, A. T. C. & Jaritngam, S. (1997), calculation of moduli from SASW tests, Journal of Computing
Prediction of pile capacity using neural networks, Journal of in Civil Engineering, ASCE, 9 (1), 1–8.
Computing in Civil Engineering, ASCE, 11 (2), 129–38. Wu, X., Ghaboussi, J. & Garrett, J. H., Jr. (1992), Use of neural
Theocaris, P. S. & Panagiotopoulos, P. D. (1993), Neural networks networks in prediction of structural damage, Computers and
for computing in fracture mechanics: Methods and prospects Structures, 42 (4), 649–59.
of applications, Computer Methods in Applied Mechanics and Yeh, I.-C. (1995), Construction site layout using annealed neural
Engineering, 106, 213–28. networks, Journal of Computing in Civil Engineering, ASCE,
Theocaris, P. S. & Panagiotopoulos, P. D. (1995), Generalised 9 (3), 201–8.
hardening plasticity approximated via anisotropic elasticity: Yeh, I.-C. (1998), Structural engineering applications with aug-
A neural network approach, Computer Methods in Applied mented neuron networks, Computer-Aided Civil and Infrastruc-
Mechanics and Engineering, 125, 123–39. ture Engineering, 13 (2), 83–90.
Thirumalaiah, K. & Deo, M. C. (1998), Real-time flood forecast- Yeh, I.-C. (1999), Design of high-performance concrete mixture
ing using neural networks, Computer-Aided Civil and Infras- using neural networks and nonlinear programming, Journal of
tructure Engineering, 13 (2), 101–11. Computing in Civil Engineering, ASCE, 13 (1), 36–42.
Topping, B. H. V. & Bahreininejad, A. (1997), Neural Com- Yeh, Y.-C., Kuo, Y.-H. & Hsu, D.-S. (1993), Building KBES for
puting for Structural Mechanics, Saxe-Coburg Publications, diagnosing PC pile with artificial neural network, Journal of
Edinburgh, UK. Computing in Civil Engineering, ASCE, 7 (1), 71–93.
Topping, B. H. V., Khan, A. I. & Bahreininejad, A. (1997), Paral- Yen, G. G. (1994), Identification and control of large struc-
lel training of neural networks for finite element mesh decom- tures using neural networks, Computers and Structures, 52 (5),
position, Computers and Structures, 63 (4), 693–707. 859–70.
Topping, B. H. V., Sziveri, J., Bahreinejad, A., Leite, J. P. B. & Yen, G. G. (1996), Distributive vibration control in flexible multi-
Cheng, B. (1998), Parallel processing, neural networks and body dynamics, Computers and Structures, 61 (5), 957–65.
genetic algorithms, Advances in Engineering Software, 29, Yu, G. & Adeli, H. (1991), Computer-aided design using object-
763–86. oriented programming paradigm and blackboard architecture,
Turkkan, N. & Srivastava, N. K. (1995), Prediction of wind load Microcomputers in Civil Engineering, 6 (3), 177–90.
distribution for air-supported structures using neural networks, Yu, G. & Adeli, H. (1993), Object-oriented finite element analysis
Canadian Journal of Civil Engineering, 22, 453–61. using EER model, Journal of Structural Engineering, ASCE,
UBC (1997), Uniform Building Code, Vol. 2: Structural Engineer- 119 (9), 2763–81.
ing Design Provisions, International Conference of Building Yun, C. B. & Bahng, E. Y. (2000), Substructural identifica-
Officials, Whittier, CA. tion using neural networks, Computers and Structures, 77,
Vanluchene, R. D. & Sun, R. (1990), Neural networks in struc- 41–52.
tural engineering, Microcomputers in Civil Engineering, 5 (3), Zadeh, L. A. (1970), Decision making in a fuzzy environment,
207–15. Management Sciences, 17 (4), 141–64.
Widrow, B. & Lehr, M. A. (1995), Perceptrons, adalines, and Zadeh, L. (1978), Fuzzy set as a basis for a theory of possibility,
backpropagation, in Arbib, M. A., ed., The Handbook of Brain Fuzzy Sets and Systems, 1 (1), 3–28.