Professional Documents
Culture Documents
Automation in Construction: Min-Yuan Cheng, Hsing-Chih Tsai, Wen-Shan Hsieh
Automation in Construction: Min-Yuan Cheng, Hsing-Chih Tsai, Wen-Shan Hsieh
Automation in Construction
j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / a u t c o n
A R T I C L E I N F O a b s t r a c t
Article history: Conceptual cost estimates, the basis of project evaluation, engineering design, cost budgeting, and cost
Accepted 12 July 2008 management, not only play an essential role in construction project feasibility studies, but are fundamental
to a project's ultimate success. As practiced today, construction cost estimates generally rely on experts'
Keywords: intuitive experience. Scientific methods should be developed and employed during project planning and
Construction cost
design stages in order to raise conceptual cost estimate accuracy. This study proposes the use of an artificial
Conceptual estimates
Genetic Algorithms
intelligence approach, the Evolutionary Fuzzy Neural Inference Model (EFNIM), to improve cost estimation
Fuzzy Logic accuracy. The advantages inherent in Genetic Algorithms, Fuzzy Logic and Neural Networks are incorporated
Neural Networks into the EFNIM, making this model highly applicable to identifying optimal solutions for complex problems.
Furthermore, this paper presents Evolutionary Web-based Conceptual Cost Estimators (EWCCE) obtained by
integrating EFNIM, WWW, and historical construction data to assist in project management. The developed
EWCCE provides two kinds of estimators that can be deployed to estimate conceptual construction cost more
precisely during the early stages of projects.
Crown Copyright © 2008 Published by Elsevier B.V. All rights reserved.
1. Introduction project costs due to the large number of significant variables and
the interactions thereof. Thus, traditional methods have limited
Cost estimates are fundamental to all project-related engineering applicability.
and greatly influence planning, design, bidding, cost management/ Artificial intelligence approaches are applicable to cost estimating
budgeting and even construction management. Such estimates allow problems related to expert systems, case-based reasoning (CBR),
owners and planners to evaluate project feasibility and control costs Neural Networks (NNs), Fuzzy Logic (FL), Genetic Algorithms (GAs),
effectively in detailed project design work. Due to the limited and derivatives of such. Many research studies have been done in this
availability of information during the early stages of a project, area. For instance, an integrated knowledge-based system for
construction managers typically leverage their knowledge, experience alternative design decisions, materials selection and cost estimating,
and estimators to estimate project costs, i.e., they usually rely on their used mainly in pre-design analysis, was proposed by Mohamed and
intuition. Researchers have worked to develop cost estimators that Celik [23]. Serpell [25] proposed a model of this problem based on
maximize the practical value of limited information available in order existing knowledge and demonstrated how the model was used to
to improve cost estimate accuracy and reliability, which should develop a knowledge-based assessment system. Arditi and Suh [3]
improve the suitability of resultant designs and subsequent project developed an expert system to present decision criteria developed to
execution work. classify available cost estimating packages. An et al. [2] developed a
Traditionally, cost estimating models have been developed using case-based reasoning model, where an analytic hierarchy process was
statistical methods [26]. Regression analysis represents a traditional employed to incorporate experience. Yau and Yang [29] applied CBR to
alternative [5,16], an inherent disadvantage of which is its estimate at the preliminary design stage the construction project
requirement of a defined mathematical form for cost functions. In implementation duration and costs. NNs represent the most fre-
addition, traditional methods are hampered in estimating accurate quently applied approach in this type of application. Wilmot and Mei
[27] developed an NN model to estimate highway construction cost
escalation over time. Adeli and Wu [1] also employed NNs to estimate
highway construction cost and identified noises in the data. Williams
[28] used NNs to predict change in the ENR construction cost index
⁎ Corresponding author. Tel.: +886 2 27376663, +886 2 27301277; fax: +886 2
27301074.
and concluded that the back-propagation neural network model
E-mail addresses: myc@mail.ntust.edu.tw (M.Y. Cheng), tsaihsingchih@gmail.com cannot accurately predict cost index movement due to the complex-
(H.-C. Tsai). ities involved.
0926-5805/$ – see front matter. Crown Copyright © 2008 Published by Elsevier B.V. All rights reserved.
doi:10.1016/j.autcon.2008.07.001
M.-Y. Cheng et al. / Automation in Construction 18 (2009) 164–172 165
Fig. 3. Membership functions: (a) trapezoidal MF; (b) triangle MF; (c) complete MF set.
Table 2
Category Estimator Impact Factors
Case no. Actual output ($/m2) Desired output ($/m2) Diff. ($/m2) EWCCE was developed to meet the goal of assisting construction
1 456.9 575.0 118.1 project planning and design through the use of evaluated cost
2 651.0 525.9 −125.1 estimates. The uses of EWCCE should be identified first. In Fig. 5, an
3 556.0 458.8 −97.3 overall construction cost estimator is used in the preliminary planning
4 830.3 790.1 − 40.2
stage, when detailed project plans have not yet been drafted.
5 596.9 661.4 64.5
Preliminary plans can be drafted with in-situ investigations and
Notations: Diff. = Actual − Desired. identified demands, after which the generated overall cost estimate
can be used to check the relevance and accuracy of those plans. Initial
inputs for temporary construction; 7 inputs for geotechnical con- design will be done next in the planning stage, and then demands and
struction; 8 for structural construction; 9 for interior decoration; 8 for designs will be checked against category estimates. Detailed planning
electromechanical infrastructure; 5 for miscellaneous construction; and design can be executed once all data and estimates meet project
and 4 for indirect construction. Construction costs used as training management needs. These conceptual estimates significantly influ-
targets reflect Taiwan's published price index for calendar year 2001. ence project construction and management.
Therefore, the proposed estimators are capable of dealing with unit Once the practicality of EWCCE has been confirmed, the system
price fluctuation of the wok and material items in the market. architecture can be designed (see Fig. 6). User interface should be done
Table 4 ranged from 1 to 5. The model uses trapezoidal MFs with widths in
Testing results for category estimates the range. Five fuzzification grades were set for each input and each
Engineering Case no. Actual output Desired output Diff. Ratio of input variable used its individual complete MFs sets. The crossover
categories ($/m2) ($/m2) ($/m2) category rate for simulation was 0.9, and the mutation rate was 0.025 per bit.
cost (%) Because the output is not binary, the sigmoid function was chosen for
Temporary 1 11.9 17.4 5.5 2.12 activation, with its slope set in the range of 1 to 6. Limitation of
construction 2 24.2 26.8 2.6 4.28
generation number was set as 3000 (iterations). The population size
3 14.0 10.5 −3.5 2.78
4 15.0 16.8 1.8 2.11 is 50.
5 14.6 17.9 3.3 2.30
Geotechnical 1 31.4 49.2 17.9 5.60 5.1. Overall construction cost estimator results
construction 2 88.9 27.2 −61.6 15.72
3 32.6 41.2 8.5 6.49
4 90.2 136.4 46.2 12.69
This estimates construction cost, obtained while the project is in
5 55.9 52.8 −3.1 8.81 the preliminary concept stage, i.e. without detailed engineering plans
Structural 1 157.4 175.9 18.5 28.08 organized by category, has significant bearing on detailed planning.
construction 2 147.1 147.5 0.4 26.03 Estimation is fraught with uncertainty and vagueness due to
3 176.5 147.3 − 29.2 35.07
incomplete and complex data. As results are not easily expressed in
4 142.2 135.7 −6.5 20.00
5 180.9 166.0 − 15.0 28.52 a formula, such typically relies heavily on the intuition of engineers
Interior decoration 1 160.4 137.5 − 22.9 28.61 working on the project. EFNIM is quite applicable for this problem
2 134.8 175.1 40.3 23.85 owing to its capacity for inference. After a process of evolutionary
3 146.8 166.7 19.9 29.16 training, achieved through applying the 23 training cases, five testing
4 241.8 239.5 −2.4 34.01
results are shown in Table 3. The result model structure of this overall
5 186.2 196.7 10.5 29.35
Electromechanical 1 134.4 136.1 1.7 23.98 cost estimate is shown in Fig. 8.
infrastructure 2 105.3 87.8 −17.6 18.64
3 82.7 64.8 −17.9 16.43 5.2. Category construction cost estimator results
4 154.3 178.3 24.0 21.71
5 112.5 135.9 23.3 17.74
Miscellaneous 1 19.1 19.4 0.3 3.40 Although the overall construction cost estimator has been devel-
construction 2 25.8 26.0 0.2 4.57 oped, construction plans in each category must still be designed.
3 20.6 6.8 − 13.8 4.09 Construction costs for engineering categories should be estimated to
4 18.6 32.6 14.0 2.62 ensure costs are controlled effectively and facilitate project manage-
5 30.2 24.8 −5.5 4.77
ment. Although it is difficult to completely categorize construction
Indirect 1 46.0 39.4 −6.6 8.21
Construction 2 39.1 35.6 −3.5 6.91 work into types of engineering, such is essential in order to estimate
3 30.1 21.6 −8.6 5.99 category cost values and helpful in project planning and design. Table 4
4 48.8 50.8 2.0 6.86 not only shows estimation results, but also category cost ratios. It is
5 54.0 67.3 13.3 8.51
apparent that category cost ratios bear significantly on project
Total construction 1 560.619 575.022 14.403 100%
cost 2 565.160 525.933 − 39.227 100%
planning and design. This result allows cost management to be
3 503.268 458.768 −44.500 100% effectively implemented into construction engineering categories.
4 710.965 790.127 79.162 100% There are seven cost estimation models trained. The structural
5 634.408 661.409 27.001 100% construction cost which significantly influences the total cost is
Notations: Diff = Actual − Desired. shown in its result model structure (see Fig. 9).
Ratio of category cost = Actual category cost / Total actual construction cost.
5.3. Comparisons between overall and category estimates
through the Internet based on the developed estimators and historical In practice, overall estimates accurate to within 25% and category
data in the host database. estimates accurate to within 15% using engineers' experience are
typically considered acceptable. Estimators developed in this paper
4.4. EWCCE system development achieve a high level of precision for construction cost estimation
during the early stages of a project (see Table 5). Estimating
The architecture of the EWCCE system involves identifying the construction costs more precisely will help make designs more
tools used to develop system function modules and user interfaces. feasible and projects more efficient by enhancing project management.
User and program interfaces for construction cost estimates are
established at three levels as follows: application user, command user, 6. Conclusions
and program data (see Fig. 7).
The primary components of the system, including Visual Basic for This paper presents comprehensive descriptions of Evolutionary
EFNIS, ASP, and a database, were developed within the Microsoft Fuzzy Neural Inference System mechanisms and the process of
Windows environment. Through the web-based system, Projects can developing construction cost estimators, i.e. EWCCE. The EFNIM
be accessed, modified and updated in accordance with users' level of mechanism is a fusion of GAs, FL, and NNs. GAs are used for a global
authority and EFNIS conceptual cost estimates. optimization, FL addresses imprecision and approximate reasoning,
and NNs are employed for input–output mapping. Therefore, EFNIM
5. Verification and testing represents a valid approach to handling complicated problems and
problems for which there is a significant level of uncertainty. The
As noted above, this study developed two kinds of estimators and EWCCE integrates EFNIM, WWW, and historical project data. Users
used 23 training cases and 5 testing cases. The capabilities of EFNIM can access EWCCE via the Internet, and EFNIM advantages are
were employed in these estimations. However, the EFNIM is time- reflected in the EWCCE.
consuming due in large part to its use of GAs. Therefore, experiments This study proposed two kinds of construction cost estimators. The
should be run to set parameters to a practicable range. The number of overall construction cost estimator was established to estimate a total
hidden layers varied from 1 to 4 and neurons in each hidden layer cost in the absence of categorized engineering plans. The category
M.-Y. Cheng et al. / Automation in Construction 18 (2009) 164–172 171
estimators, with additional data inputs, was established to evaluate execution even when only a minimal amount of available data
engineering costs within categories. The advantages of EWCCE include: is available.
2. Category construction costs, categorized by engineering type, offer
1. A overall construction cost estimate can be provided during an alternative to overall estimates that provides results that are
the preliminary project planning stage to facilitate project more reasonable and practicable.
3. Category estimators supply useful information on the relative ratios
of engineering categories, which is essential for detailed construc-
tion cost management.
Table 5 4. EFNIM, WWW, and historical data are all incorporated into EWCCE,
Result comparisons of overall and category estimates making EWCCE a web accessible platform that can be applied to
Case no. Error of overall estimates (%) Error of total category estimates (%) various construction projects.
1 20.541 2.504 This paper presents the application of EWCCE to estimate
2 23.783 7.458
construction costs during the early stage of construction projects in
3 21.201 9.699
4 5.082 10.018 order to facilitate designer, owner and contractor for decision-
5 9.755 4.082 making. Results show that EWCCE is relevant and applicable to
Avg. 16.072 6.753 construction management in Taiwan and may be implemented
Notations: Error = [(Actual − Desired)/Desired] × 100%. worldwide with modifications to account for specific regional/
Avg. = error average. national factors.
172 M.-Y. Cheng et al. / Automation in Construction 18 (2009) 164–172
References [16] F. Khosrowshahi, A.P. Kaka, Estimation of project total cost and duration for
housing projects in the UK, Building and Environment 31 (4) (1996) 373–383.
[17] G.H. Kim, D.S. Seo, K.I. Kang, Hybrid models of neural networks and genetic
[1] H. Adeli, M. Wu, Regularization neural network for construction cost estimation,
algorithms for predicting preliminary cost estimates, Journal of Computing in Civil
Journal of Construction Engineering and Management 124 (1) (1998) 18–24.
Engineering 19 (2) (2005) 208–211.
[2] S.H. An, G.H. Kim, K.I. Kang, A case-based reasoning cost estimating model using
[18] G.J. Klir, B. Yuan, Fuzzy Sets and Fuzzy Logic: Theory and Applications, Prentice Hall
experience by analytic hierarchy process, Building and Environment 42 (7) (2007)
PTR, Upper Saddle River, New Jersey, 1995.
2573–2579.
[19] P. Liatsis, Y.J.P. Goulermas, Minimal optimal topologies for invariant higher-order
[3] D. Arditi, K. Suh, Expert system for cost estimating software selection, Cost
neural architectures using genetic algorithms, Proceedings of the IEEE Interna-
Engineering 33 (6) (1991) 9–19.
tional Symposium on Industrial Electronics 2 (1995) 792–797.
[4] A.H. Boussabaine, T.M.S. Elhag, A neurofuzzy model for predicting cost and
[20] L. Liu, K. Zhu, Improving cost estimates of construction projects using phased cost
duration of construction projects, RICS Research, The Royal Institution of Chartered
factors, Journal of Construction Engineering and Management 133 (1) (2007)
Surveyors, 1997, 9 pages.
91–95.
[5] P.A. Bowen, P.J. Edwards, Cost modeling and price forecasting; practice and theory
[21] H.R. Maier, T. Sayed, B.J. Lence, Forecasting cyanobacterial concentrations using b-
in perspective, Construction Management and Economics 3 (1985) 199–215.
spline networks, Journal of Computing in civil engineering 14 (3) (2000) 183–189.
[6] M.Y. Cheng, C.H. Ko, Computer-aided decision support system for hillside safety
[22] Z. Michalewicz, Genetic Algorithms + Data Structures = Evolution Programs,
monitoring, Automation in Construction 11 (4) (2002) 453–466.
Springer-Verlag, Berlin, 1996.
[7] M.Y. Cheng, C.H. Ko, Object-oriented evolutionary fuzzy neural inference system
[23] A. Mohamed, T. Celik, An integrated knowledge-based system for alternative
for construction management, Journal of Construction Engineering and Manage-
design and materials selection and cost estimating, Expert Systems with
ment 129 (4) (2003) 461–469.
Applications 14 (3) (1998) 329–339.
[8] M. Gen, R. Cheng, Genetic Algorithms and Engineering Design, Wiley, New York,
[24] G.N. Rao, F. Grobler, S. Kim, Conceptual cost estimating: a hybrid neural-expert
1997.
system approach, Computing in Civil and Building Engineering (1993) 423–430.
[9] H. Ghezelayagh, K.Y. Lee, Training neuro-fuzzy boiler identifier with genetic
[25] A.F. Serpell, Towards a knowledge-based assessment of conceptual cost estimates,
algorithm and error back-propagation, IEEE Power Engineering Society 2 (1999)
Building Research and Information 32 (2) (2004) 157–164.
978–982.
[26] S. Singh, Cost model for reinforced concrete beam and slab structures in building,
[10] M.B. Gorzalczany, P. Gradzki, A neuro-fuzzy-genetic classifier for technical
Journal of Construction Engineering and Management 116 (1) (1990) 54–67.
applications, Proceedings of IEEE International Conference on Industrial Technol-
[27] C.G. Wilmot, B. Mei, Neural network modeling of highway construction costs,
ogy 1 (2000) 503–508.
Journal of Construction Engineering and Management 124 (3) (2005) 210–218.
[11] I. Hayashi, M. Umano, T. Maeda, A. Bastian, L.C. Jain, Acquisition of fuzzy knowledge
[28] T.P. Williams, Predicting changes in construction cost indexes using neural
by NN and GA, IEEE International Conference on Knowledge-based Intelligent
networks, Journal of Construction Engineering and Management 120 (2) (1994)
Electronic Systems, 1998, pp. 69–78.
306–320.
[12] S. Haykin, Neural Networks: A Comprehensive Foundation, Prentice-Hall, New
[29] N.J. Yau, J.B. Yang, Case-based reasoning in construction management, Computer-
York, 1999.
Aided Civil and Infrastructure Engineering 13 (2) (1998) 143–150.
[13] T. Hegazy, A. Ayed, Neural network model for parametric cost estimation of
[30] W.D. Yu, C.C. Lai, W.L. Lee, A WICE approach to real-time construction cost
highway projects, Journal of Construction Engineering and Management 124 (3)
estimation, Automation in Construction 15 (1) (2006) 12–19.
(1998) 210–218.
[31] L.A. Zadeh, Fuzzy sets, Information and Control 8 (3) (1965) 338–353.
[14] J.H. Holland, Adaptation in Neural and Artificial Systems, The University of
Michigan Press, Ann Arbor, 1975.
[15] I. Jagielska, C. Matthews, T. Whitfort, An investigation into the application of neural
networks, fuzzy logic, genetic algorithms, and rough sets to automated knowledge
acquisition for classification problems, Neurocomputing 24 (1–3) (1999) 37–54.