Professional Documents
Culture Documents
(Asce) 0733-9364 (2003) 129 4
(Asce) 0733-9364 (2003) 129 4
(Asce) 0733-9364 (2003) 129 4
Abstract: This paper presents a predictive tool that uses safety factors to estimate accident risk for processes commonly employed on
construction sites. The tool is demonstrated in the context of the utility-trenching process. Based on expert surveys, preplanning,
supervision, and training are identified as critical safety factors needed to predict accident risk when evaluating safety performance related
to trenching operations. Preplanning has a greater impact on excavation tasks than supervision and training, while conversely, supervision
and training are more important during the pipe installation phase of trenching operations.
DOI: 10.1061/共ASCE兲0733-9364共2003兲129:4共431兲
CE Database subject headings: Accident prediction; Safety factors; Construction site accidents; Trenching.
Safety Issues in Construction The safety practices encountered on construction sites are as
varied as the sites themselves 共Wilson and Koehn 2000兲, and the
Though improvements in construction worker safety have been assessment of the risk of accidents is, in most cases, more sub-
achieved, the construction industry continues to lag behind most jective than objective 共Levitt and Samuelson 1993兲. However, the
other industries with regard to safety 共National 1998兲. A dispro- contention of this paper is that a priori data can be used to esti-
portionate number of disabling injuries and fatalities occur in the mate the risk of accidents based on current safety practices and
construction sector 共Hinze 1997兲. Construction accounted for the performance expected on the proposed project.
nearly one-fifth of all industrial fatalities in 2000 共increasing from An analysis of the risk of accidents at a construction site be-
one-sixth in 1995兲, which amounts to three times its 6.6% share gins with a determination of the work practices and conditions
of total employment in 2000 共BLS 2001兲. Construction work is that create an unsafe work place 共Gambatese and Stewart 1999兲.
intrinsically hazardous, but inadequate task planning, poor safety This paper will illustrate a system that identifies the factors af-
training, lack of safety incentives, and insufficient incident fecting site safety and the probability of accidents for particular
investigation—that is, poor safety management—play a major work processes 共for example, trenching, slab construction, and so
role in this poor level of safety within the construction industry. forth兲.
These safety factors will be used to estimate the probability of
accidents based on a priori analysis. The analysis uses fuzzy
Problem Statement mathematical techniques and input from safety experts to assess
factors that impact field operations and influence accident poten-
Accident statistics have played an important role in measuring tial. The methodology developed for trenching operations demon-
safety performance 共BLS 1997兲. However, accident statistics are strates a predictive tool that can be used to assess accident prob-
based on actual accident data and the compilation of postaccident ability before work begins.
information. Very little research has been focused on assessment
of accident potential before the fact 共Staley and Foster 1996兲. Ex
post facto data provide factual data about accident frequency, se- Factors Affecting Safety Performance
verity, and so on, which are not necessarily helpful in predicting
accidents or assessing accident risk. Lack of an accident does not In this demonstration, factors affecting safety performance are
mean there is no risk of an accident; rather, there is a need to limited exclusively to those factors applicable to utility trenching.
estimate the risk level of accidents based on current safety prac- For identification of the safety factors, 35 experts were consulted:
tices. 13 safety directors in corporate construction provided manage-
ment perspectives, and 22 respondents 共project managers, super-
1 intendents, and foremen兲 provided field perspective. The group
Research Associate, Division of Construction Engineering and Man-
agement, School of Civil Engineering, Purdue Univ., West Lafayette, averaged 24.2 years of experience in the construction industry
IN 47907-1294.
2
Professor and Head, Division of Construction Engineering and Man-
agement, School of Civil Engineering, Purdue Univ., West Lafayette,
IN 47907-1294. Table 1. Experience of Experts Consulted for Safety Factor
Note. Discussion open until January 1, 2004. Separate discussions Identification
must be submitted for individual papers. To extend the closing date by
one month, a written request must be filed with the ASCE Managing Perspectives Management Field Total
Editor. The manuscript for this paper was submitted for review and pos- Number of experts 13 22 35
sible publication on September 4, 2001; approved on June 17, 2002. This Experience
paper is part of the Journal of Construction Engineering and Manage-
In construction 共years兲 24.3 24.1 24.2
ment, Vol. 129, No. 4, August 1, 2003. ©ASCE, ISSN 0733-9364/
Only in trenching 共years兲 16 19.7 18.3
2003/4-431– 436/$18.00.
Small 1.00 1.00 0.94 0.78 0.31 0.00 0.00 0.00 0.00 0.00 0.00
Very small 1.00 1.00 0.88 0.61 0.10 0.00 0.00 0.00 0.00 0.00 0.00
This framework of analysis was originally proposed by Ayyub medium-severe linguistic parameters were used to indicate ad-
and Eldukair 共1990兲; the variables are shown in Table 3. For verse consequences, and finally high-low reflected the frequency
instance, if the condition C of the training factor is poor and the level of occurrence 共accident兲. The values of the grading scales
frequency F of poor training is large, the adverse consequences from all survey respondents were analyzed to define the member-
AC of this poor training are severe, leading to an accident fre- ship functions for the primary values. The defined grading scales
quency R, which is high. Conversely, if the condition C of super- can be extended by the effect of linguistic hedges 共modifiers that
vision is good and the frequency F is large, the AC is small and limit or extend the fuzzy values, such as ‘‘Fairly’’ or ‘‘Very’’兲 to
the R is low. The fuzzy relations between the values of these four translate high and low levels of linguistic measures. The transla-
variables are given in the appendix. tion of linguistic hedges such as Very and Fairly into fuzzy sets
may be assumed to be Very⫽Con(A)⫽ 关 U A (x i ) 兴 2 , and Fairly
⫽Dil(A)⫽ 关 U A (x i ) 兴 1/2, where U(x i ) is the membership function
Data Acquisition of the element x i , and Con and Dil are the concentration and
dilation operations, respectively 共Tsoukalas and Uhrig 1997兲. The
To translate the values of the four linguistic variables into safety fuzzy membership functions developed for the grading measures
factors, they must be converted to fuzzy mathematical variables, of all variables such as C, F, and AC are given in Tables 4 –7.
and a membership function reflecting a grading scale must be
developed. For the acquisition of the data for the grading scale
and the linguistic measure of the variables, a separate set of 35
Linguistic Evaluation of Variables
experts was consulted for an interview survey who had an aver-
age of 22.3 years of experience in the construction industry and
16.6 years of experience in the area of trenching operations. Utility trenching using a trench shield was considered in order to
quantify the effects of accidents for this study. The interview sur-
vey asked questions designed to linguistically determine the level
Grading Scale Development of variables 共for example, poor, good, high, low, and so on兲. Con-
dition assessment of the three factors 共training, supervision, pre-
The survey asked the experts to respond, based on their experi- planning兲 as they relate to the trenching operations—excavation
ence, as to the condition levels of three factors: training, supervi- and pipe installation—was developed. The survey asked such
sion, and preplanning. This was done to develop membership questions as the following: ‘‘When the condition of the safety
sets. For instance, the expert responded as to what degree training factor is very poor, what is the frequency of occurrence for the
共in his or her experience兲 is good, moderate, or poor. Similarly, very poor condition?’’ 共for example, very large, large, fairly large,
the expert indicated at what frequency level good training oc- medium, fairly small, small, and very small兲.
curred: large 共often兲, medium, and so on. This information was Additional questions to determine the level of adverse conse-
used to develop membership sets for C, F, AC, and R. quences and frequency were also included in the questionnaire:
To establish a general numerical guideline for the condition- ‘‘What is the adverse consequence 共that is, negative impact兲 de-
related safety factors, poor conditions were given a value of 0 and riving from the very poor condition?’’ 共very severe, severe, and so
extremely good conditions a value of 10. The questionnaire asked on兲; or ‘‘What is the frequency level of the occurrence of acci-
the experts to assign appropriate numerical condition 共C兲 values dents?’’ 共very high, high, and so on兲.
as poor-moderate-good. The values small-medium-large were The most probable values of the associated variables resulting
used for the frequency level of occurrence 共condition兲, small- from the value of the variable C in each factor are presented with
fuzzy joint membership functions, integrating the effect of the In Figs. 1 and 2, the variation in the value of variable C, ⌬C,
four variables. To measure the joint membership value for each doesn’t result in a significant change of P( f ) for either excava-
level of probability, the membership value for each level of vari- tion or pipe installation when other factors stay in good or mod-
able R, P f ⫽10⫺n is developed by evaluating six matrices for the erate condition. When the other factors stay in poor condition,
joint relation. The probability mass function of P f 共Zadeh 1968兲 however, the variation in P( f ), ⌬ P( f ), shows significant
is used to estimate the fuzzy probability of accidents in utility- changes. This means the probability of accidents in utility-
trenching operations. Finally, the mean value of the fuzzy prob- trenching operations with shielding systems is affected more by
ability of accidents, P f , can be calculated as the interaction of the performance of all these factors rather than
by one factor, and particularly when all are in poor condition.
N⫽⫺6⫻ P 共 P f ⫽10⫺6 兲 ⫺5⫻ P 共 P f ⫽10⫺5 兲 ⫺4⫻ P 共 P f ⫽10⫺4 兲 Thus the P( f ) value used for the simulation of accident effects is
limited to the case of the varied condition of one factor, ⌬C, with
⫺3⫻ P 共 P f ⫽10⫺3 兲 ⫺2⫻ P 共 P f ⫽10⫺2 兲 ⫺1⫻ P 共 P f ⫽10⫺1 兲 the others in poor condition, as shown in Table 10.
References
Downloaded from ascelibrary.org by Nat Inst Of Const Manage & Research Hyderabad (NICMAR) on 08/09/23. Copyright ASCE. For personal use only; all rights reserved.
Acknowledgments
Ayyub, B. M., and Eldukair, Z. A. 共1990兲. ‘‘Safety assessment methodol-
The writers express their special gratitude to Dr. Lefteri H. ogy for construction operations.’’ Structural safety and reliability, A.
Tsoukalas, professor and head of the School of Nuclear Engineer- H.-S. Ang, M. Shinozuka, and G. I. Schuëller, eds., ASCE, New York,
ing at Purdue University for providing review and insight regard- 771–777.
Bureau of Labor Statistics 共BLS兲. 共1997兲. ‘‘Perils in the workplace com-
ing the use of fuzzy mathematics in this paper.
pensation and working conditions.’’ Compensation and working con-
ditions, Census of Fatal Occupational Injuries, Washington, D.C.,
2共3兲.
Appendix. Fuzzy Set Operations with Four Bureau of Labor Statistics 共BLS兲. 共2001兲. Census of fatal occupational
Linguistic Variables „C, F, AC, R… injuries summary, U.S. Dept. of Labor, Washington, D.C., August 17.
Gambatese, J. A., and Stewart, P. J. 共1999兲. ‘‘Application of risk mapping
The four linguistic variables that characterize the safety factors— to construction project jobsites.’’ Implementation of safety and health
training, supervision, and preplanning—are on construction sites, Proc., 2nd Int. Conf. of CIB Working Commis-
1. Condition of factors affecting safety performance, referred to sion W99, Balkema, Rotterdam, The Netherlands, 785–792.
as C; Hinze, J. 共1997兲. Construction safety, Prentice-Hall, Upper Saddle River,
2. Frequency of occurrence of condition C, referred to as F; N.J.
3. Adverse consequences resulting from condition C, referred Lee, S. 共2001兲. ‘‘Simulation analysis of productivity variation affected by
accident risk in underground construction operations.’’ PhD disserta-
to as AC; and
tion, School of Civil Engineering, Purdue Univ., West Lafayette, Ind.
4. Frequency of accident occurrence R.
Levitt, R. E., and Samuelson, N. M. 共1993兲. Construction safety manage-
The major linguistic variables are C and F 共Ayyub and Eldu- ment, Wiley, New York.
kair 1989兲. The total effect of all factors 共training and so on兲 is the National Safety Council. 共1998兲. Accident facts, Itasca, Ill.
variable E, which is determined by taking the fuzzy union of the Occupational Safety and Health Administration 共OSHA兲. 共1995兲. OSHA
membership function U c⫻ac (c i ,ac j )⫽min关Uc(ci),Uac(ac j)兴, com- 2226, 1995–revised, U.S. Dept. of Labor, Washington, D.C.
bining the condition C 1 and the adverse consequences AC j . Staley, B. G., and Foster, P. J. 共1996兲. ‘‘Investigating accidents and inci-
Similarly, the total effects of the frequency of occurrence of the dents effectively.’’ Mining Technology, 78共895兲, 67–70.
C i , T, are also developed by taking the fuzzy union of Suraji, A., Duff, A. R., and Peckitt, S. J. 共2001兲. ‘‘Development of causal
U f ⫻r ( f j ,r k ), combining the frequency of occurrence of the C i , model of construction accident causation.’’ J. Constr. Eng. Manage.,
F j and the frequency level of the accident’s occurrence, R k . The 127共4兲, 337–344.
Tsoukalas, L. H., and Uhrig, R. E. 共1997兲. Fuzzy and neural approaches
standard fuzzy relation K, between the adverse consequences AC j
in engineering, Wiley, New York.
and the frequency level of the accident’s occurrence R k , is devel- Wilson, Jr., J. M., and Koehn, E. 共2000兲. ‘‘Safety management: Problems
oped based on fuzzy condition statements represented by ‘‘if encountered and recommended solutions.’’ J. Constr. Eng. Manage.,
AC 1 , then R 1 , else if AC 2 , then R 2 ... else if can, then R n .’’ The 126共1兲, 77–79.
fuzzy composition relation, M, between E and K, U eok (c i ,r k ) Zadeh, L. A. 共1968兲. ‘‘Probability of measure of fuzzy events.’’ J. Math.
⫽max ac j(min关UE(ci ,ac j),UK(ac j ,rk)兴), has been developed. Fi- Anal. Appl., 23, 421– 427.