Abstract

Monte Carlo simulation has been used extensively for addressing probabilistic uncertainty in range estimating for construction projects. However, subjective and linguistically expressed information results in added non-probabilistic uncertainty in construction management. Fuzzy logic has been used successfully for representing such uncertainties in construction projects. In practice, an approach that can handle both random and fuzzy uncertainties in a risk assessment model is necessary. This article discusses the deficiencies of the available methods and proposes a Fuzzy Monte Carlo Simulation (FMCS) framework for risk analysis of construction projects. In this framework, we construct a fuzzy cumulative distribution function as a novel way to represent uncertainty. To verify the feasibility of the FMCS framework and demonstrate its main features, the authors have developed a special purpose simulation template for cost range estimating. This template is employed to estimate the cost of a highway overpass project.

Introduction

Performing risk analysis using Monte Carlo simulation is very common in construction project management; traditionally, probability theory is used to model uncertainty regarding simulation model inputs. In practice, the probability of an event can be estimated according to the frequency of that event occurring in a number of experiments (Pedrycz, 1998). However, if the number of experiments is not large enough to be significant, and more experiments cannot be performed, it is not possible to accurately estimate the event’s probability. In these circumstances, we can engage human experts who are usually good at supplying the required information. Some researchers try to convert experts’ knowledge into probabilistic distributions. This estimated probability may be used directly in a risk analysis problem (Ahuja et al., 1994), or it may be combined with available data using Bayesian methods to estimate a parameter that considers both subjective judgment and historical data (Garthwaite et al., 2005). However, there are some criticisms on performing probabilistic analysis on subjective and linguistically expressed data because subjective reasoning of individuals may not be appropriate for objective scientific conclusions (Goldstein, 2006). In other words, the information gained from experts is subjective and contains ambiguity, and there is a chance of introducing artificial knowledge that is not actually available to the model using probability values gained from experts (Guyonnet et al., 2003).