Overcoming Uncertainties in Risk Analysis: Trade-Offs among Methods of Uncertainty Analysis

Abstract

Industrial risk analysis suffers from many problems of uncertainty due to the difficulty of estimating various parameters of concern for the analysis. In the real world, we usually use many qualitative and/or uncertain parameters for risk evaluation. While the quantification of parameters is an important task, it is usually practiced according to the experience of the analyst (discrete approach) or by using probabilistic models (probabilistic approach). The discrete approach is very limited because it does not take into account the variability or the uncertainty of parameters. On the other hand, the probabilistic approach requires knowledge of the parameter’s statistical distribution, which may be very difficult or even impossible. Furthermore, in both approaches, qualitative variables are not easy to deal with. Over years of research, we have developed a general approach to overcome these problems. The estimation of parameters and the treatment of available data are based upon fuzzy logic models, with some improvements in the fuzzy reasoning mechanism. This paper presents a comparison between our fuzzy approach and the discrete and probabilistic approaches. A geotechnical application was developed to evaluate the risk of natural ground movements in a rock cliff that would have severe impact on the surrounding environment. We have ended up with a general approach to the problem of uncertainty and with some recommendations on how to approach different parameters according to their nature (using either the discrete, probabilistic or fuzzy method). The improvements we have made to the fuzzy reasoning process (beta cuts reasoning technique) has been approved by specialists in the domain of fuzzy logic and are applicable to all branches of science.