Contribution à l’application des techniques de l’intelligence artificielle à l’analyse quantitative des risques industriels

Abstract

Today, the industrial world has become more sensitive to the control of major accidents because of the serious and even catastrophic material, human and environmental consequences. For this, considerable efforts are made in terms of risk management in order to prevent these accidents. However, as our national economy is based on the very high risk hydrocarbon industry, where large quantities of flammable, explosive and toxic liquids and gaseous products, etc., are processed, stored and transported, the threat of an accident is always imminent, even no space is safe from disasters associated with complex oil and gas industrial installations. As an example, we can cite the catastrophic explosion of the Skikda liquefaction complex in January 2004, which in addition caused very significant material damage, 27 deaths and 73 injuries, not to mention the psychosis installed in the inhabitants and residents. In order to control these phenomena, several techniques and mathematical models of risk prediction have been developed. Among these methods, we find quantitative risk analysis (QRA); the latter is a rigorous approach and is essential for a good estimation and management of industrial risks. It mainly consists of identifying potential accident scenarios, estimating their frequencies and analyzing their consequences. The purpose is to estimate individual and societal risks and, therefore, to study and implement effective measures that adequately respond to this estimate. In some cases, the data may be available and known with precision by referring to experience feedback. However, these data are not always suitable for the analysis of rare and often complex events such as major accidents for which the statistical data are not satisfactory. Databases and expert judgments are another source that provides data used by risk analysis methods, but which is also fraught with uncertainty and imprecision. Fuzzy approach can offer a very adequate framework for the representation and treatment of these uncertain and / or imprecise aspects.