A hybrid reasoning system for the prevention of rail accidents

International audienceThis article describes a contribution to improving the usual safety analysis methods used in the certification of railway transport systems. The methodology is based on the complementary and simultaneous use of knowledge acquisition and machine learning. To demonstrate the feasibility and soundness of the proposed approach to assist in the analysis and evaluation of railway safety, we have developed a software tool. This tool is composed of two main modules: A module for the classification and capitalization of historical accident scenarios and a module for evaluating the completeness and consistency of the new scenarios proposed by the manufacturer of the transport system. The first classification module is an inductive, incremental and interactive learning system. The second evaluation module, which is based on the use of a rule-learning system, aims to provide experts with suggestions of potential failures which have not been considered by the manufacturer and which are capable of jeopardizing the safety of a new rail transport system. In contrast to traditional failure Diagnosis Systems, the developed software tool can be seen as an aid which reveals potential risk during the design stage of rail system