Early Detection of Gas Dispersion Accident through a Neural Network Based Expert System

Typical major accidents are gas dispersion, fire (pool, flash, jet) and explosion (UVCE, VCE). Gas dispersion accidents are the most dangerous because of the dimension of the impact area or of the potential fire or explosion which can be generated in case of ignition. Human interpretation of fault that could result in an accident is normally based only on a part of the incoming information, so a forecast of the event evolution is very difficult. The in-site gas detectors could give a big amount of information about an accident (the substances involved, the concentration, the leak position, the gas mass flow, the loss duration) since the earliest phases of event evolution; these information, integrated with environmental and meteorological conditions, concur to define effective and efficient protective actions. This work describes how we can use data registered by gas detectors in order to know evolution characteristics of dispersion and to quickly identify the real event among the potential accident events reported in the risk analysis of the Safety Report.