Dynamic systems reliability evaluation using uncertainty techniques for performance monitoring

The authors suggest the use of fuzzy measures and fuzzy integrals in evaluating the reliability of control systems by approximating an expert´s view on a complex system when assessing the performance. The class of systems considered have structural complexity exhibiting a closed form model of the underlying process. The approach may be described in three parts where in the first stage a rule based classifier (`spy´) extracts `states of performance´ from the process. It is shown that the rule premise resembles a possibility based control chart and that the possibilistic version, embedded in a rule based system offers a comprehensive man process interface while having a similar or slightly improved speed of detection. The reliability can be quantified based on a finite set of abstract states over which a certainty measure is defined. A prediction for a specified reliability interval of time is done by using a qualitative model akin to Markov stochastic processes and consequently decisions are made to alter the system structure. This framework allows distinct classes of uncertainty to be considered