A hybrid approach to the automatic design of adaptive fuzzy control and diagnostic systems

After a brief presentation of the technique of dynamic optimization of fuzzy reasoning methods, a hybrid approach to determine the most adequate reasoning strategy to be applied to a fuzzy control/diagnostic system from a set of available alternatives is proposed. This approach consists of two steps. First the automatic extraction of relevant knowledge for the design of the reasoning method used in the rule-base of the fuzzy control/diagnostic system is performed. This step is based on the ability of the dynamic switching fuzzy system introduced by Smith (1994) to cluster automatically the reasoning parameters (fuzzy operators or/and defuzzification methods) which assure an optimal performance of the respective fuzzy rule-based system. Then, in a second step, fuzzy meta-rules for tuning the control/diagnostic system are manually extracted by the human expert from the reasoning tables obtained in the first step. Illustration on an inverted pendulum shows the ability of this approach to cope with complex control and diagnostic situations as automatic rejection of unknown disturbances and sensor failures. Application to the control of the acrobot and to the diagnosis of a DC motor prove the ability of this technique to manage critical tasks in engineering of continuous dynamic systems.