A Trainable Transparent Universal Approximator for Defuzzi cation in Mamdani Type Neuro-Fuzzy Controllers

|A novel technique of designing application speci c defuzzi cation strategies with neural learning is presented. The proposed neural architecture considered as a universal defuzzi cation approximator is validated by showing the convergence when approximating several existing defuzzi cation strategies. The method is successfully tested with fuzzy controlled reverse driving of a model truck. The transparent structure of the universal defuzzi cation approximator allows to analyse the generated customised defuzzi cation method using the existing theories of defuzzi cation. The integration of universal defuzzi cation approximator instead of traditional methods in Mamdani type fuzzy controllers can also be considered as an addition of trainable non linear noise to the output of the fuzzy rule inference before calculating the defuzzi ed crisp output. Therefore, non linear noise trained speci cally for a given application shows a grade of con dence on the rule base, providing an additional opportunity to measure the quality of the fuzzy rule base. The possibility of modelling Mamdani type fuzzy controllers as a feed forward neural network with the ability of gradient descent training of the universal defuzzi cation approximator and antecedent membership functions ful ll the requirement known from Multi-layer Preceptrons in nding solutions to non linear separable problems. Keywords|Fuzzy control, Fuzzy neural networks, Mechatronics, Defuzzi cation