Adaptive Fault Diagnosis Method Based on FNN

Abstract

In complex manufacturing, the system parameters have dynamic and nonlinear characters. Existing parameters setting methods show low efficiency and accuracy, and some setting experience accumulated in engineering practice can not be fully used. Therefore, an online parameter setting method with improved adaptive neuro-based fuzzy inference model is proposed in this paper. The advantages of ANFIS in self-learning and fuzzy inference method of fuzzy mathematics are combined effectively to determine the fuzzy rules and fuzzy membership. Then the neural network predictor is introduced to enhance the adaptive ability and nonlinear approximation ability of fuzzy neural network. The parameters and the structure of fuzzy rules are regulated and updated by the nearest neighbor clustering algorithm, which improves the accuracy of fuzzy inference and enhances the fault-tolerance and robustness of system. In our experiments, by the simulations of transformer fault diagnosis and soldering technology, we find improved ANFIS model can better reflect the actual operating state of transformer and it can make modeling and control for nonlinear I/O system in the manufacturing process like soldering. The improved system is also proved by experiments to have better performance in robustness and convergence.