Fault detection of rotating machine parts using novel fuzzy neural network
S. Taniguchi, Dauren F. Akhmetov, Y. Dote · 2003
This paper proposes a novel fuzzy neural network for fault detection of rotating machine parts. Firstly, soft computing which is the fusion or combination of fuzzy systems, neural networks and genetic algorithms is studied. Then, by taking advantages of fuzzy systems and neural networks a novel fuzzy-neural network with a general parameter learning algorithm and system structure determination is developed. The network is based on one of local basis function networks. The general parameter method (GP) is based on GMDH (group methods of data handling). The GP is used for a learning algorithm and the structure determination of the developed fuzzy neural network. As the resulting network needs only fuzzy inference computation with GP calculations, which is, generally speaking, the combination of soft and hard computing, called computational intelligence, is suitable to solve nonlinear problems, it especially needs a little computation time. Therefore, it is easy to implement with a HITACHI RISC+DSP microprocessor fast enough for real time operations. The developed signal processor is self-organizing, self-tuning and automated designed. In order to confirm the feasibility of fault diagnosis performance by the developed network, it is experimentally applied to fault detection (diagnosis) of rotational machine parts (automobile transmission gears). It is found that the developed method is superior to other diagnosis methods by comparison.