Fuzzy knowledge system for machine maintenance

This paper describes an application of a fuzzy knowledge-based neural network to the area of machine maintenance for the Mass Transit Railway Corporation (MTRC) of Hong Kong. The model utilizes expert knowledge and transforms this into fuzzy membership functions through control rules. An error backpropagation network was selected for the network training using various activation functions. After extensive training of the network, the use of FastProp with a hyperbolic tangent was recommended. The decomposition of input patterns to facilitate training was also suggested to overcome the long training time of the backpropagation network and eliminate the effect of local minima. Both the test and forecast results indicated that it is an excellent network for machine maintenance planning. The result shows a 20.08% improvement over the existing maintenance methodology. Integrating the above features, the proposed model can smoothly handle more types of industrial machine maintenance problems and generate benefits for MTRC in terms of improved customer service and better corporate image