FUZZY NEURAL NETWORKS FOR MACHINE MAINTENANCE IN MASS-TRANSIT RAILWAYSYSTEM

This paper describes an application of fuzzy knowledge-based neural-ne twork system (FKNNS) being developed by the Hong Kong Mass Transit Rai lway Corporation (MTRC) for the maintenance of its ticket machines in one of the busiest transit systems in the world. The model utilizes sp ecific experts' knowledge which is transformed into fuzzy membership f unctions through certain control rules. The error backpropagation netw ork was selected for the network training in which various activation functions were tested. After extensive training of the network, the Fa stProp with hyperbolic tangent was recommended. Input patterns were de composed to facilitate the training process and eliminate the effect o f local minima. Both the test and forecast results indicated that the FKNNS is an excellent aid for machine maintenance planning since there are too much difficulties in deriving the analytical solution the oth erwise. Beta test result shows a 20.08% improvement over the existing maintenance methodology. Moreover, the developed model can smoothly ha ndle more types of industrial machine maintenance problems and generat e intangible benefits toward MTRC in terms of improved customer servic e and better corporation image.