Evaluating case-based reasoning and evolution strategies for machine maintenance

Outlines a study to evaluate case based reasoning (CBR) and evolution strategies (ES) for machine maintenance in the Mass Transit Railway Corporation (MTRC) of Hong Kong. It utilizes specific expert's knowledge, which is transformed into case-base and fuzzy membership functions through certain control rules. Three learning algorithms: adaptive gradient learning of CBR, time series prediction using a time lagged recurrent network (TLRN), and a radial basis function (RBF) neural network of ES were investigated. To improve the learning procedure, constructive backpropagation is adopted to develop a case-based reasoning network. The same database as in Baluja (1994) was applied to the present study. Experimental results indicate that TLRN is the best in terms of training result. It has achieved an improvement of 99% and 274% against CBR and RBFs respectively. Compared with that in the above paper, there is 651% improvement on the model which was based on a genetic algorithm with FastProTank learning. An integration of CBR and ES to further improve the automation of the scheduling process for machine maintenance is undergoing.