EANN 2014: a fuzzy logic system trained by conjugate gradient methods for fault classification in a switch machine

Abstract

This paper discusses and analyzes the performance of a technique for classifying possible faults (lack of lubrication, lack of adjustment and malfunction of a component) that can occur in an electromechanical switch machine, which is an equipment used for handling railroad switches. This technique makes use of a type-1 and singleton fuzzy logic system trained through the conjugate gradient method (i.e., second-order information is now considered). Combinations of feature extraction techniques based on higher-order information, feature selection technique based on Fisher's discriminant ratio and three classifiers (Bayes based, multilayer perceptron neural network and type-1 and singleton fuzzy logic system) show the effectiveness of the discussed technique when a data set provided by a company of the Brazilian railway sector, which addresses the possible faults in a switch machine, is considered. Additionally, the reported results show that the type-1 and singleton fuzzy logic system trained by the conjugated gradient method can offer higher convergence rate and performance for a limited number of epochs than that one trained by the steepest descent method. Finally, but not the least, based upon the attained results, the proposed technique enables the railway company to adopt solutions to achieve operational excellence.