Failure prognosis of DC starter motors using hidden Markov models

This paper deals with the prognosis of gear faults in DC machines using time frequency distribution methods. The proposed method presents future state prediction of the machine faults using Hidden Markov models. Different methods for estimating the parameters of hidden Markov model with limited data are discussed. The proposed method uses Matching Pursuit decomposition and projections of the training data on linear discriminant planes for estimation of model parameters. Experimental results are presented to illustrate the method.