Skip to Main Content
Reliable and fault free operation of machines needs not only timely fault detection and classification, but also an estimate of its remaining useful life, resulting in two phases of systems health monitoring, diagnosis and prognosis. Both share commonalities, with prognosis being the succeeding phase of diagnosis. In this paper, a prognosis algorithm based on the statistical hidden Markov model, is presented for the electrical faults of permanent magnet AC machines. The model parameters are computed by using the training outputs of the diagnosis phase. The algorithm estimates the failure state probability for each sampled observation. Time-frequency features extracted from the torque producing component of the machine current is used as the health indicator. The remaining useful life is estimated in terms of the probability of failure state. Parameter training of Hidden Markov Models generally need huge amounts of historical data, which are often not available in the case of highly reliable electrical machines. A method, which uses experimental observations, is presented for the computation of the state dependent observation probability densities from the limited data.