The use of hidden Markov models for condition monitoring electrical machines

This contribution is concerned with the application of a statistical pattern recognition method to the diagnostic function of electric machine condition monitoring. It describes the hidden Markov modelling technique (HMM), which uses historical data as a training set against which it constructs and tests models of the processes under observation. Operating under the classification mode it fits multi-sensor inputs to appropriate models which allow simple rule based decision making to take place. The technique may also be regarded as possessing the properties of a data fusion centre, making it very applicable to process monitoring and performance mapping of systems. A description of the basic hidden Markov method is given, and experimental results, which give evidence of its utility for monitoring the condition of electrical machines, are presented.