Hidden Markov Models for failure diagnostic and prognostic

This paper deals with an estimation of the Remaining Useful Life of bearings based on the utilization of Mixture of Gaussians Hidden Markov Models (MoG-HMMs). The raw signals provided by the sensors are first processed to extract features, which permit to model the physical component and its degradation. The prognostic process is done in two phases: a learning phase and an evaluation phase. During the first phase, the sensors' data are processed in order to extract appropriate and useful features, which are then used as inputs of dedicated learning algorithms in order to estimate the parameters of a MoG-HMM. The obtained model represents the behavior of the component including its degradation. In addition, the model contains the number of health states and the stay durations in each state. Once the learning phase is done, the generated model is exploited during the second phase, where the extracted features are continuously injected to the learned model to assess the current health state of the physical component and to estimate its remaining useful life and the associated confidence. The proposed method is tested on a benchmark data taken from the “NASA prognostic data repository” related to bearings used under several operating conditions. Moreover, the developed method is compared to two methods: the first using traditional HMMs with exponential time durations and the second using regular Hidden Semi Markov Model (HSMM). Finally, simulation results are given and discussed at the end of the paper.