Health States Estimation and Prediction of Failure Occurrence Time of Rolling Element Bearing Using Hidden Markov Model

The fourth industrial revolution and internet of things applications have led to significant advancements in sensor technology and cloud services. Modern systems can easily gather, transport, and store large amounts of data using the advanced powerful processing capabilities and cutting-edge GPU. Such massive data may be exploited extravagantly for a variety of purposes, including the development of intelligent maintenance strategies. To ensure the safe and reliable operations of any industrial assets, prognostics and health management plays a significant role. Among all the aspects in prognostics and health management (PHM) used for efficient deployment of predictive maintenance, machine health states estimation and remaining useful life prediction are the most pivotal tasks. This study thus proposes a novel technique based on hidden markov model (HMM) for the health states estimations and prediction of the failure occurrence time of a roller element bearing. Given the success of HMMs with its wide variety of applications in speech recognition, language processing, and image processing. This study proposes to use it in the domain of machine prognostics as well. The three prime benefits of HMM are (i) the number of clusters is not required to be pre-determined, (ii) HMM works best with sequential data (iii) it can extract hidden states of the system that are not directly observable in the concerned problem. Finally, the efficacy of the proposed model has been tested on the IEEE-PHM-2012 challenge data sets. The results obtained are much better when compared with the existing literatures.