Bearing Fault Predictive Maintenance using LSTM

In the recent times, industrial automation has accelerated the use of sophisticated and contemporary machine tools. Rotating machines are the most critical equipment in today's modern industry set-ups. Operating conditions of these equipment must be maintained in order to avoid production interruptions and financial loss. Rotating machines operate for extended periods of time, and as a result, they frequently experience mechanical and electrical stresses. Because of these electrical and mechanical stresses, various parts of rotating machines will fail. Machine Learning and Deep Learning algorithms have been used effectively in industry for predictive maintenance of machines to increase the operational reliability and reduced costs. In this research paper, the problem of predictive maintenance using the IMS bearing dataset from NASA Prognostic Data Repository is undertaken and Long-short term memory (LSTM) model is deployed. For this purpose, root means square (RMS) value is taken as condition indicator of the machine which is provided the information about machine’s health. KernelRidge regression is introduced for comparing its result with LSTM based model. After exhaustive simulation and the simulation performance measures in terms of root means square error (RMSE) score of training, validation and testing set are 0.0017, 0.0101 and 0.0012 respectively for proposed LSTM model.