Design and Deployment of a Remaining Useful Life Estimation Algorithm of Power Switches in a Cloud Computing Environment

This paper presents an improvement in the estimation of the Remaining Useful Life (RUL) of power switches towards the tail-end of their life cycle. The tail-end phase is crucial since it is during this time period that decisions on repair or replacement are made. The collector-emitter voltage across the power switch is considered the signal that could indicate a tentative upcoming failure. Feature extraction of this signal is thereby carried out to train machine learning models based on Long Short-Term Memory (LSTM) and Physics Informed LSTM (PILSTM) to estimate the RUL. Better estimation results compared to the state-of-the-art approaches are reported. Furthermore, the paper details the implementation of the developed ML algorithms on the Microsoft Azure cloud computing platform, providing a comprehensive guide for end-to-end deployment.