A new Approach for Remaining Useful Life Prediction of Bearings Using 1D-Ternary Patterns with LSTM

Abstract

AbstractBearings are one of the components that frequently malfunction in mechanical systems and their failure directly affects the system's performance. Therefore, accurately predicting bearing failures helps personnel with maintenance planning and prevents unexpected failures. Data-driven prognostic techniques are commonly used to predict the remaining useful life (RUL) of high-speed bearings. RUL prediction relies on determining the fundamental relationship between bearing degradation and current health status, and its accuracy depends on the effectiveness of the features extracted from the bearing. In this study, a new approach has been proposed to predict the remaining life of bearings. Two different feature vectors, LOWER and UPPER, are obtained by applying the 1D-TP method to vibration signals, and RUL prediction is performed using LSTM. The proposed approach has been tested on a dataset obtained from the PRONOSTIA platform, and performance metrics such as MAE, RMSE, SMAPE, RA, and Score values have been determined. The results show that the 1D-TP + LSTM method helps to successfully predict the remaining life of bearings. As a result, accurate RUL assessment or reliability analysis will help personnel make appropriate maintenance decisions, prevent losses due to mechanical system damage, improve production safety, and prevent damage to the mechanical system.