Journal bearing seizure degradation assessment and remaining useful life prediction based on long short-term memory neural network

Abstract

Accurate bearing degradation assessment and remaining useful life (RUL) prediction may effectively avoid major disasters in manufacturing. With the rapid development of the computer industry, deep learning has emerged as a reliable algorithm for time-series prediction and has shown good performance. In this paper, the journal bearing seizure experiment was performed. The collected multi-sensor failure dataset is used for feature extraction and degradation indicator (DI) construction. The DI and working condition information are applied for degradation stage (DS) division by the fuzzy c-means (FCM) algorithm. Considering the transition of different DSs, the one-stage and multi-stage iteration prediction models based on the Long Short-Term Memory (LSTM) neural network for RUL prediction are proposed. The particle swarm optimization (PSO) is used to optimize model hyperparameters. The results show that the multi-stage iteration prediction may achieve the early warning of seizure failure and outperform the one-stage iteration prediction and traditional machine learning prediction.