Long short-term memory network with multi-resolution singular value decomposition for prediction of bearing performance degradation

Abstract

Evaluating and predicting bearing performance degradation is essential to the reliability and safety of mechanical equipment systems. However, due to the complex working conditions, bearing vibration signals always suffer from serious noise, which is reflected in bearing degradation features and makes the performance prediction more difficult. To solve the problem, this paper proposes a novel method that utilizes long short-term memory network with multi-resolution singular value decomposition to predict bearing performance degradation. To explore feature expressions that are more conducive to trend prediction, the fault features from original vibration signals are enhanced, and multi-resolution singular value decomposition (MRSVD) is used for decomposition and reconstruction to accurately detect the fault point in vibration signals, and suppress the influence of interfering noise. Finally, a long short-term memory (LSTM) network is used to predict bearing performance degradation. Case studies with accelerated bearing degradation tests verified the effectiveness of the proposed method.