Remaining Useful Life Prediction for Circuit Breaker Based on SM-CFE and SA-BiLSTM

Abstract

Targeting at the uncertainty of the degradation of conventional circuit breakers (CCBs) and the perfect mechanical degradation characterization by vibration signals, a remaining useful life (RUL) prediction method based on similarity measure (SM), comprehensive features extraction (CFE) and self-attention bidirectional long short-term memory network (SA-BiLSTM) is proposed in this paper. Firstly, the degradation-sensitive intrinsic mode functions (IMFs) of the vibration signals are separated by the SM method, and the reconstructed signals are divided into intervals based on the working process of CCBs to obtain the vibration signals for life prediction. Secondly, the explicit parametric features (EPFs) are obtained based on the vibration events and waveform features, and the implicit parametric features (IPFs) are extracted using the multi-scale convolutional auto-encoder (MSCAE). Next, the comprehensive features are obtained by combination of the EPFs and IPFs. On this basis, the degradation mode judgment method is formed by using the least square method (LSM) linear fitting of the EPFs. Finally, a quantitative life prediction model based on SA-BiLSTM is constructed, and greater weights are assigned to the important time steps. The proposed method is proved to show high prediction accuracy and good stability, which is more advantageous compared with other hybrid prediction models.