Composite Fault Prediction of Bearing Based on Cascaded Long Short Term Memory Neural Network

Abstract

At present, resonance demodulation and data envelopment analysis, etc are usually used for health evaluation and remaining life prediction of bearings, but there are some problems such as difficulty in extracting health degree, single fault prediction type, and lack of reliable models. Aiming at these deficiencies, this paper proposes a method which realizes bearing health assessment and remaining life prediction through two-stage cascaded long short term memory neural network. In this paper, the bearing data set of Xi'an Jiaotong University is used for experimental comparison. The results show that, compared with the resonance demodulation method, the monotonicity, robustness and trend of the health degree curve are comprehensive in the health degree evaluation of single fault data using this method. The evaluation indicators increase by 12%, 24.8% and 5% respectively, and the comprehensive evaluation index in the composite fault data increases by 15.1%, and the remaining service life can be predicted according to the health evaluation curve, which proves that the method is effective in health evaluation and remaining life.

Keywords

• neural network
• lstm (long short term memory)
• compound fault
• remaining service life
• predictive maintenance