Multi-bearing remaining useful life collaborative prediction: A deep learning approach

Abstract

Rolling bearing health analysis and remaining useful life prediction have become an increasingly crucial research area that can promote reliability and efficiency in the modern manufacturing industry. Internet-of-Things and cyber manufacturing techniques make it convenient to collect large volumes of sensor data that can provide powerful support for efficient data analytics such as deep learning. The combination of a massive amount of available data and advanced machine learning models brings new opportunities for bearing remaining useful life prediction. This paper proposes an integrated deep learning approach for multi-bearing remaining useful life collaborative prediction by combining both time domain features and frequency domain features. The method can extract high-quality degradation patterns of rolling bearing from vibration signals. Regarding features extracted from bearing vibration signals, in addition to three conventional time domain features, a novel frequency domain feature is adopted in the proposed method as well. Based on the extracted features, the deep neural network model is introduced to predict the remaining useful life of rolling bearing. We evaluate the performance of the proposed method on a real dataset and compare it with several commonly used shallow prediction methods Numerical experiment results show the effectiveness and superiority of the proposed approach.