Residual Life Prediction of Mechanical Equipment Based on Feature Learning in Big Data Background

For the traditional data-driven methods of remaining useful life prediction, health indicators are generally built with some hand-crafted feature extraction methods. However, in the big data era, hand-crafted feature extraction methods require specific expert knowledge and time-costing. In order to solve this problem, a feature learning based method (adaptive feature learning based remaining useful life prediction, AFLRULP) was proposed to predict remaining useful life of the mechanical equipment. Firstly, a window data matrix was constructed to solve the data fluctuation problem. Then, a one-dimensional convolutional neural network with multi-layers was built to map the data matrix to the health conditions of the mechanical equipment. Finally, the remaining useful life was predicted with a failure threshold. In order to verify the effectiveness of the proposed method, it is validated by a bearing life-cycle dataset. Additionally, the proposed method was compared with some manual feature extraction based methods. The results show that the proposed method (AFLRULP) does not need to manually extract features, it can map the original monitoring data to the performance status and remaining life of mechanical equipment. Additionally, compared with the existing life prediction methods based on manual feature extraction, the proposed method improves the cumulative relative accuracy of bearing life prediction by an average of 0.20.