Prediction of remaining life of bearings based on integral correction and global attention mechanism

Abstract Rolling bearing failure is an important cause of equipment failure, and its remaining service life prediction is a key research direction to detect the health status and service life of machinery. However, the feature indicators extracted based on expert knowledge cannot better characterize the trend of bearing degradation in the early and middle term, and the neural network has poor prediction effect for longer sequences. To address the above problems, a method is proposed to predict the remaining life of rolling bearings based on integral transformation and global attention mechanism. Firstly, the time domain feature indicators of vibration signals are integrally corrected and transformed into integral corrected indicators to solve the problem of inconspicuous information of bearing degradation in the front and middle term; then, a long and short time memory neural network based on global attention mechanism is built to deeply explore the mapping relationship between the integral feature indicators of long sequences and the remaining life of bearings, which gives full play to the advantage of global attention on weight optimization allocation; finally, a linear regression function is used to construct health indicators and achieve the prediction of the remaining life of rolling bearings. Experiments were conducted on the bearing data of PRONOSTIA and compared with other methods, and the results showed that the method has better accuracy and prediction precision than other methods.