At present, the life model based on statistics still plays a leading role in bearing life prediction, but it is found in test and engineering application that the life calculated by the statistical life model is usually conservative and the dispersion of bearing life is large. Therefore, how to improve the bearing life model through the study of bearing performance degradation mechanism is a main problem. With the development of new information technology and artificial intelligence, the life prediction method based on condition monitoring has become a hot research field of bearing life prediction. With the help of big data, artificial intelligence information and other technologies, we can obtain the dynamic signal reflecting the service performance of the bearing, obtain the signal characteristic parameters characterizing the decline of the bearing performance, and establish the mapping relationship between the signal characteristic parameters and the remaining life, so as to realize the prediction of the remaining life. However, there is a lack of appropriate characteristic parameters to measure the evolution law of gradual decline of bearing performance during operation, and compared with the traditional life prediction model, artificial intelligence methods such as neural network have unclear physical meaning and large influencing factors. How to carry out in-depth research on the difficulties is very important for bearing life prediction technology.