Research On Bearing Fault Prognostics Algorithm Based On Transfer Learning

With the advent of the era of big data and artificial intelligence,the health maintenance work in industrial systems has gradually changed from manual maintenance to intelligent maintenance era.There are a large number of rotating mechanical parts in industrial systems,among which the failure rate of bearings is high.However,the traditional bearing health maintenance methods have various problems such as poor timeliness and inaccurate positioning.People hope to improve the reliability and stability of bearings in industrial systems and complete the predictive maintenance of bearings through intelligent management and intelligent diagnosis.Therefore,the Prognostics and health management(PHM)technology came into being.Guided by bearing PHM technology,this paper studied bearing fault prognostics algorithm based on transfer learning to improve the prediction ability of bearing fault prognostics algorithm.The main research contents were as follows:(1)Based on the bearing vibration data collected by vibration sensors,the feature extraction algorithm of bearing vibration data based on convolution autoencoder was studied.In this paper,one-dimensional convolution autoencoder was used as feature extractor,and the hidden layer output of convolution autoencoder was used as the features which were extracted,so that the feature extraction of bearing vibration data could be realized and the error of manual labelling could be avoided.(2)Based on the features extracted from convolution autoencoder,the bearing health indicator construction algorithm based on support vector regression was studied,and the bearing health indicator with better monotonicity and trend was constructed,which effectively reflected the health status of bearings in the whole life cycle,and also verified the feature extraction ability of convolution autoencoder for bearing vibration data.(3)Based on the features extracted from convolution autoencoder,the remaining useful life prognostics algorithm based on transfer learning was studied.This paper proposed a neural network model with multi-scale model structure,multi-scale attention mechanism,channel attention mechanism and spatial attention mechanism,and adopted the method of parameter transfer learning to improve the prediction ability of the prediction algorithm.