Study On Gated Recurrent Unit And Convolution Auto Encoder For Fault Diagnosis And Remaining Useful Life Prediction

With the development and application of mechanical equipment,it plays an increasingly important role in social activities.The normal operation of mechanical equipment is of great significance to ensure social production and life.Traditional fault diagnosis methods are difficult to deal with big data,and the modeling process requires a lot of prior knowledge and expert experience.The emergence of deep learning provides a new idea for fault diagnosis and prediction.Based on the Gate Recurrent Unit and convolutional self encoder,this paper studies the fault types and residual service life prediction of mechanical equipment.The main research contents are as followsFirstly,considering that the mechanical equipment can not obtain the label data in the actual operation process,and the traditional fault diagnosis methods do not make full use of the time sequence features in the data,this paper studies the fault diagnosis method of deep Gate Recurrent Unit based on convolutional self coding.This method uses the unsupervised learning feature of convolutional self coding to automatically learn fault features from unlabeled training data,and uses deep Gate Recurrent Unit to complete the learning of time sequence features.Finally,the time sequence features are analyzed by the classifier to complete the fault classification.This method overcomes the problems of insufficient label data and incomplete utilization of data sequence information in practical application,and achieves better fault diagnosis performance.Secondly,in the process of remaining service life prediction,considering that there are few prediction methods using deep data features,this paper uses convolution self coding to learn the deep features in the data,and optimizes the features through attention mechanism to enlarge the weight of important features.Then,the optimized deep feature data is fused with the original data,and input to the Gate Recurrent Unit for temporal feature extraction;Finally,the time series features are mapped to the label space through the full connection layer to realize the prediction of the remaining service life,and the prediction performance of this method is analyzed by comparing with other methods.The fault diagnosis method is verified in the bearing data set provided by Western Reserve University,and the remaining service life method is verified in the phm2012 data set provided by IEEE 2012 PHM data challenge and the C-MAPSS data set provided by NASA.The influence of network parameters on the model performance is discussed,The effectiveness of the proposed method is verified.