Study On Long Short-Term Memory Network For Fault Diagnosis And Remaining Useful Life Prediction

Modern machinery and equipment are becoming more and more complex,automated and sophisticated,and its safe operation is particularly important.Therefore,the study on diagnosis and prediction of mechanical failures has attracted much attention,and data-driven methods based on vibration signal analysis are widely used.The traditional intelligent diagnosis and prediction methods depend on the quality of the extracted features and were greatly affected by prior knowledge and diagnosis experience.In recent years,the introduction of deep learning has provided new ideas for fault diagnosis and prediction.This paper mainly studies fault diagnosis and remaining life prediction based on deep learning of recurrent neural network.Firstly,aiming at the problems that traditional methods are difficult to adaptively extract sensitive features and cannot fully utilized the time information of failure evolution,a bearing fault diagnosis method based on convolutional network and long-short term memory network is proposed.This method constructed a deep neural network model to adaptively extract the robust features from original bearing signals,and then utilized the long short-term memory network to learn the time-dependent relationship in these features.Finally,the fault classification is completed by analyzing the timing characteristics.The proposed method overcame the problems existed in the traditional feature extraction methods,such as heavy dependence on expert experience and incomplete utilization for time series information,and realized intelligent and accurate diagnosis of faults.Secondly,in view of the requirement for continuous-time characteristics in the remaining life prediction and the incapability of learning the complete data information,this paper studied a remaining useful life prediction method based on deep bidirectional LSTM network.The monitoring data was preprocessed by a sliding time window and then as the input of network.The deep neural network accomplished the extraction of high-dimensional features from sensor data and prediction of remaining useful life by learning the degradation information,and this method has achieved better prediction performance than other machine learning methods.The methods proposed in this paper were verified on the Western Reserve bearing Dataset and turbofan engine Dataset,respectively.In addition,the effects of key parameters were analyzed and compared with the commonly used methods.The experimental results showed the effectiveness of the research methods.