Research On Fault Diagnosis And Prognosis Of Industrial Equipment Based On Deep Learning

With the rapid development of modern industry,the complexity,precision and intelligence of industrial equipment are increasing,the operating environment and conditions of industrial equipment are also more complex.The planned maintenance and post-maintenance of equipment have problems such as excessive maintenance and insufficient maintenance,which can no longer meet the needs of modern industry.Prognosis and Health Management(PHM)is a preventive maintenance approach,aiming to find faults,determine fault types and predict the time of fault occurrence timely and accurately.Fault detection,fault diagnosis and fault prognosis are important contents of PHM,among them,fault detection is the basis of fault diagnosis and fault prognosis,fault prognosis is the core of PHM and more advanced maintenance based on the first two.This paper focuses on the data-driven PHM approach,which is based on deep learning for fault detection,fault diagnosis,and fault prognosis of industrial equipment and its key components,so as to improve the reliability and safety of industrial equipment,reduce the maintenance cost of industrial equipment,and help enterprises optimize maintenance decisions.The main research contents of this paper are as follows:(1)Aiming at the problems of poor timeliness and low accuracy in fault detection of complex industrial equipment,a fault detection method of Improved Structure based Transformer(IST)model is proposed.The important features are automatically extracted by the improved Transformer structure,the relationships between features and features,between labels and labels,between features and labels are deeply mined.The experimental results show that the proposed method can detect the early faults of complex industrial equipment quickly and accurately,and has good timeliness and accuracy.(2)Aiming at the problems of fault class imbalance and multi-fault coupling in fault diagnosis of complex industrial equipment,a multi-label fault diagnosis method of Improved Dual Attention based Transformer(IDAT)model under class imbalance is proposed.The combined oversampling method of ADASYN and Borderline-SMOTE1 is proposed to balance the positive and negative fault samples,the improved Dual Attention mechanism is used to extract important features from multiple dimensions,and the binary relevance method is used to reduce the difficulty of training multi-label model.The experimental results show that the proposed method can accurately diagnose multiple faults occurring simultaneously under fault class imbalance,and has good universality and robustness.(3)Aiming at the problems of complex conditions interference and uncertainty effects in fault diagnosis of complex industrial equipment’s key component,a fault diagnosis and uncertainty quantification method of Improved Mixed Attention based Transformer(IMAT)model under complex conditions is proposed.The improved Mixed Attention mechanism is used to extract the important features and their influence from different dimensions simultaneously,the K-means condition recognition method is proposed to reduce the interference of the conditions on the model,and the MC Dropout approximate Bayesian method is used to construct confidence intervals to quantify the uncertainty of the model.The experimental results show that the proposed method has high accuracy,reliability,and generalization under complex conditions.