Research On Fault Diagnosis Method Of Industrial Process Based On Deep Learning

With the development of science and technology,industrial equipment presents the characteristics of complexity,informatization and intelligence,which also puts forward higher requirements for real-time fault monitoring and diagnosis.In recent years,deep learning has made great progress in industrial process fault diagnosis by virtue of its good feature learning ability.However,the data generated by modern industrial processes often have the characteristics of multivariable,nonlinear,high dimension,timing and multi-scale.Existing deep learning models still have many problems such as insufficient feature extraction and inefficient use of feature information.In order to solve the above problems and further improve the accuracy of fault diagnosis,this paper proposes two industrial process fault diagnosis models based on deep learning.The specific contents are as follows:(1)Aiming at the fact that the existing methods cannot fully extract the characteristic information in the data,and in order to avoid model degradation,an end-to-end multi-scale feature learning method based on model fusion is proposed for fault diagnosis of complex chemical processes.First,by combining CNN and residual learning,a MRCNN model is designed,and then it is cleverly integrated with the LSTM network,so that it can simultaneously extract non-linear high-dimensional spatial features of different scales and temporal features in the data,effectively reducing the loss of feature information and avoiding model degradation,thereby greatly improving the diagnostic effect of the model.(2)Aiming at the problem that the features extracted by the existing methods contain a lot of redundant information and the diagnosis accuracy is low when using the traditional softmax classifier,a new method of complex chemical processes fault diagnosis based on hybrid DRSN and XGBoost algorithm is proposed.This method combines the DRSN and XGBoost algorithms,and introduces the Nadam optimization algorithm to update the network parameters,so that it can effectively filter the redundant information in the features during training and improve the learning efficiency of the network.The traditional softmax classifier is discarded during classification,and the better-performing XGBoost classifier is selected,which significantly improves the final diagnosis accuracy of the model.