Research On Process Fault Identification Based On Improved Depth Learning Model

Deep learning model is a powerful learning tool for feature extraction ability,because it can mine the information hidden in the raw data deeper than the special network structure,it is a new paradigm that deep learning model are applied to fault diagnosis model of chemical process.Therefore,this paper mainly studies the two fault recognition methods based on deep learning model are proposed on the system for complex multivariable conditions,and the proposed method is simulated by Tennessee-Eastman(TE)process.The effectiveness of the proposed method is analyzed by experimental results.For the multi-modality of the operational state data of modern industrial processes,the characteristics of nonlinear,single-sample data are less and other characteristics,the industrial process fault identification model based on three kinds of data dynamic expansion(independent component analysis(ICA),principal component analysis(PCA),Gaussian Noise)convolutional neural network(CNN)is proposed.First,the variables are reconstructed by selecting faults and the data is expanded by the training set.Then,the preprocessed data is input into the constructed convolutional neural network model,the training effect is observed.Finally,the data expansion is verified by the recognition effect of the process fault data,the rationality and effectiveness of the latter CNN model identification method,the recognition effect of the input fault variable structure changes indicates the limitations of the CNN model.For the problem of the limitation of the CNN model,the low-efficiency feature extraction of fault variables,the number of classifications is small and the fault recognition rate is low in complex industrial processes,based on asymmetric convolution kernels of the industrial process of convolution neural network(CNN)fault diagnosis model was proposed in this paper.The model adopts fault variable reconstruction to preprocess the fault data;randomly select the training set for the processed fault data and label the fault data of different categories;the asymmetric convolution kernel model is introduced to extract the features of the input fault variables,improve the efficiency of feature extraction;The TE(Tennessee-Eastman)process fault online test set sample is identified based on the CNN model improved AC-CNN model with AC architecture,the experimental results show that the proposed method has obvious recognition effect on the TE process fault dataset,which demonstrates the effective and excellent of the model.