Fault Diagnosis Of Chemical Process Based On Convolutional Neural Network Research

The chemical production process is characterized by high non-linearity,uncertainty,vulnerability to interference and correlation,etc.The chemical production process is also a special dynamic system.Its environment is different from other types of industrial production,and the production environment is unstable and dangerous,making the safety management in the production process very difficult.In order to run the chemical equipment normally,it is necessary to detect faults in time and diagnose faults accurately.A fault diagnosis method based on convolutional neural networks(CNN)is proposed for high dimensional non-linearity of chemical process in this paper.After wavelet transformation for denoising and standardized preprocessing of data,CNN can extract the hidden information features from the original data of chemical process through feature extraction and learning.And then carries out fault classification through softmax classifier.According to the results of fault diagnosis research applied to TE process,the average fault detection rate is 80.31%,and the false alarm rate is 2.837%,indicating that the proposed convolutional neural network method is effective and can detect faults in a timely manner.In view of the complex structure of CNN itself and such problems as its own parameters and structure needed to be optimized,a particle swarm optimization algorithm is proposed to optimize the super parameters of convolutional neural network.In order to verify the effectiveness and performance of the fault diagnosis of the PSO-CNN method,the TE process was taken as an experimental case for the fault diagnosis study.The optimal structure and parameters of the convolutional neural network model are determined in the process of particle swarm optimization.According to the experimental results,the average fault detection rate based on the PSO-CNN method is increased to 84.72% compared with the CNN method,and the false alarm rate is only 2.26%.The average fault detection rate is higher than that of PCA,KPCA and MICA.And the fault detection speed is also faster than the CNN method,indicating that the PSO-CNN method is more sensitive to some certain faults.Therefore the proposed PSO-CNN method has better fault diagnosis performance.Finally,the PSO-CNN method proposed in this paper is applied to the fault diagnosis of cyclohexanone production process in a petrochemical company,and compared with the results of the traditional KPCA method,it is concluded that the PSO-CNN method has better fault diagnosis effect.The fault detection rate of cyclohexanone production process reached 92.3%,which was much higher than that of the KPCA method which is 75.14%,and the false alarm rate was 3.247%,which could ensure the safe operation of the production process.