Fault Diagnosis Method Based On The Deep Learning Research

With the development of science and technology,more and more advanced intelligent equipments are used in the chemical industry,the degree of automation is getting higher and higher,and the production process is more and more complicated.However,due to the equipment working in a long time high-intensity,sometimes there will be a variety of faults.In this paper,we use a typical chemical process as the simulation object,and propose two kinds of deep learning methods to prove the feasibility of the method to a certain extent.Compared with the traditional fault diagnosis method,the diagnostic accuracy has been relevantly improved.Firstly,this paper introduces the purpose and significance of the fault diagnosis of chemical process,and summarizes the development process of deep learning and the current application and application of deep learning.This paper introduces two classic artificial neural networks,BP neural network and probabilistic neural network,based on the development of deep learning,and presents the existing problems,which leads to the deep learning method,and lists the application of deep learning in fault diagnosis Difficulties.Then,it has been introduced a deep depth of faith model of deep learning model.the network structure and its derivation process can be introduced in detail.A fault diagnosis is proposed in which the model is based on deep belief network.The initialization method are also given,so does the training method.Some shortcomings of the fault diagnosis are found out that the model is based on deep belief network.Based on the deep belief network,this paper presents another deep learning model-the stack encoder.The network structure and derivation process of the stack encoder are introduced in detail.A fault diagnosis model based on the stack encoder is proposed,the initialization method and training method of the model are introduced.the advantages of the stack encoder are introduced.comparing with the depth belief network.Finally,in this paper,it presents the data of TE chemical process fault.The simulation process and data preprocessing are listed.According to the fault diagnosis model of simple Softmax classifier,the fault diagnosis model based on deep belief network and the fault based on stack encoder Diagnosis model of the three models respectively use the data for fault diagnosis simulation to generate the fault function to evaluate the pros and cons of the cost function curve,as well as the confusion matrix of each model.Based on the simulation results,the advantages and disadvantages of deep learning applications in fault diagnosis are analyzed,and which of the two deep learning models is more in line with TE's chemical data.