Fault Identification Method Of Chemical Process Based On Spatial-temporal Information Fusion

The continuous expansion of modern industrial production scale has put forward higher requirements for safety and reliability.The chemical industry has the characteristics of being flammable and explosive,and once an accident occurs,it will cause huge property losses and even casualties.Therefore,using real-time monitoring and fault diagnosis technology in the chemical process has become a necessary means to ensure production safety.The traditional method of fault diagnosis based on mechanism modeling and knowledge modeling has many limitations,and due to the large-scale and complex production scale,it is increasingly difficult to obtain the mechanism and knowledge models of the system.With the rapid development of data storage technology and artificial intelligence technology,a large amount of historical data of chemical processes can be saved,and data-driven fault diagnosis technology has ushered in new opportunities.Among them,deep learning has attracted much attention and made significant research progress in the field of fault diagnosis due to its powerful feature extraction ability.In this context,this thesis conducts research on the fault identification method of chemical processes based on spatiotemporal information fusion.The main research work and achievements are as follows:(1)Under the convolutional neural network(CNN)deep learning framework,different input representations have varying degrees of impact on fault diagnosis performance.The algorithm discusses the influence of three categories of seven different representations: numerical representation at different scales,image mapping representation(radar chart mapping,Gramian angular summation fields(GASF)mapping)and signal transformation representation(fast fourier transform(FFT),wavelet transform).Testing on the TE process dataset shows that the FFT method has the best average performance.On this basis,a universal fault detection and diagnosis(FDD)ensemble learning model is proposed by integrating multiple base learners.Finally,the effectiveness of the ensemble learning model in fault detection and diagnosis was verified on Tennessee Eastman process data,and the framework provides an effective method for fault detection and diagnosis based CNN.(2)Industrial process data is mostly high-dimensional temporal data,and the types of faults are closely related to the temporal and spatial distribution characteristics of the data.Therefore,it is particularly crucial to reasonably characterize and effectively mine the deep spatiotemporal information contained in the data.To solve this problem,the graph convolution network(GCN)is introduced to extract the spatial distribution information in the data,and the long short memory network(LSTM)and the onedimensional convolutional neural network(1D CNN)are introduced to extract the temporal information in the data.The experimental results show that the proposed algorithm effectively extracts and utilizes the spatiotemporal information of chemical process fault data,greatly improving the accuracy of fault recognition.(3)To further improve the usability of the algorithm,a graphical interactive interface for chemical process fault identification was designed and developed based on Py Qt5.This interface includes five major functions: dataset import,parameter setting,display of running process,visualization of network layer features,and display of experimental results.The interactive interface framework is clear,the content is complete,the use is simple,and the operation is convenient.At the same time,the two major functions of process display and network layer feature visualization can provide ideas for model optimization and give users a good experience.