Research On Process Monitoring Data Generation And Fault Diagnosis Based On Deep Learning

Modern industry is becoming more and more automated,intelligent and complex with the rapid development of modernization and computer science,and industrial data has become the core of modern industrial informatization.However,the substantial increase in the amount of industrial data has brought great challenges to existing industrial data processing methods.There are significant limitations to using traditional methods to process existing industrial data.In recent years,deep learning become one of the best methods for big data processing and analysis.It is good at discovering intricate relationships in multidimensional data.Although the appllication of deep learning has been relatively perfect in the field of images,more exploration and research are needed in the processing of industrial time series data.In modern industry,because most of the time is normal working conditions,there are fewer faulty working conditions,so the number of normal samples collected.Therefore,the number of normal samples collected is much larger than the number of faulty samples.When performing fault diagnosis on unbalanced datasets,the traditional diagnostic methods have a low classi fication accuracy.Therefore,this paper proposes a research framework for process monitoring data generation and fault diagnosis based on deep learning model Generative Adversarial Networks(GAN)and Convolutional Neural Networks(CNN)for the imbalance of categories in industrial data.The main contents and contributions of this paper mainly include the following three points:First,this paper proposes a data reconstruction method based on matrix construction,to be able to represent and model the time series data,time series data by deep learning.This method constructs the time series data into a matrix form and then converts it into a picture form.The obtained data in the form of picture retains the time-dependent characteristics of the time series data to the greatest extent.Then,this paper uses a variety of improved GANs to solve the problem of fault sample generation for the category imbalanced dataset.The newly generated data has very similar characteristics to the original data,which shows that GAN has a strong capability of time series data generation.Finally,for the problem of difficult diagnosis of unbalanced datasets in the process industry,the fault data generated by the GANs are added to the original unbalanced dataset firstly,and enhanced the original dataset into category-balanced,and then CNN is used for fault diagnosis on this balanced dataset.The research framework of data generation and fault diagnosis proposed in this paper is verified in the actual polyethylene gas-solid fluidized bed equipment.The experimental results show that the use of the enhanced dataset generated by GANs combined with the CNN fault diagnosis method can greatly improve the accuracy of fault classification.