Fault Detection And Diagnosis Method For Batch Process Based On Deep Learning

With the development of science and technology,modern industrial process is developing towards large-scale and complex direction,in which batch production process occupies a large share in the market.In order to improve the reliability and safety of complex large-scale industrial systems,reduce the occurrence of major accidents,or reduce the production downtime of industrial processes,reduce manufacturing costs,etc.,it is essential to accurately detect and diagnose industrial system failures in the early stages of production.Thanks to the upgrade of information technology and traditional measurement equipment,the acquired process measurement data is becoming more and more abundant,which makes the traditional data-driven method difficult to use for fault detection of complex intermittent processes.Because such processes usually have the characteristics of large amount of data,non-linearity,non-Gaussian and so on.In recent years,deep learning has achieved many achievements in the fields of image and natural language processing with its powerful feature extraction ability,attracting many scholars to study its theory and application,and began to try to solve some problems in their respective fields with deep learning.This paper will also introduce the deep learning method and research on the fault detection problem in the batch process.The specific research is as follows:1)Study the basic framework of the deep learning model and its algorithm,and discuss the problems and challenges faced in introducing the deep learning model into fault detection,and propose the modeling strategy and the direction that needs attention.2)For the batch production process with multi-stage operation characteristics,the convolution-auto-encoder network model based on unsupervised learning algorithm is established.Then,the Gaussian mixture model and clustering are set up on the coding layer of the network,and the computational quantity of the model is reduced greatly while the feature is extracted.Finally,the global probability detection index is presented with Markov distance and confidence interval,and the fault detection is realized.The results show that this method can effectively improve the fault detection rate by simulating experiments on a class of semiconductor etching process.3)Aiming at the large amount,large variety and low value density of process data in batch process,a supervised learning method based on long-short-term memory network and batch standardization is established.Firstly,the feature learning is improved by using the improved deep long-term and short-term memory network.By adding the batch normalization layer and the representation method of cross entropy loss,the characteristics of the intermittent process data can be effectively extracted and quickly learned.Finally,a simulation experiment was performed on a semiconductor etching process.The simulation experiment can effectively detect and diagnose various faults,and the overall detection rate of faults is over 95%.At the end of the article,the paper summarizes the paper and puts forward its own ideas and prospects for future research work.