Research On Fault Diagnosis Of Industrial Process Based On Data Driven

In recent years,with the swift development of science and technology,especially computer and information management technology,the scale of industrial production equipment is becoming more intelligent,large-scale,continuous,automatic and high-speed.An increasing number of variables can be measured,handled and monitored,and conventional fault diagnosis methods such as mathematical model and process knowledge cannot meet the demand.Datadriven approaches rely on historical data obtained from industrial production to build a model,rather than depending on a precise mathematical model,which has gradually been widely concerned.This paper takes Tennessee Eastman Process(TEP)as research background,based on principal component analysis(PCA)and kernel independent component analysis(KICA)to study fault detection and diagnosis of industrial production process.This thesis elaborates on the research status of fault diagnosis technology and industrial fault diagnosis technology,and the basic principle of PCA,Fast ICA algorithm,KICA and decomposition of wavelet packet.That how to establish monitoring statistics,set the control limits of monitoring norms and utilize wavelet package to filter is introduced.PCA and KICA algorithm are simulated and analyzed in the article,on the basis of analyzing the advantages and disadvantages of two algorithms respectively,fault detection method of kernel independent component analysis and principal component analysis(KICA-PCA)is introduced to combine KICA and PCA.Non-Gaussian characteristic information and Gaussian feature information of the data are extracted by two-step method in the paper.According to the results of TEP simulation,it is found that KICA-PCA algorithm achieves synchronous monitoring of non-Gaussian and Gaussian processes.But on account of the neglect of dynamic characteristics of autocorrelation and hysteresis in process sampling data,the arithmetic detects small fault and gradual failure poorly,and lacks available variable contribution analysis method to diagnose faults.In this treatise,in order to put forward a fault diagnosis method based on wavelet packets filtering dynamic kernel independent component analysis and principal component analysis(FDKICA-PCA),theory of wavelet packets filtering and Characteristics of AR model predicting data are incorporated into KICA-PCA algorithm.In this method,for the sake of eliminating the influence of instrument precision and measurement error on the diagnosis process,wavelet packet is used to filter the process data.What's more,AR model is utilized to predict the process data and make data dynamic.Finally,data is analyzed by the two-step KICA-PCA method,and non-linear diagram analysis is used for fault diagnosis.FDKICA-PCA algorithm and KICA-PCA algorithm are used for fault detection of TE process separately.Simulation results show that detection rate of FDKICA-PCA is higher,and the algorithm can extract dynamic and nonlinear of the data and filter the data effectively.