Research On Fault Diagnosis Of Batch Process Based On Fisher Discriminant Analysis

The production process of modern process industry has complex structure,large scale and high degree of automation.Once a failure occurs,it often leads to very serious consequences.Therefore,the research on the monitoring method of industrial process becomes particularly important.The process monitoring based on the theory of statistical analysis is based on the data collected in the production process,and determines the running condition by mining the laws in the data.Based on the typical batch process of beer fermentation,this paper studies the fault diagnosis algorithm based on Fisher discriminant analysis(FDA).1.This paper proposes an improved FDA fault detection algorithm.When the distance statistics is not clear,the control limit is determined by kernel density estimation,which improves the accuracy of the model.2.Aiming at the nonlinearity of the process data,combining Kernel Theory with FDA,this paper proposes a KFDA fault detection algorithm and further exploits the advantage of independent element analysis(ICA)in data independence analysis to develop an ICA-KFDA fault detection model.The performance of the algorithm has been significantly improved in terms of missing detection and accuracy.3.For the above detected fault,through the feature extraction to establish fault database,according to the classification results of the classifier to determine the fault.Two methods of similarity discrimination and distance discrimination are adopted,in which the fault database used in the similarity discrimination is the optimal discriminant vector,which is achieved by comparing the angle between the current fault vector and the discriminant vector of the historical fault.Distance discrimination is to determine the type of fault by comparing the distance between the feature vector of the current fault data and the known fault feature vector.On this basis,a kernel method is proposed to solve the problem of separating data incompletely by FDA.An algorithm for kernel parameter optimization is presented.The simulation results verify the superiority of the kernel Fisher feature extraction method in fault diagnosis.4.By simulating the micro-beer production process data,it can be seen from the simulation results of the three algorithms that the performance of the ICA-KFDA in fault detection is better than the first two methods.However,The kernel Fisher feature extraction fault diagnosis method can effectively classify the faults and fully prove the feasibility of the algorithm.