Industrial Process Fault Detection And Diagnosis Based On KPCA And SSVM

It is unavoidable that the machine will break down during process of production, in this case, it not only effect qaulity of products but also cause large production accident. Traditional monitor motheds cannot slove complicated the quality control problem relying on manpower completely. The monitor method based on multiple statistic analysis doesnot rely on the precise mathematical model and it can finish fault detection and diagnosis, taking use of datas caused in course of production fully. The multiple statistic method has the advantages such as more realizable and simple. With its development over the past three decades, The monitor method based on multiple statistic analysis gets a series of great achievement and applies widespreadly in the modern industry.This paper adopts multiple statistic theory, and it is set in the Tennessee Eastman Process to carry out fault detection and diagnosis. The paper finishes the work as follow:1) The paper uses Kernel Principal Components Analysis(KPCA) method to detect and diagnose the faults in Tennessee Eastman Process, and takes the results to compare with Principal Components Analysis(PCA) method. The results show KPCA, which can extract nonlinear PCs, has better performance.2) The paper introduces Simple SVM(SSVM), meanwhile, adopts a method based on KPCA and SSVM to build fault detection identificational model. This method takes use of KPCA to extract nonlinear PCs of fault datas, as the input data of SSVM to moelling. The results show this method improves the faults identification rate.3) Because of the paramaters of SSVM effecte the model accuracy, the paper adopts Particle Swarm Optimization(PSO) method to optimizme the paramaters. Through researching on the faults further, the paper adopts a way to extract main variables to build model. The results of simulation show this method reduces the misclass rate greatly.