Research On Fault Diagnosis Based On Multivariate Statistical Analysis Process

Nowadays, due to the strong coupling of the modern industrial process system, when one part malfunctions, it will cause a chain reaction and even catastrophic accidents. So, it is very important to ensure the security and the reliability of the system, and the fault diagnosis technology provides an important way.The multivariate statistical technique method based on the signal processing is an important branch in the research of fault diagnosis.This thesis is based on the multivariate statistical theory, and studies the principal component analysis and its improvement in nonlinear factors, the reconstruction method, also an improved reconstruction method is proposed, and these methods are used in TE process and four-tank process system for fault diagnosis. The main research contents are as follows:(1) The basic theory of the principal component analysis is explained. On this basis, this thesis gives a comprehensive statistic by combing the SPE statistic and Hotelling T2statistic. And by introducing the wavelet denoising into the principal component analysis, an improved PCA method is proposed, so as to improve the fault detection rate.(2) The PCA can’t implement effective fault diagnosis for the nonlinear system, so the kernel principal component analysis method is studied. This thesis studies a KPCA fault identification method based on the contribution plots based on the partial derivative of kernel function. By contrasting the simulation results, it verifies the effecttiveness and advantage of KPCA.(3) The SPE-based fault identification method and the Hotelling T2-based fault identification method are studied. Then a relative fault isolation ability index, which is constructed by using the fault separable amplitude of two subspaces, is used to compare the fault separation ability of these two fault identification methods. Combing the two methods, a fault dentification method based on the comprehensive reconstruction is studied, and a fault identification index based on the comprehensive reconstruction is defined, to improve the accuracy of fault identification. At the same time, a singular value decomposition method is put forward to extract the fault characteristics of two subspaces.On this basis, a fault identification method based on the wavelet denoising and fault reconstruction is put forward, and two kinds of relative fault isolation ability index are defined to compare the fault isolation ability before and after the filter. By comparing the fault identification simulation results in tennessee-eastman (TE) process, it is validated to be effective and practical to introduce the wavelet denoising into the fault reconstruction. (4) The simulation model of a four-tank process system is built by MATLAB/Simulink, and the normal process data and fault data are collected, then using this thesis’s theory method to implement the simulation of fault diagnosis, so verifying the effectiveness of the method.