IMPROVED MSPCA WITH APPLICATIONTO PROCESS MONITORING

摘要

This paper proposes a improved Multi-scale PrincipalComponent Analysis (MSPCA). A key problem of faultdetection in process monitoring lies in how to enhance theaccuracy of fault detection to reduce detection costs. In thispoint, MSPCA has improved to a great degree, but it can beimproved in the accuracy of selecting the coefficients ofwavelet used in reconstructing. Because the accuracy ofselecting the coefficients of wavelet is directly concernedwith the accuracy of fault detection in using PrincipalComponent Analysis (PCA) reconstructed. When selectingthe coefficients of wavelet based on selecting the coefficientsof wavelet in using MSPCA, The improved MSPCA shouldbe proposed to detect the fault in process monitoring. Itutilizes Principal-component-related Variable Residuals (PVR)statistic and Common Variable Residuals (CVR) statistic atdifferent scales to replace the statistic Q and combine themwith the statistic T2 to select the coefficients of wavelet.According to the analysis of simulation of algorithm'sexample, and comparing the improved MSPCA withMSPCA and conventional PCA, it shows that the improvedMSPCA has enhanced the accuracy of fault detection inprocess monitoring.