Nonlinear Process Monitoring Based on Kernel PCA
AbstractIn this paper, a new fault detection and identification method based on kernel principal component analysis (KPCA) for nonlinear process is developed. KPCA can efficiently compute principal components in high-dimensional feature spaces by means of integral operators and nonlinear kernel functions. The nonlinear problems in the input spaces are translated to linear ones the feature spaces. Based on the statistics T2 and squared prediction error (SPE) charts in the feature space, the fault is detected using KPCA methods. Then a new fault identification method for KPCA is developed using the gradient of kernel function. The both new statistics CT2 and CSPE are defined respectively which represent the contribution of each variable to the monitoring statistics, Hotellings T2and SPE of KPCA. To demonstrate the performance, the proposed method is applied to Tennessee Eastman processes. The simulation results show that the proposed method effectively identifies the source of various types of faults.
cuimei Bo Jun Li Jinguo Lin Aijing Lu
College of Automation NanJing University of Technology NanJing, JiangSu 210009,P.R. China
国际会议
南宁
英文
2007-07-20（万方平台首次上网日期，不代表论文的发表时间）