Fault diagnosis method based on TKPCR and its application in superheated steam system

The issue of multivariate statistical process monitoring is a popular research topic, within which fault diagnosis method faced to nonlinear systems is an important difficulties. This paper proposed a partial derivative contribution plot based on total kernel principal component regression (TKPCR) method which is a solution of the nonlinear process fault diagnosis problem. The TKPCR model is constructed to build the potential correlation between the input and the output matrix. Then, the input matrix X is decomposed into two orthogonal subspace, and two monitoring statistics T
_o
and T
_y
are designed for fault detection. Then, we propose a comprehensive index C for fault diagnosis, which is composed of T
_o
and T
_y
. Furthermore, we applied the partial derivate of kernel function to acquire the contribution of variables. The threshold values of each variable's contribution are given to obtain better fault diagnosis results. Finally, the model of the boiler steam superheat system is used to confirm the advantages of this method. In the simulation, the proposed method can obtain the faulty variables accurately for the nonlinear systems.