Partial Derivate Contribution Plot Based on KPLS-KSER for Nonlinear Process Fault Diagnosis

In the process monitoring of nonlinear systems, kernel function is the main means to solve the nonlinear data by mapping low-dimensional nonlinear data to high-dimensional linear data. However, the use of kernel function has two disadvantages: 1. Kernel function needs a lot of calculation, especially under the condition of large number of training samples, 2. Kernel function leads to the inability to obtain the relationship between input variables and statistics. So that the identification of fault variables is difficult. In this paper, Taylor series expansion is used to remove the high order infinitesimal term, so that the Gaussian kernel function is replaced by the input matrix, which greatly reduces the amount of computation required for fault detection and diagnosis. The replacement method is introduced into the KPLS model, and the input matrix is successfully decomposed into quality-related and unrelated parts by using SVD decomposition. Based on the detection model, through the gradient theory, the partial derivate is used to calculate the gradient of each variable in the statistics to isolate the fault variables. In order to verify the effectiveness of the algorithm, this paper uses the TEP model to carry on the simulation experiment, has obtained the very good process monitoring effect, at the same time has greatly reduced the simulation experiment time.