Nonlinear Process Monitoring Using Dynamic Sparse Kernel Classifier

Abstract Nonlinear process monitoring method based on kernel function is effective but has great computation complexity for all training samples are introduced in model training. This paper proposes a novel sparse kernel method based on dynamic sparse kernel classifier (DSKC) for nonlinear dynamic process monitoring. In the proposed method, monitoring model is built using a nonlinear classifier technique based on kernel trick. In order to reduce the complexity of kernel model, a forward orthogonal selection procedure is applied to minimize the leave one out error. A monitoring statistic is developed and confidence limit is computed by kernel density estimation. For identify fault source variables, contribution plot is constructed based on the idea of sensitivity analysis. Simulation of a continuous stirred tank reactor system shows that the proposed method performs better compared with kernel principal component analysis in terms of fault detection performance and computation efficiency.