Nonlinear Batch Process Monitoring Using Phase-Based Kernel-Independent Component Analysis-Principal Component Analysis (KICA-PCA)

In this article, the statistical modeling and online monitoring of nonlinear batch processes are addressed oil the basis of the kernel technique. First, the article analyzes the conventional multiway kernel algorithms, which were just simple and conservative kernel extensions of the original multiway linear methods and thus inherited their drawbacks. Then, ail improved nonlinear batch monitoring method is developed, This method captures the changes of the underlying nonlinear characteristics and accordingly divides the whole batch duration into different phases. Then, focusing on each subphase, both nonlinear Gaussian and non-Gaussian features are explored by a two-step modeling strategy usingkernel-independent component analysis-principal component analysis (KICA-PCA). Process monitoring and fault detection can be readily carried out online without requiring the estimation of future process data. Meanwhile. the dynamics of the data are preserved by exploring time-varying covariance structures. The idea and performance of the proposed method are illustrated using a real three-tank process and a benchmark simulation of fed-batch penicillin fermentation production.