RBF networks-based nonlinear principal component analysis for process fault detection

Recognizing the shortcomings of the traditional principal component analysis (PCA) used in fault detection of the nonlinear process, a nonlinear PCA (NLPCA) method based on radial basis function (RBF) neural networks is developed for fault detection. Firstly, a NLPCA model composing two RBF networks is proposed where the first network achieves the nonlinear transformation of the input variables to principal component and the second network performs the inverse transformation to reproducing the original data. Secondly, the principal curves algorithm is used to resolve the acquirement of the training data. Finally, the fault detection method using the RBF networks-based NLPCA is presented and then the validity of the proposed approach is illustrated by a simulation example of a 3-order nonlinear system.