Fault detection of uncertain nonlinear process using reduced interval kernel principal component analysis (RIKPCA)

The interval kernel principal component analysis (IKPCA) is an extension of the KPCA method to deal with data with uncertainties. However, for a large data set the IKPCA method suffers high computation complexity. To avoid these disadvantages, a new fault detection method for uncertain nonlinear process entitled reduced interval kernel principal component analysis is proposed in this paper. The concept of the developed method consists of determine a reduced data set by choosing variables with the variance of highest projection in the direction of the selected principal components. Two RIKPCA models are developed: the first model is based on the midpoints-radii KPCA (RIKPCA(CR)) and the second one is based on the lower and upper bounds of intervals (RIKPCA(UL)). The purpose of the developed RIKPCA technique is to improve the efficiency of the IKPCA technique and to minimize the computation time. The efficiency of the developed method is illustrated by an application to the Tennessee Eastman process (TEP), and the desired performance is satisfied.