Entropy principal component analysis and its application to nonlinear chemical process fault diagnosis

Most chemical processes generally exhibit the characteristics of nonlinear variation. In this paper, an improved principal component analysis method using information entropy theory, called entropy principal component analysis (EPCA), is proposed for nonlinear chemical process fault diagnosis. This approach applies information entropy theory to build an explicit nonlinear transformation, which provides a convenient way for nonlinear extension of principal component analysis. The information entropy can capture the Gaussian and non-Gaussian information in the measured variables via probability density function estimation. With the entropy of original measured variables, the entropy principal components are calculated using a simple eigenvalue decomposition procedure, and two monitoring statistics are built for fault detection. Once a fault is detected, EPCA similarity factors between the occurred fault dataset and historical fault pattern datasets are computed for fault recognition. Simulations on a continuous stirred tank reactor system show that EPCA performs well in terms of fault detection and recognition. (C) 2014 Curtin University of Technology and John Wiley & Sons, Ltd.