Industrial monitoring based on moving average PCA and neural network

Abstract

For industrial process monitoring, instead of conventional principal component analysis (PCA), dynamic principal component analysis (DPCA) is a powerful tool to deal with dynamic relationship among process variable series. Similar to PCA, conventional DPCA methods are not suitable for many industry processes with variables containing nonlinear relationship. Neural networks possess the ability to approximate any nonlinear functions, it have been used is to capture the nonlinear function in DPCA. However, it often needs too many nodes in its layers. It is known that moving average techniques can capture time-dependent without adding the dimensionality of raw data set. Along the line, a method of industrial monitoring discussed in this paper combines moving average PCA (MAPCA) and neural network. As an application case, industrial chemical separation process is used to illustrate the method.