Process monitoring based on wavelet packet principal component analysis

Abstract

For improving performance of principal component analysis (PCA) for process monitoring with noise, this paper proposes a wavelet packet PCA (WPPCA). It integrates ability of wavelet packet in de-noise and ability of PCA to de-correlate the variables by extracting a linear relationship. Process monitoring is simulated in a two-inputs, two-outputs dynamic process with noise. The simulation result shows that the wavelet packet PCA eliminate effects of noise effectively with better monitoring performance. Finally, the proposed approach is successfully applied to Tennessee Eastman process for dynamic monitoring.