Chapter 4 - Linear and nonlinear multiscale latent variable-based generalized likelihood ratio for fault detection

Abstract

The objective of this chapter is to enhance the quality of fault detection by suppressing the effect of these errors using wavelet-based multiscale representation of data, which is a powerful feature extraction tool. Multiscale representation of data has been used to improve the fault detection abilities of latent variable (LV) (like PLS, PCA, kernel PCA and kernel PLS). Thus, combining the advantages of multiscale representation with those of hypothesis testing should provide even further improvements in FD. To do that, multiscale latent variables (MSLV)-based on GLRT are proposed for fault detection. The advantages of MSLV-based GLRT methods are threefold: (i) the dynamical multiscale representation is proposed to extract accurate deterministic features and decorrelate autocorrelated measurements; (ii) the MSLV model evaluates the LV of the wavelet coefficients at each scale. Due to its multiscale nature, MSLV is appropriate for modeling of data that contain contributions from events whose behavior changes over time and frequency; (iii) the GLRT is a composite hypothesis testing method and is known to have better fault detection performance compared to conventional LV statistics. The fault detection performances are illustrated through two examples, one using synthetic data and the other using simulated Tennessee Eastman Process (TEP) data. The results demonstrate the effectiveness of the multiscale-based approaches over the conventional methods.