MONITORING

Time-Varying Processes

Most large-scale processes present non-stationary behaviour at different time scales. Several approaches have been developed to handle this feature. Among them, are included the multivariate statistical process control approaches (MSPC) based on recursive principal component analysis (RPCA) and recursive partial least squares (RPLS) using several updating formulas such as rank-one updating or block updating through Lanzos tridiagonalization (Dayal & MacGregor, 1997; Li et al., 2000; Wold, 1994), moving window PCA and PLS (MWPCA, MWPLS), multiscale decomposition and recombination of relevant scales (Rosen & Lennox, 2001; Teppola & Minkkinen, 2001), as well as the use of explicit modelling frameworks for non-stationary processes, such as the ARIMAX model family. We have conducted an extensive comparison study and developed new approaches for handling this difficult but ubiquitous problem in industry.