A Multivariable Statistical Process Monitoring Method Based on Multiscale Analysis and Principal Curves

This study aims to develop an algorithm by integrating multi-resolution analysis (MRA) and principal curves (PC) for monitoring multivariate processes. This may pave the way for handling nonlinear data by means of principal curves in process monitoring area. We succeed in utilizing PC technique for monitoring without the assistance of neural networks, a traditional tool to deal with nonlinear model in papers, and get ideal results. The methodology proposed is tested with a mathematical example and a simulated benchmark process: the continuous stirred tank reactor (CSTR). The results demonstrate that, compared with traditional principal component analysis (PCA), PC, nonlinear PCA and multiscale PCA, the proposed approach can extract the nonlinearity and decorrelate the autocorrelated measurements effectively and is, hence, suitable for multivariate process monitoring.