A Multidimensional Scaling based Process Monitoring Technique

摘要

This paper presents a new multivariate process monitoring method based on thernclassical scaling (CMDS) technique. CMDS technique projects the multivariate scoresrnby using the inter-dissimilarity values of the multivariate process measurements orrnvariables and, hence, complex computation of nonlinear principal components can bernavoided. Modified T2 and squared prediction errors are introduced as the monitoringrnstatistics and are developed based on the multivariate scores of variable configurations.rnA moving window mechanism is proposed in evaluating the variation in therndissimilarities among process variables. The proposed method is applied to the on-linernmonitoring of a simulated continuous stirred tank reactor with recycle. The applicationrnresults demonstrate that the proposed method is very effective in detecting variousrnabrupt and incipient faults.