Process monitoring with principal components and partial least squares

Statistical process monitoring in a multivariate setting is a common problem in many industries, including both continuous (chemical and process) and discrete manufacturing. Traditionally, techniques based on multivariate control charts have been suggested for this application. However, these procedures suffer several disadvantages, including insensitivity to process shifts when many process variables are simultaneously monitored, and inefficiency in determining which subsets of process variables are responsible for out-of-control signals. This paper presents a review and comparison of principal component analysis, partial least squares, and traditional control charting.