Process monitoring based on measurement space decomposition

J.L. GODOY; J.R. VEGA; J.L. MARCHETTI

Roma

Conferencia; IMAACA 2011 - 5th International Conference on Integrated Modeling and Analysis in Applied Control and Automation (part of the 8th International Mediterranean and Latin American Modelling Multiconference); 2011

A useful decomposition of the input and output variable spaces is obtained by using the relationships supporting Partial Least Squares Regression (PLSR). The resulting technique is capable to classify faults or anomalies according to four types: those associated to measurements of input variables, those related to measurements of output variables, those linked to the inner latent structure of the complex processes data and those related to upsets that follow its correlation structure. This classification is suggested by the different subspaces in which the whole measurement space can be decomposed by using PLSR modeling and specific statistics actuating on each subspace. Hence, using an in-control PLSR model, the tool is able to detect anomalies and to diagnose its kind. The approach can be used for monitoring closed loop systems and to detect abnormal controller functioning. Several features of the proposed classification technique are analyzed through static and dynamic simulation examples.