Compensation terms to improve fault detection in multivariate auto-correlated processes

This paper analyses the use of ARMA filters for detecting abnormal conditions in complex processes. Such filters were recently introduced in the multivariate statistical process control (MSPC) framework to address the issue of auto-correlation in the recorded variables [1]. While these filters can indeed remove auto-correlation from the associated MSPC monitoring scheme, this paper shows that their application influences the sensitivity for fault detection. A compensation term is introduced here to correctly identify the magnitude of abnormal conditions.