A comparison and evaluation of key performance indicator-based multivariate statistics process monitoring approaches

Abstract

In this paper, the key performance indicator (KPI)-based multivariate statistical process monitoring and fault diagnosis (PM-FD) methods for linear static processes are surveyed and evaluated using the multivariate statistics framework. Based on their computational characteristics, the possible methods will be broadly classified into three categories: direct, linear regression-based, and PLS-based. The three categories are respectively presented in the first part, then the comparison study in aspects of their interconnections, geometric properties, and computational costs are shown, and finally their performance for PM-FD of KPIs is evaluated using a new evaluation index called expected detection delay, where a numerical case and the Tennessee Eastman process are used to provide a demonstration of the evaluation result.