Comparison Study of Multivariate Statistics Based Key Performance Indicator Monitoring Approaches

摘要

In this paper, multivariate statistical process monitoring approaches for key performance indicator (KPI) related static processes are reviewed under a unified framework. Based on their key nature in extracting KPI-related information from process variable space for performance monitoring, those approaches are analyzed and sorted into three categories: direct cross-correlation based decomposition method, modified least square regression based approaches, partial least square based approaches. In addition, their numerical properties and monitoring performance are compared in details. Finally the well-accepted TE benchmark process is utilized to demonstrate the theoretical comparison results and their monitoring performance from industrial viewpoint.