Key-Performance-Indicator-Related Orthogonal Trend Subspace Analysis for Nonstationary Process Monitoring

Key-performance-indicator (KPI)-related monitoring is crucial to maintain product quality and operational efficiency of industrial processes. However, nonstationary characteristics are often encountered in modern industrial systems due to fluctuations in raw materials, external disturbances, and the change in operational modes. In this context, monitoring of KPI-related variables is hard because statistical characteristics (e.g., mean and variance) of measurement data change over time. In this article, a KPI-related orthogonal trend subspace analysis (KOTSA) is proposed for monitoring nonstationary industrial processes. The idea of KOTSA is to decrease the trend variation of KPI-related variables by means of KPI-unrelated variables. Subsequently, thanks to the well-defined principal angle between two subspaces, two new statistics are proposed for nonstationary process monitoring, which leads to a significant improvement in monitoring performance by evaluating the variation of the nonstationary subspace rather than the nonstationary sample. The monitoring performance of the proposed method is theoretically analyzed and validated on three case studies, and superior monitoring performance is achieved by the proposed method.