Data-driven Key Performance Indicator Fault Detection Approach Based on Sparse Direct Orthogonalization

In recent years, key performance indicator (KPI) detection has attracted much attention in large-scale process plants. Several methods have been developed to solve this issue. However, further studies find that post-processing methods have relatively high false alarm rates (FARs) for quality-unrelated faults. Also, methods combined with preprocessing, like orthogonal signal correction-modified partial least squares (OSC-MPLS), sometimes lack robustness. To deal with this problem, this paper proposes an enhanced pretreatment method, namely sparse direct orthogonalization (SDO), and a novel KPI-related fault detection approach called SDO-MPLS is developed. Compared with OSC-MPLS, the proposed approach has more robust performance and better interpretability, while a numerical case and the Tennessee Eastman process (TEP) are used to demonstrate the effectiveness of the proposed approach. Copyright (C) 2020 The Authors.