Abstract Plant‐wide industrial processes are increasingly prevalent in modern industry. To address safety challenges in such large‐scale systems, numerous distributed monitoring schemes have been developed. However, most existing methods rely solely on data‐driven, block‐wise monitoring, focusing primarily on statistical correlations among variables while neglecting their physical interconnections. This paper proposes a multi‐block monitoring method grounded in a knowledge‐based causality diagram, offering a distinct alternative to conventional data‐driven approaches for large‐scale process monitoring. By integrating domain knowledge into data processing, the method improves interpretability and enhances fault detection and diagnosis capabilities. First, process knowledge is used to construct a knowledge‐based causality diagram, which decomposes the system into physically meaningful sub‐blocks, thereby increasing interpretability. Second, each sub‐block is monitored locally, followed by global monitoring, to improve fault detection performance. Third, a hybrid fault diagnosis method that combines knowledge and data is introduced. By analyzing variable contributions across sub‐blocks and leveraging the knowledge‐based causality diagram, the root cause and fault propagation path can be effectively identified. Finally, the proposed method is evaluated using the Tennessee Eastman benchmark process.
Zhou et al. (Fri,) studied this question.