Hybrid framework for on‐the‐fly diagnosis of energy inefficiency in multi‐unit processes based on data‐driven and knowledge‐based integration

Abstract Pinpointing the true roots of energy waste in large, multi‐unit industrial systems is notoriously difficult: the data are high‐dimensional, and process units are tightly interlinked. This paper presents a powerful hybrid diagnostic framework that integrates explainable AI (XAI), Granger causality (GC), and fault tree analysis (FTA). By combining data‐driven variable selection using XAI, causal structure discovery through GC, and expert engineering knowledge encoded in FTA, the proposed framework enables real‐time root cause identification. It ranks the most influential variables and systematically connects them to subsystem interactions, cleanly tracing abnormal behaviour back to its origin. We validate the approach on a high‐fidelity dynamic simulator of a heat recovery network in a pulp and paper mill, showing that it can reliably detect and localize root causes in real time without relying on any labelled fault data.