摘要
The rapid transition toward Industry 5.0, driven by pervasive automation, dense sensor networks, and cyberphysical integration, has significantly increased the structural and dynamic complexity of modern industrial processes. Conventional process monitoring approaches often struggle to cope with such complexity, particularly when transparent fault interpretation and causal explainability are required. To address these limitations, this paper proposes a Knowledge-Informed Causal Neuro-Fuzzy (KI-CNF) framework for explainable, real-time process monitoring. The proposed framework synergistically integrates a dynamic inner global-local preserving projection (DiGLPP) for informative feature learning, an adaptive neuro-fuzzy inference system (ANFIS) for nonlinear fault detection and classification, and a causal signed directed graph (CSDG) for organizing processinformed signed directed dependencies. This unified architecture enables timely fault detection, systematic identification of influential and root-related variables, and quantitative tracing of plausible fault-propagation paths across process units in terms of influence magnitude, propagation probability, traversal time, and statistical edge-impact score. The effectiveness of the KI-CNF framework is validated against established monitoring benchmarks, including Wavelet-based Principal Component Analysis (WT-PCA) and the Dual-Attention Long Short-Term Memory Autoencoder (DALSTM-AE). Comprehensive case studies are conducted on a binary distillation column system and the Tennessee Eastman Process, representing both small-scale and large-scale industrial benchmarks. Quantitatively, KI-CNF achieved an average fault detection rate (FDR) of 99.63% with a false alarm rate (FAR) of 0.25% on the DCS benchmark and an average FDR of 99.38% with a FAR of 0.40% on the TEP benchmark, while attaining overall accuracies of 99.69% and 99.49%, respectively. Overall, the KI-CNF framework delivers a unified, explainable, and knowledge-guided approach that bridges advanced machine learning and real-world industrial process monitoring.