Publication
In the Industrial Internet of Things (IIoT), condition monitoring sensor signals from complex systems often exhibit nonlinear and stochastic spatial-temporal dynamics under varying conditions. These complex dynamics make fault detection (FD) particularly challenging. While previous methods effectively model these dynamics, they often neglect the evolution of relationships between sensor signals. Undetected shifts in these relationships can lead to significant system failures. Furthermore, these methods frequently misidentify novel operating conditions (OCs) as faults. Addressing these limitations, we propose dynamic edge via graph attention (DyEdgeGAT), a novel approach for early-stage FD in IIoT systems. DyEdgeGAT's primary innovation lies in a novel graph inference scheme for multivariate time series that tracks the evolution of relationships between time series, enabled by dynamic edge construction. Another key innovation of DyEdgeGAT is its ability to incorporate OC contexts into node dynamics modeling, enhancing its accuracy and robustness. We rigorously evaluated DyEdgeGAT using both a synthetic data set, simulating varying levels of fault severity, and a real-world industrial-scale multiphase flow facility benchmark with diverse fault types under varying OCs and detection complexities. The results show that DyEdgeGAT significantly outperforms other baseline methods in FD, particularly in the early stages with low severity, and exhibits robust performance under novel OCs. The code and data associated with our research are publicly accessible via https://github.com/EPFL-IMOS/dyedgegat.