Cross-scale information fusion for fault detection and hierarchical localization of distributed parameter systems

Accurate fault diagnosis in distributed parameter systems (DPSs) is fundamentally challenged by the inherent difficulty in quantifying and decoupling spatiotemporally coupled anomalies from pervasive system noises via distributed measurements. In this research, a novel cross-scale information fusion framework is proposed for rapid fault detection and hierarchical localization of DPSs. First, the inter-module similarity is designed to monitor voltage abnormalities using the recursive correlation computation with the square wave perturbation. The dominant and residual statistics of the distributed temperature are extracted by the proposed spatiotemporal feature decomposition. A weighted fusion strategy based on the genetic algorithm optimization is proposed to construct a multiscale indicator by combining the inter-module similarity, dominant statistic, and residual statistic. The hierarchical localization framework first identifies fault-affected modules through the inter-module similarity, then localizes the specific cells using extracted spatiotemporal features. From a measurement science perspective, this framework advances multi-sensor data fusion methodology by integrating heterogeneous measurands with distinct measurement uncertainties and spatial resolutions. The proposed approach enhances the overall measurement sensitivity and spatial measurement resolution for health monitoring of complex distributed systems. Experimental verification implemented on a Li-ion battery pack under different working conditions demonstrates the effectiveness and superiority of the proposed method with high detection rate, small detection delay, and high localization accuracy. © 2026 Elsevier Ltd.