Fault diagnosis and early warning for distribution network equipment operation and maintenance using artificial intelligence and digital twins

Abstract

A collaborative framework integrating artificial intelligence and digital twin modeling is developed for high-precision fault diagnosis and early warning in distribution network equipment operation and maintenance. The proposed system features a hierarchical edge-cloud data architecture using MQTT protocols to enable low-latency, multi-modal data sharing among physical assets, virtual digital replicas, and AI models. Hybrid training techniques combine real-world operational records with physics-based digital twin simulations, addressing data scarcity and enhancing the detection of rare and compound fault types. An extended Kalman filter synchronizes virtual and physical system states at millisecond resolution, while closed-loop feedback dynamically calibrates both AI inference and twin models. Experimental results using a semiphysical platform with twelve transformers and ten fault categories demonstrate that the composite diagnostic metric of the collaborative approach exceeds conventional AI and expert system baselines in both accuracy and response speed. The framework markedly reduces false alarms and generalization error when exposed to evolving grid scenarios, proving robust under data imbalance and operational uncertainty. This work substantively advances digital-integrated predictive maintenance and delivers an engineering blueprint for intelligent operation of modern distribution networks.