Motion-to-Risk: Physics-Guided Multi-Source State Assessment for High-Voltage Vacuum Circuit Breakers

High-voltage vacuum circuit breakers are critical switching devices in power systems, and their reliable condition assessment is essential for safe operation and maintenance decision-making. However, breaker abnormalities are often reflected by heterogeneous operation-related evidence, and existing methods based on single-source measurements or generic feature fusion may weaken source-specific diagnostic roles and limit the recognition of compound abnormal conditions. To address this problem, this paper proposes the Physics-Guided Multi-Source State Assessment Network (PMSA-Net), a reliability-aware framework that integrates mechanical motion, opening- and closing-position limit events, and infrared thermography. Source-specific encoders first extract dynamic, end-position, and thermal representations. Reliability-aware Asymmetric Selective Interaction (RASI) then calibrates primary and auxiliary evidence, using mechanical motion as the operational context and the other sources as complementary constraints. Thermal Frequency-aware Selective Modulation (TFSM) stabilizes the low-frequency thermal field and enhances high-frequency hotspot responses. Task-conditioned evidence allocation jointly predicts state category, travel anomaly, limit-event consistency, thermal risk, and overall risk. On a laboratory-simulated benchmark covering 15 operating conditions, PMSA-Net achieved 90.18±0.36% state-category accuracy and an 80.00±0.50% overall-risk macro-F1 score over five independent runs. Under cross-source abnormalities, it exceeded the variational-fusion baseline by 2.59 and 3.05 percentage points on these metrics, respectively. These results indicate that reliability-aware calibration improves multi-task assessment of compound breaker abnormalities.