Abstract: Vacuum on-load tap changers (OLTC) are key components of electrical systems, relying on vacuum arc extinguishing technology to ensure stable voltage regulation. However, under high load and high current conditions in large hospitals, vacuum degradation, contact ablation, and insulation contamination pose significant challenges to operational reliability. This study proposes an intelligent multi-modal diagnostic method for vacuum OLTCs by integrating vacuum-specific vibration characteristics and electrical parameter analysis. First, based on the mechanical dynamics of vacuum arc suppression, a Hankel matrix is constructed from vibration signals, and Singular Value Decomposition (SVD) with Fast Fourier Transform (FFT) filtering is applied to eliminate noise interference caused by vacuum chamber micro-vibrations. Second, from the perspective of vacuum arc energy balance, key electrical parameters including transient current harmonics and dielectric recovery voltage are extracted as fault indicators. Finally, a multi-class Support Vector Machine (SVM) optimized by a binary tree structure is employed to fuse vibration and electrical features, achieving accurate identification of six typical vacuum-related faults (e.g., vacuum leakage, contact oxidation). Experimental results demonstrate 98.2% diagnostic accuracy, with particular advantages in early-stage vacuum pressure decline detection. This method provides a novel solution for real-time monitoring of vacuum integrity in OLTCs, significantly enhancing the safety of vacuum-based power regulation systems.

Key words: vacuum arc extinguishing, on-load tap changer, singular value decomposition, dielectric recovery strength, multimodal fault diagnosis