The Fault Diagnosis Method of On-Load Tap Changers Based on Phase Space Tensor Decomposition and Deep Learning Models

During the switching process of on-load tap changers (OLTCs), the vibration signals generated by the motion of mechanical components are closely related to their mechanical states. This study utilizes vibration signals from multiple locations, representing them using phase space tensors. The constructed third-order phase space tensor is then decomposed using High-Order Singular Value Decomposition (HOSVD) to extract the core tensor. Based on this, a CNN-Transformer-based mechanical fault diagnosis model for OLTCs is proposed. The results show that the core tensor information from the vibration signals during OLTC operation effectively captures signal features. The proposed CNN-Transformer model demonstrates excellent performance in mechanical fault diagnosis, achieving an accuracy of up to 95%. This model provides an important reference for fault identification and condition-based maintenance of OLTCs.