A wavelet elastic metric network for limited-data mechanical fault diagnosis of transformer on-load tap changers.

Abstract

Reliable fault diagnosis of on-load tap changers (OLTCs), a critical component for transformer voltage regulation, is essential for ensuring the safe operation of power systems. However, the nonstationary characteristics of switching vibration signals and limited fault data restrict the effectiveness of conventional diagnostic methods. To address this issue, a Wavelet Elastic Metric Network (WEMNet) is proposed. A Wavelet-enhanced Multiscale Feature Encoder is designed to extract transient impacts and multiscale time-frequency features from vibration signals, while an elastic metric mechanism adaptively adjusts feature distances according to fault-related variations, improving discriminability under limited-data conditions. Experimental results under typical OLTC mechanical fault conditions demonstrate that the proposed method achieves stable diagnostic performance in few-shot scenarios, providing an effective solution for condition monitoring of OLTC.