A Non-Contact Vibration Detection Method for Transformer Fastening Bolts Based on Video Phase Analysis

View at publisher Summarise with VIRAT
Authors: Yujie Hu, Shenglong Zhu, Cheng Zhao, Shaolei Wu, Min Pan, Zhengkai Zhang, Yuxiang Zuo, Ruonan Hu
Cited by: 0
References: 14

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Loosening of fastening bolts in power transformers can induce severe faults such as structural resonance and winding deformation, posing a serious threat to the safe and stable operation of the power grid. To address the limitations of conventional contact-based vibration detection methods—namely complex installation procedures, potential safety risks, and insufficient representation of global modal characteristics—this paper proposes a non-contact vibration detection method for transformer bolts based on amplitude-enhanced video phase motion estimation. The proposed method first performs multiscale and multidirectional time-frequency decomposition of the video sequence using a two-dimensional Gabor filter bank. Subpixel-level motion fields are then estimated based on the phase constancy principle. To suppress noise, local amplitude responses are introduced as confidence weights, enabling the reconstruction of time-domain displacement signals. Finally, vibration characteristics are extracted through frequency-domain analysis to identify the bolt condition. To validate the effectiveness of the proposed approach, an oil-immersed transformer experimental platform was established. Vibration video data of bolts in both tightened and loosened states were acquired under no-load and load operating conditions. Feature parameters, including the mean value, peak-to-peak value, fundamental frequency ratio, and the ratio of odd harmonics at 50 Hz, were extracted and analyzed. Experimental results demonstrate that, in the loosened state, the absolute mean value increases, the peak-to-peak value rises significantly, the fundamental frequency ratio decreases, and the proportion of odd harmonics at 50 Hz increases; moreover, these feature differences are more pronounced under load conditions. Across all four operating conditions, the fundamental frequency remains stable at approximately 100 Hz, confirming the reliability of the experimental data. The proposed method is easy to implement and exhibits strong robustness to interference, providing efficient technical support for predictive maintenance of transformer fastening bolts.

Fields of Study & Subjects

Machine Fault Diagnosis Techniques Power Transformer Diagnostics and Insulation Structural Health Monitoring Techniques

Publication details

Journal / source
International Conference on Electrical Automation and Artificial Intelligence (ICEAAI)
Pages
1212–1218
Publication date
16 January 2026
Language
EN