Deep Learning Based Transformer Anomaly Detection

The detection of transformer abnormalities is a critical aspect of substation safety and reliability. Traditionally, this task relies on human judgment, where duty officers assess transformer conditions based on audible cues during routine patrols. However, this method are labor-intensive and fail to provide real-time anomaly detection, leaving substations vulnerable during monitored periods. Additionally, inspection robots currently equipped with cameras lack audio processing capabilities, further limiting their effectiveness in identifying transformer anomalies. To address these limitations, this paper proposes a deep learning -based algorithm for real-time transformer anomaly detection. By deploying a voice processing device near transformers, audio signals are continuously collected and transmitted to a server for analysis. On the server side, a deep learning model leverages speech processing techniques to detect abnormalities with high accuracy in real time. The proposed method offers significant advantages over traditional approaches. Unlike manual inspections, it eliminates dependency on the presence and expertise of duty officers, achieving higher real-time performance. Moreover, the system’s classification accuracy of 97% substantially reduces the reliance on human labor and the associated costs. This study demonstrates the potential of integrating deep learning with voice signal processing to enhance the efficiency and reliability of transformer monitoring systems.