Machine Learning Approaches for Fault Detection in Switchgear Equipment

This Power distribution networks' overall efficiency is significantly impacted by the resilience of its switch gears. In the end, larger problems arise from utility agencies dealing with a lot of unwelcome complexities brought on by defective switchgear. The variety of defects that arise in switchgear scenarios in real life makes it difficult for traditional approaches to diagnose faults in switchgear. Algorithms for ML provide a data-driven method that may adapt and learn from datasets, perhaps getting beyond the drawbacks of conventional methods. This research looks at the use of ML methods for LSTM-based switchgear equipment arcing fault detection. The methodology involved collecting sound data from various ultrasonic test instruments and transforming this data into Mel- Spectrogram images for pre-processing. The LSTM model is employed to classify the faults based on these. Performance measures including recall, accuracy, precision, F1-score, and cross-entropy loss are utilized to assess a results of an Adam optimizer-optimized LSTM model. According to the results, the LSTM model had a defect detection accuracy of 95.8%, surpassing other ML models such as RF and Extreme Learning Machine (ELM). This method improves the performance and reliability of switchgear systems by showing that LSTM networks can forecast arcing issues. Therefore, it can be inferred that the suggested method can detect arcing problems in switchgear.