Next-Generation Defect Detection in High Voltage Electrical Equipment via Deep Learning and Image Processing

The paper presents a deep-learning-based, thermal image-analysis advanced defect detection method for high-voltage electrical equipment. Based on the features extracted from the variant AlexNet of the convolutional neural network (CNN), the classification model designed with the classifier Random Forest (RF) achieved 94.8% accuracy. The photographs used for investigation were obtained on an infrared thermal camera at several substations in Chongqing, China. They were taken under cold weather conditions. The electrical components' thermal conditions were classified into two defective and non-defective types based on the temperature differential. The proposed approach supersedes other methods in terms of precision: the precision values are high at 93.2% and the recall values at 95.6%. The combination of CNN and RF forms a computationally efficient solution for achieving the enhancement of defect detection reliability in high-voltage systems. The results outline the potential of this technique in enhancing maintenance practices, minimizing equipment failure probabilities, and ensuring safe electrical infrastructure usage. The scope of future work will be on optimization techniques along with their application within some different environments.