Infrared Image-Based Fault Diagnosis and Condition Assessment of Power Equipment Using Deep Learning

With the advancement of smart grid development, the safe operation of power equipment is critical to ensuring reliable energy supply.Infrared imaging technology serves as an effective tool for fault detection in power systems; however, traditional image processing methods struggle to extract robust features under complex environmental conditions, limiting their effectiveness in achieving high-precision fault diagnosis.The emergence of deep learning has opened new avenues in this field.Nonetheless, existing studies based on YOLO-series algorithms face challenges such as low accuracy in detecting small-scale faults and limited capacity for analyzing multiple fault types.Additionally, current approaches to condition assessment often fail to integrate thermal distribution from infrared images with operational parameters, resulting in incomplete evaluations.To address these issues, this study proposes two major contributions: (1) an improved YOLOv5 algorithm enhanced with attention mechanisms for optimized feature fusion and an adaptive anchor box strategy to boost the detection accuracy of small-scale faults, such as bushing overheating and connector oxidation; and (2) a condition assessment framework that integrates features from infrared images, operating parameters, and historical fault data.A comprehensive indicator system is developed based on multi-source information fusion, and the Analytic Hierarchy Process (AHP) combined with a fuzzy comprehensive evaluation model is employed to quantify the health status of equipment.The proposed method significantly enhances the accuracy and robustness of fault identification and establishes an integrated "fault diagnosis-condition assessment" framework.These results offer a complete solution for intelligent maintenance of power equipment and provide valuable theoretical and practical insights for applying deep learning in the field of power system monitoring.