Pattern Recognition of Typical Defect Discharge Models Based on Infrared Images

Partial discharge is a common fault mode in operating electrical equipment, making its accurate detection and diagnosis crucial. Owing to its advantages such as high sensitivity, thermal imaging technology has become a significant method for early fault detection in electrical equipment. Infrared thermography is widely applied in the maintenance and fault diagnosis of substation equipment; however, research predominantly focuses on fault location detection, with relatively few studies dedicated to discharge pattern recognition. In this study, three discharge models simulating surface discharge, needle-plate discharge, and air-gap discharge were constructed in insulating oil. An infrared detection device was designed and developed based on the MLX90640 module. This device comprises an MLX90640 infrared imaging sensor for data acquisition, STM32 Microcontroller for processing, TFT LCD display for real-time visualization, and USB Type-C interface for data transmission to a computer. Infrared images were captured for three PD defect models under applied voltage 7 kV. Following over 100 repeated captures and outlier removal, 50 valid images were retained per model group. The image resolution was enhanced using a bilinear interpolation algorithm. Subsequent preprocessing included cropping, grayscale conversion, filtering, and enhancement. Feature parameters such as complexity index and length-width ratio were extracted. Pattern recognition of discharge models was accomplished through a comprehensive analysis utilizing distinct feature parameters: preliminary identification was performed based on complexity index, mean difference, and standard deviation, while length-width ratio and contrast ratio were subsequently employed to further verify accuracy. The reasons for the differences in feature parameters among the various discharge Models were analyzed.