Broad Learning System Based Visual Fault Diagnosis for Electrical Equipment Thermography Images

The basic principle of fault diagnosis using infrared thermography is that the temperature of electrical equipment reveals its service condition. However, the crucial information is often masked by complex backgrounds. Besides, the thermal image sequence of electrical equipment in power systems usually reaches tens of thousands of frames, which also increases the difficulty of diagnosis. Moreover, previous work mostly focuses on temperature information of images at pixel-wise level. Here, the spatial-temporal properties of infrared image sequence is recognized for fault diagnosis. An automatic electrical equipment diagnosis strategy for preventive maintenance is proposed in this work, in which spatial-temporal properties of infrared image sequence including spatial thermography distribution and temporal temperature features have been considered. Temperature soft-sensing and target segmentation are first accomplished, which is the basis of subsequent comprehensive utilization of temperature information and image texture. Then features describes time-variation and spatial distribution are extracted from infrared image sequence. Here, the utilization of infrared image information is no longer restrained to pixel level or single detection time point. Finally, classification model is developed with broad learning system to determine whether equipment is under a normal operating state. The efficacy of the proposed method is verified through a power system application.