Research on Substation Thermal Defect Identification Method based on Infrared Point Cloud

Substation is the basic unit of power system production and operation, but the long-term operation of equipment is prone to thermal fault, so thermal defect detection is an important means to ensure power safety. Current detection methods rely on two-dimensional infrared images, which can only obtain the temperature distribution information of the equipment, and lack the spatial depth information, so it is difficult to realize the accurate identification of thermal defects. In order to improve the detection accuracy, this paper proposes an infrared point cloud detection method based on deep learning. Through the joint calibration of multi-sensor and the registration and fusion of point cloud, a three-dimensional infrared point cloud model integrating spatial geometry and temperature information is constructed. The typical equipment is selected to establish the infrared point cloud data set, and then the PointNet++ network is used to extract the characteristics of the three-dimensional infrared point cloud data and segment the thermal defects, so as to realize the automatic identification of equipment overheating defects. For the problem that the measurement temperature is affected by the angle of view and distance, the feedforward neural network is used to compensate and correct the temperature data. The experimental results show that the PointNet++ method for detecting thermal defects has achieved 98.86% overall accuracy (OA) and 97.50% average intersection union (mIoU) on the test set. The results show that the detection method based on two-dimensional and three-dimensional data fusion and deep learning can provide reliable technical support for thermal defect detection of substation equipment, and has important practical application value.