3-D Segmentation and Surface Reconstruction of Gas Insulated Switchgear via PointNet-MLS Architecture

High quality 3D reconstruction technique is essential for digital twin (DT) application of power equipment. This work presents a PointNet-MLS combined architecture to realize component segmentation and surface reconstruction of gas insulated switchgear (GIS) with complex background interference. In order to make the GIS ontology point cloud obtained continuous and smooth, greedy projection triangulation is then applied. Lastly, the local features of the GIS point cloud are enhanced, and the three-dimensional geometric properties of the GIS apparatus are better restored using the moving least squares approach. The results show that the mean intersection over union (mIoU) of PointNet++ algorithm for on-site GIS point cloud segmentation can reach 92.1%, which is higher than 32.8% and 13.7% of K-means and PointNet algorithms, respectively. The proposed MLS algorithm can effectively repair the defects of GIS point cloud after greedy projection triangulation, so that the repaired surface part can maintain the three-dimensional shape characteristics of the GIS point cloud.