"As part of the digitalization of power grids, we process hundreds of point clouds to create digital twins of high-voltage switchgear. The degree of automation of object recognition depends heavily on the quality of the point clouds. Together with the team at Fraunhofer IPA, we have successfully evaluated ML methods that we can use to carry out automated quality checks on the input data."
Wolfgang Eyrich
entegra eyrich + appel gmbh
Contact at the AI Innovation Center
Ira Effenberger
AI-supported evaluation of 3D point clouds of high-voltage switchgear
Initial situation
Digital twins of high-voltage switchgear are being created for the digitalization of electricity grids. The digital models are based on 3D point clouds of the systems, which are recorded using laser scanners. However, the degree of automation in the creation of the digital twin depends largely on the quality of the point cloud. The weather conditions, the devices and software products used, as well as the procedure during scanning can influence the quality of the resulting 3D point cloud. For efficient digitization of high-voltage switchgear, the quality of the point clouds is therefore checked first. This previously manual inspection and evaluation of the point cloud quality is very time-consuming and also not objective. This process is therefore to be automated and an AI-based assessment of the point cloud quality is to be investigated as part of the quick check.
Solution idea
In the scans of the high-voltage switchgear, statements about the quality of the scan data are to be generated automatically. The evaluation is to be carried out on the basis of the lines. This approach has several advantages. Lines have a high availability in the data. Furthermore, the geometry of the cables is known, as they can be modeled approximately as cylinders. Therefore, the pipes in the 3D point clouds are first segmented using AI. The quality of the point clouds is evaluated by analyzing the segmented pipes. The point cloud of the system is analyzed with AI at several hundred points so that a clear statistical statement about the quality of the point cloud can be made.
Benefit
Models of the existing switchgear are necessary for the digitalization of the electricity grid. Due to the large number of switchgears, a high degree of automation is required for digitization. In this Quick Check, the first step of the processing pipeline - evaluating the scan quality - was considered. By automatically evaluating the quality of the point cloud, the manual, monotonous and time-consuming inspection of the scans can be replaced. Since the AI-based assessment of scan quality is carried out on hundreds of locations in the point cloud, it is not a random check, but provides a complete and detailed picture. In addition, the assessment is reproducible and does not depend on experience or the subjective assessment of a manual inspection.
Implementation of the AI application
Already digitized switchgear serves as the basis for training the AI algorithm. Training data is automatically generated from these systems by feeding the information from the already generated digital twins back into the initial clouds. The developed network derives local geometric features from the point cloud. These local features are combined with each other until features with high semantic meaning are created, on the basis of which the lines can be segmented from the point cloud. For a point-by-point segmentation of the 3D point cloud, the global features are combined with the local features.