Classifying elements in an infrastructure model using convolutional graph neural networks

Inventors

LAPOINTE, Marc-André • Asselin, Louis-Philippe • Jahjah, Karl-Alexandre • Rausch-Larouche, Evan

Assignees

Bentley Systems Inc

Abstract

In example embodiments, techniques are provided for classifying elements of infrastructure models using a convolutional graph neural network (GNN). Graph-structured data structures are generated from infrastructure models, in which nodes represent elements and edges represent contextual relationships among elements (e.g., based on proximity, functionality, parent-child relationships, etc.). During training, the GNN learns embeddings from the nodes and edges of the graph-structured data structures, the embeddings capturing contextual clues that distinguish between elements that may share similar geometry (e.g., cross section, volume, surface area, etc.), yet serve different purposes.

Core Innovation

The invention relates to classifying elements in an infrastructure model using a convolutional graph neural network (GNN). A method accesses an inference dataset that includes an infrastructure model having elements with missing class information or misclassifications, and generates a graph-structured data structure in which nodes represent elements and edges represent contextual relationships among elements.

The contextual relationships comprise at least geometric relationships or textual metadata-based relationships. The graph-structured data structure is applied to the convolutional GNN to produce one or more class predictions for each element, where the convolutional GNN produces the class predictions using embeddings learned using recursive neighborhood diffusion.

The method outputs indications of at least some of the produced class predictions, where the output includes correcting misclassifications or filling in missing classes in the infrastructure model. The method further includes storing the infrastructure model with the corrections or filled-in missing classes, and producing outputs that update the classified infrastructure model for further use.

Claims Coverage

The document contains three independent claims: a method for classifying infrastructure-model elements, a non-transitory electronic-device readable medium storing instructions to perform the method, and a computing device configured to perform the method. The independent claims share the same core inventive approach of graph construction from an inference dataset and per-element class prediction using a convolutional GNN with embeddings learned via recursive neighborhood diffusion.

Across the independent claims, classification is performed by converting an inference dataset’s infrastructure model into a graph with contextual edges, applying a convolutional GNN whose embeddings are learned using recursive neighborhood diffusion, and outputting corrected or completed element class information, optionally including display or storage of the updated infrastructure model.

Stated Advantages

Documented Applications