Graph Neural Network for Irregular BIM Data Representation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies struggle to efficiently process and query Building Information Models (BIMs) due to their complex and irregular data structures, making it difficult to feed them into deep learning models for predictive tasks.
Innovation Solution
The approach involves translating BIM models into graph structures and utilizing recent advances in deep learning, such as self-supervised learning and multimodal learning, to map BIM models into dense vector representations using a stack of deep models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If BIM models are processed using traditional deep learning architectures, then the processing speed and efficiency improve, but the compatibility and accuracy deteriorate due to irregular data structures
Solution Approach 1:
The patent introduces a graph structure as an intermediary representation layer between the irregular BIM data and the deep learning model. BIM models are first converted into graph structures where building elements become nodes and their relationships become edges, creating a standardized intermediate format that preserves the irregular data semantics while making it compatible with graph neural network processing
Solution Approach 2:
The patent transforms the parameter representation of BIM data by converting geometric and semantic attributes into graph-based features. Node features include element properties (area, volume, material), while edge features capture spatial relationships (adjacency, connectivity, hierarchy). This parameter transformation enables the irregular BIM data to be processed by structured deep learning architectures
2Ease of operation
If BIM models are treated as independent files, then the file management and processing simplicity improve, but the loss of cross-model patterns and rich information deteriorates
Solution Approach 1:
The patent merges multiple independent BIM models into a unified graph-based database structure. Instead of processing models as isolated files, the system combines them into a single graph database where nodes and edges can span across multiple original model boundaries, enabling cross-model pattern recognition and information retrieval while maintaining efficient query operations
3Device complexity
If manual search methods are used to query BIM models, then the implementation complexity and computational resources required are reduced, but the search speed and retrieval efficiency deteriorate
Solution Approach 1:
The patent replaces manual search mechanisms with an automated graph neural network-based retrieval system. The system uses graph embeddings to represent BIM models in a continuous vector space, enabling semantic similarity search through distance metrics. This substitution of mechanical search with intelligent algorithms dramatically improves retrieval speed and accuracy while the graph database structure keeps the overall system complexity manageable
Data Source
AI summary
A method and system provide the ability to utilize three-dimensional (3D) models to perform a predictive task. Multiple 3D models, consisting of non-Euclidean data, are obtained. Each 3D model is translated into a relational graph with nodes and edges. Each relational graph is processed using a graph neural network (GNN) that computes a node representation per node. The node representations are aggregated into a structural representation of the 3D model. Multiple different views of the 3D model are captured and passed through a convolutional neural network (CNN) to compute a view representation of each view. The view representations are aggregated into a single visual representation. The GNN and CNN are trained using a multiview contrastive training objective to maximize agreement between the structural representation and the single visual representation to form final learned representations. The final learned representation is utilized to perform the predictive task.