3D Multi-Graph Neural Networks for Faster Physical Object Behavior Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer-based systems face challenges in learning 3D object geometries for various physics problems with complex material properties and interactions, often relying on pre-labeled parameters that lack necessary information for adapting to new geometric features, and require computationally expensive physics calculations.
Innovation Solution
Integrating information from 3D geometries, finite element models, and simulation data to construct graph/multi-graph representations, using advanced machine learning techniques to predict physics solutions efficiently, without re-running costly simulations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional finite element systems are used to determine object behavior, then physics-based simulation accuracy is maintained, but computational cost and time consumption increase significantly
Solution Approach 1:
The patent creates a graph-based copy of the finite element model structure, where nodes and elements are represented as graph nodes and edges. This graph representation captures the essential topological relationships of the FE model without requiring full physics calculations, enabling fast prediction of object behavior while maintaining reasonable accuracy.
Solution Approach 2:
The patent replaces the traditional mechanical physics-based finite element calculation system with a machine learning-based prediction system. The system learns from historical FE simulation data and uses the learned patterns to predict object behavior, substituting computationally expensive physics calculations with faster ML inference.
2Ease of operation
If pre-labeled parameters are used in conventional systems, then parameter extraction is simplified, but adaptability to new geometric features is lost
Solution Approach 1:
The patent performs preliminary extraction of geometric features and model parameters during the graph construction phase, before the actual prediction task. This preliminary action captures essential geometric characteristics in a structured graph format that can be efficiently processed by the machine learning model, simplifying subsequent operations while maintaining adaptability.
Solution Approach 2:
The patent transforms traditional pre-labeled parameters into a graph-based representation that captures geometric features in a more flexible and adaptive manner. The graph structure allows the system to naturally adapt to new geometric configurations by learning from the topological relationships rather than relying on fixed parameter labels.
3Reliability
If historical physics-based data is leveraged in traditional systems, then learning from past designs is enabled, but computational expense must be repeated for each new design
Solution Approach 1:
The patent performs comprehensive physics-based simulations and data extraction in advance during the training phase, storing the results in a structured graph format. When new designs need to be evaluated, the system only requires fast graph-based ML inference rather than repeating expensive physics calculations, significantly reducing computational energy consumption while maintaining the ability to learn from historical data.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Embodiments perform neural network based behavior determination of physical objects. One such embodiment processes a three-dimensional (3D) numerical-method model representing a physical object to extract (i) 3D geometric data associated with the physical object and (ii) simulation data. The extracted 3D geometric data and simulation data are transformed into a 3D multi-graph. The 3D multi-graph is processed with one or more deep neural network (DNN) and one or more operators to determine behavior of the physical object.