Machine Learning Stiffness Matrix Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The generation of stiffness matrices in structural analysis simulations can be computationally intensive, leading to extended execution times and bottlenecks in the simulation process, especially when using traditional finite element methods.
Innovation Solution
A machine learning model, specifically a graph convolutional neural network, is trained to estimate element stiffness matrices by generating feature vectors based on node locations, elastic moduli, and stress applied to nodes, performing convolutional operations, and producing stiffness matrices more efficiently than traditional methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional finite element methods are used to generate stiffness matrices, then accuracy of structural analysis is maintained, but execution time increases and computational bottlenecks occur
Solution Approach 1:
The patent creates a simplified copy of the stiffness matrix generation process using machine learning. Instead of performing full finite element calculations, a neural network is trained to replicate the essential mapping from geometric and material inputs to stiffness matrix outputs, dramatically reducing computation time while preserving accuracy
Solution Approach 2:
The patent performs preliminary training of the machine learning model using data from traditional finite element analyses. By pre-computing and storing the relationships between input parameters and stiffness matrices in the neural network, the system eliminates the need for repeated heavy computations during actual structural analysis
2Reliability
If traditional finite element methods are used to generate stiffness matrices, then comprehensive mechanical properties are computed, but computational complexity increases
Solution Approach 1:
The patent replaces the traditional mechanical/computational system of finite element matrix generation with a machine learning system. The neural network substitutes the complex iterative numerical computations with learned patterns, reducing computational complexity while maintaining the ability to predict mechanical properties
3Productivity
If machine learning models are used to estimate stiffness matrices, then execution time is reduced, but model training complexity increases
Solution Approach 1:
The patent performs the complex model training operation in advance, before actual structural analysis is needed. The trained model can then be deployed for rapid predictions, separating the heavy training phase from the fast inference phase to achieve both speed and accuracy
Data Source
AI summary
An information processing apparatus generates a feature vector for each of a plurality of nodes included in an element in mesh data, based on a location of the corresponding node, an elastic modulus at the corresponding node, and stress applied to the corresponding node. The information processing apparatus enters the feature vectors of the nodes to a trained machine learning model, performs a convolutional operation on the feature vectors of the nodes, and estimates an element stiffness matrix indicating stiffness of the element based on a result of the convolutional operation.