Graph Convolutional Network for Structural Displacement Inference

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The existing finite element method for structural analysis simulations requires solving large-scale simultaneous equations, which results in high calculation amounts and long execution times, making it inefficient for inferring displacement amounts of nodes directly from boundary condition data.

Innovation Solution

A computer-readable medium storing a program that calculates a stiffness matrix and a force vector from mesh data and boundary condition data, and generates feature data from these, which is then used to infer displacement amounts through convolutional operations based on the connection relationships of nodes, employing a graph convolutional network for improved accuracy and efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the finite element method is used to solve large-scale simultaneous equations for structural analysis, then the displacement amounts of nodes can be calculated, but the calculation amount becomes large and the execution time becomes long

Engineering Contradiction:

Improveinference accuracy of displacement amountsVSAvoidexecution time of structural analysis simulation

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The patent replaces the traditional iterative mechanical solving process of the finite element method with a neural network-based inference system. The neural network is trained on simulation data to directly predict displacement amounts from boundary conditions, eliminating the need to solve large-scale simultaneous equations during actual analysis, thus reducing execution time while maintaining accuracy

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent performs preliminary training of the neural network using simulation data generated from the finite element method. By pre-computing the relationship between boundary conditions and displacement amounts during the training phase, the system prepares a ready-to-use inference model that can quickly predict results without repeating the computationally intensive solving process

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

Engineering Contradiction:

Improveinference accuracy of displacement amountsVSAvoidcalculation amount

Core Design Contradiction:

Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent substitutes the computationally intensive iterative solving mechanism of the finite element method with a neural network inference system. The neural network processes boundary condition data through learned patterns rather than mechanical matrix operations, significantly reducing the calculation amount while preserving the ability to accurately predict displacement amounts

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Productivity

If displacement amounts are directly inferred from boundary condition data using convolutional operations, then the execution time is reduced, but the inference accuracy becomes low

Engineering Contradiction:

Improveexecution time of structural analysis simulationVSAvoidinference accuracy of displacement amounts

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent performs preliminary training of the convolutional neural network using simulation data that captures the complex relationship between boundary conditions and displacement amounts. This pre-training phase allows the network to learn accurate mapping patterns, ensuring that subsequent direct inference operations maintain high accuracy while achieving fast execution times

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20230011312A1Inference method and information processing apparatus

Publication Date: 2023.01.12 FUJITSU LTD

AI summary

An information processing apparatus stores mesh data including a plurality of nodes and a plurality of edges and boundary condition data indicating force applied to an object represented by the mesh data. The information processing apparatus calculates a stiffness matrix including a plurality of stiffness values corresponding to the plurality of edges and a force vector including a plurality of force values corresponding to the plurality of nodes. The information processing apparatus generates feature data from the stiffness matrix and the force vector. The information processing apparatus infers a plurality of displacement amounts corresponding to the plurality of nodes by performing a convolutional operation on the feature data in accordance with a connection relationship of the plurality of nodes.