A Hybrid GNN approach for predicting node data for 3D meshes

Shwetha Salimath,Francesca Bugiotti,Frederic Magoules

2023-10-23

Abstract:Metal forging is used to manufacture dies. We require the best set of input parameters for the process to be efficient. Currently, we predict the best parameters using the finite element method by generating simulations for the different initial conditions, which is a time-consuming process. In this paper, introduce a hybrid approach that helps in processing and generating new data simulations using a surrogate graph neural network model based on graph convolutions, having a cheaper time cost. We also introduce a hybrid approach that helps in processing and generating new data simulations using the model. Given a dataset representing meshes, our focus is on the conversion of the available information into a graph or point cloud structure. This new representation enables deep learning. The predicted result is similar, with a low error when compared to that produced using the finite element method. The new models have outperformed existing PointNet and simple graph neural network models when applied to produce the simulations.

Machine Learning,Numerical Analysis

What problem does this paper attempt to address?

The paper attempts to address the problem of how to efficiently find the optimal set of input parameters to improve the efficiency of mold manufacturing in the metal forging process. The current method is to generate simulations under different initial conditions using the Finite Element Method (FEM) to predict the optimal parameters, but this method is time-consuming and requires high computational resources. To this end, the authors propose a hybrid Graph Neural Network (GNN) approach, aiming to handle and generate new data simulations through a graph convolution-based surrogate model, thereby significantly reducing time costs. This method converts existing 3D mesh data into graph or point cloud structures and uses deep learning techniques to predict results, achieving accuracy comparable to the Finite Element Method but with faster speed and lower cost. Specifically, the model proposed in the paper can generate a new simulation within 300 milliseconds after training, whereas the traditional FEM method takes about 45 minutes. Additionally, the model shows a lower mean absolute error when predicting output features such as wear, demonstrating its effectiveness and efficiency in metal forging simulations. In this way, the researchers hope to provide a faster, more economical, and accurate parameter optimization method for the metal forging process.

A Hybrid GNN approach for predicting node data for 3D meshes

Graph Neural Networks for Mesh Generation and Adaptation in Structural and Fluid Mechanics

Scientific Computing Algorithms to Learn Enhanced Scalable Surrogates for Mesh Physics

Topology-Agnostic Graph U-Nets for Scalar Field Prediction on Unstructured Meshes

Bayesian Mesh Optimization for Graph Neural Networks to Enhance Engineering Performance Prediction

MeshPointNet: 3D Surface Classification Using Graph Neural Networks and Conformal Predictions on Mesh-Based Representations

X-MeshGraphNet: Scalable Multi-Scale Graph Neural Networks for Physics Simulation

Knowledge and Data Dual-Driven Graph Network for Tumbler Strength Prediction in Sintering Process

Learning Mesh-Based Simulation with Graph Networks

Graph Neural Networks (gnns) Based Accelerated Numerical Simulation

MeshingNet: A New Mesh Generation Method based on Deep Learning

Predicting Material Properties Using a 3D Graph Neural Network with Invariant Local Descriptors

Physics-informed MeshGraphNets (PI-MGNs): Neural finite element solvers for non-stationary and nonlinear simulations on arbitrary meshes

Geom-DeepONet: A Point-cloud-based Deep Operator Network for Field Predictions on 3D Parameterized Geometries

Virtual Foundry Graphnet for Metal Sintering Deformation Prediction

Multi-GPU Approach for Training of Graph ML Models on large CFD Meshes

Enhancing Graph U-Nets for Mesh-Agnostic Spatio-Temporal Flow Prediction

Multiscale graph neural networks with adaptive mesh refinement for accelerating mesh-based simulations

Mesh Neural Networks Based on Dual Graph Pyramids

Prediction of nano metal matrix composites based on hybrid approach