Abstract:In engineering design, surrogate models are widely employed to replace computationally expensive simulations by leveraging design variables and geometric parameters from computer-aided design (CAD) models. However, these models often lose critical information when simplified to lower dimensions and face challenges in parameter definition, especially with the complex 3D shapes commonly found in industrial datasets. To address these limitations, we propose a Bayesian graph neural network (GNN) framework for a 3D deep-learning-based surrogate model that predicts engineering performance by directly learning geometric features from CAD using mesh representation. Our framework determines the optimal size of mesh elements through Bayesian optimization, resulting in a high-accuracy surrogate model. Additionally, it effectively handles the irregular and complex structures of 3D CADs, which differ significantly from the regular and uniform pixel structures of 2D images typically used in deep learning. Experimental results demonstrate that the quality of the mesh significantly impacts the prediction accuracy of the surrogate model, with an optimally sized mesh achieving superior performance. We compare the performance of models based on various 3D representations such as voxel, point cloud, and graph, and evaluate the computational costs of Monte Carlo simulation and Bayesian optimization methods to find the optimal mesh size. We anticipate that our proposed framework has the potential to be applied to mesh-based simulations across various engineering fields, leveraging physics-based information commonly used in computer-aided engineering.

What problem does this paper attempt to address?

The problems that this paper attempts to solve are as follows: In engineering design, traditional surrogate models face the problems of information loss and difficult parameter definition when dealing with complex 3D CAD data. Specifically: 1. **Information loss**: When simplifying 3D CAD data into low - dimensional representations, critical geometric information is often lost. 2. **Difficult parameter definition**: Especially for complex 3D shapes common in industrial datasets, traditional methods are difficult to effectively define parameters. 3. **Trade - off between computational efficiency and accuracy**: Inappropriate mesh sizes will lead to either over - simplification (losing important geometric details) or excessive computational burden. To address these challenges, the paper proposes a graph neural network (GNN) framework based on Bayesian optimization to directly learn geometric features from the mesh representation of CAD models to predict engineering performance. This framework determines the optimal mesh element size through Bayesian optimization, thereby improving computational efficiency while maintaining high prediction accuracy. ### Specific problem description - **Limitations of low - dimensional representations**: Traditional surrogate models usually rely on geometric parameters or design variables, and these parameters have limitations when representing complex 3D CAD data. Low - dimensional representations cannot capture the inherent geometric complexity of 3D shapes, while high - dimensional data is difficult to parameterize, affecting data resolution. - **High computational cost**: Traditional computer simulations can provide detailed representations of physical systems, but the computational cost is very high, especially in the design optimization process that requires multiple iterations. - **Deficiencies of existing methods**: Existing 3D data representation methods (such as voxels, point clouds, etc.) have their own limitations when dealing with complex 3D structures and cannot simultaneously take into account memory efficiency, rendering ability, and the ability to handle irregular inputs. ### Solution The solution proposed in the paper is to use graph neural networks (GNN) combined with Bayesian optimization to build an efficient surrogate model. The specific steps are as follows: 1. **Data pre - processing**: Convert 3D CAD data into polygon meshes and adjust the mesh size through re - partitioning and clustering. 2. **Model training**: Use the GNN model for training and minimize the mean square error (MSE) between the predicted values and the real values. 3. **Bayesian optimization**: Determine the optimal mesh element size through Bayesian optimization to ensure that the model reduces unnecessary computational costs while maintaining high prediction accuracy. Through this method, the paper aims to develop a surrogate model that can efficiently and accurately predict engineering performance in complex engineering simulations, thereby replacing expensive computer simulations.

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

Scientific Computing Algorithms to Learn Enhanced Scalable Surrogates for Mesh Physics

An FEA surrogate model with Boundary Oriented Graph Embedding approach

Extrapolative Bayesian Optimization with Gaussian Process and Neural Network Ensemble Surrogate Models

Deep Neural Network-Based Surrogate-Assisted Inverse Optimization for High-Speed Interconnects

MAgNET: A Graph U-Net Architecture for Mesh-Based Simulations

DEBOSH: Deep Bayesian Shape Optimization

Deep Learning-based surrogate models for parametrized PDEs: handling geometric variability through graph neural networks

Enhanced data efficiency using deep neural networks and Gaussian processes for aerodynamic design optimization

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

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

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

Dynamic graph-based convergence acceleration for topology optimization in unstructured meshes

Gaussian process regression + deep neural network autoencoder for probabilistic surrogate modeling in nonlinear mechanics of solids

MMGP: a Mesh Morphing Gaussian Process-based machine learning method for regression of physical problems under non-parameterized geometrical variability

Neural P$^3$M: A Long-Range Interaction Modeling Enhancer for Geometric GNNs

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

Proposing an intelligent mesh smoothing method with graph neural networks

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

Bayesian Optimization on Large Graphs via a Graph Convolutional Generative Model: Application in Cardiac Model Personalization

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