Bridging Data and Physics: A Graph Neural Network-Based Hybrid Twin Framework

TL;DR

The paper tackles the challenge of bridging physics-based simulations with real-world behavior by introducing a Graph Neural Network–based hybrid twin that learns the spatial discrepancy (ignorance gap) between a FEM solution and observed data. The approach combines a physics-driven FEM model with a data-driven GNN to predict and correct the gap, enabling accurate nonlinear heat-transfer predictions using sparse spatial measurements. Through extensive synthetic experiments across multiple meshes, loading patterns, and geometries, the method demonstrates strong generalization and data efficiency, outperforming a pure data-driven baseline. The work suggests practical benefits for industrial simulations by reducing data requirements and improving interpretability, with future directions including attention-based enhancements and real-world validation.

Abstract

Simulating complex unsteady physical phenomena relies on detailed mathematical models, simulated for instance by using the Finite Element Method (FEM). However, these models often exhibit discrepancies from the reality due to unmodeled effects or simplifying assumptions. We refer to this gap as the ignorance model. While purely data-driven approaches attempt to learn full system behavior, they require large amounts of high-quality data across the entire spatial and temporal domain. In real-world scenarios, such information is unavailable, making full data-driven modeling unreliable. To overcome this limitation, we model of the ignorance component using a hybrid twin approach, instead of simulating phenomena from scratch. Since physics-based models approximate the overall behavior of the phenomena, the remaining ignorance is typically lower in complexity than the full physical response, therefore, it can be learned with significantly fewer data. A key difficulty, however, is that spatial measurements are sparse, also obtaining data measuring the same phenomenon for different spatial configurations is challenging in practice. Our contribution is to overcome this limitation by using Graph Neural Networks (GNNs) to represent the ignorance model. GNNs learn the spatial pattern of the missing physics even when the number of measurement locations is limited. This allows us to enrich the physics-based model with data-driven corrections without requiring dense spatial, temporal and parametric data. To showcase the performance of the proposed method, we evaluate this GNN-based hybrid twin on nonlinear heat transfer problems across different meshes, geometries, and load positions. Results show that the GNN successfully captures the ignorance and generalizes corrections across spatial configurations, improving simulation accuracy and interpretability, while minimizing data requirements.

Figures (14)

• Figure 1: Scheme of the proposed approach of the heat transfer model. The model takes a linear simulation frame as an input. Node and edge features are passed through the encoder that projects them into the hidden space. The resulting features are then passed through $K$ message passing layers that update node features in the hidden space. The decoder outputs the difference of temperature between linear and nonlinear simulations. The output is added to the input to obtain the nonlinear simulation frame.
• Figure 2: The schematic domain representation of the data used for the use cases: (a) dataset A1; (b) dataset A2; (c) dataset A3; (d) dataset A4; (e) dataset A5; (f) dataset A6; (g) dataset A7; (h) dataset A8. The mesh nodes on the heat source are in red, the nodes of the Dirichlet BC are in blue.
• Figure 3: Schematic representation of the domain in dataset B1. The heat source is normally distributed on a plate (red), each design has a different placement of the heat source center (yellow). The Dirichlet BC is on the left side of the plate (blue).
• Figure 4: Parameterization of the domain shape in B2 dataset.
• Figure 5: Schematic representation of the simulation domains used in dataset B2. The mesh nodes on the heat source are in red, the nodes of the Dirichlet BC are in blue.