Interpreting CFD Surrogates through Sparse Autoencoders

2025-07-21

AI Summary

Graph neural network (GNN)-driven CFD surrogate models suffer from limited interpretability, hindering their deployment in safety-critical applications. Method: We propose a post-hoc interpretability framework based on sparse autoencoders (SAEs), which constructs an overcomplete dictionary in the pre-trained GNN’s node embedding space to learn physically meaningful implicit primitives—e.g., vorticity structures and shear layers—marking the first application of SAEs to interpret graph-based CFD surrogates. Contribution/Results: The learned dictionary exhibits unambiguity and model-agnosticism, enabling cross-architecture interpretation and visualization-based validation. Experiments demonstrate precise identification and disentanglement of key flow physics without modifying the original surrogate model, significantly enhancing decision transparency and trustworthiness. This approach provides a novel pathway toward regulatory compliance and engineering adoption of high-fidelity CFD surrogates.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Computer Vision: Interpretability, Explainability, and Transparency
• Reasoning under Uncertainty: Graphical Models

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semantics
• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding

Abstract

Learning-based surrogate models have become a practical alternative to high-fidelity CFD solvers, but their latent representations remain opaque and hinder adoption in safety-critical or regulation-bound settings. This work introduces a posthoc interpretability framework for graph-based surrogate models used in computational fluid dynamics (CFD) by leveraging sparse autoencoders (SAEs). By obtaining an overcomplete basis in the node embedding space of a pretrained surrogate, the method extracts a dictionary of interpretable latent features. The approach enables the identification of monosemantic concepts aligned with physical phenomena such as vorticity or flow structures, offering a model-agnostic pathway to enhance explainability and trustworthiness in CFD applications.

Problem

• Interpret latent representations of CFD surrogate models
• Enhance explainability of graph-based CFD surrogates
• Align learned features with physical phenomena

Innovation

• Leveraging sparse autoencoders for interpretability
• Extracting interpretable latent features dictionary
• Identifying monosemantic concepts with physical phenomena