A graph neural network based chemical mechanism reduction method for combustion applications
arXiv:2603.22318v1 Announce Type: new Abstract: Direct numerical simulations of turbulent reacting flows involving millions of grid points and detailed chemical mechanisms with hundreds of species and thousands of reactions are computationally prohibitive. To address this challenge, we present two data-driven chemical mechanism reduction formulations based on graph neural networks (GNNs) with message-passing transformer layers that learn nonlinear dependencies among species and reactions. The first formulation, GNN-SM, employs a pre-trained surrogate model to guide reduction across a broad range of reactor conditions. The second formulation, GNN-AE, uses an autoencoder formulation to obtain highly compact mechanisms that remain accurate within the thermochemical regimes used during training. The approaches are demonstrated on detailed mechanisms for methane (53 species, 325 reactions), ethylene (96 species, 1054 reactions), and iso-octane (1034 species, 8453 reactions). GNN-SM achieve
arXiv:2603.22318v1 Announce Type: new Abstract: Direct numerical simulations of turbulent reacting flows involving millions of grid points and detailed chemical mechanisms with hundreds of species and thousands of reactions are computationally prohibitive. To address this challenge, we present two data-driven chemical mechanism reduction formulations based on graph neural networks (GNNs) with message-passing transformer layers that learn nonlinear dependencies among species and reactions. The first formulation, GNN-SM, employs a pre-trained surrogate model to guide reduction across a broad range of reactor conditions. The second formulation, GNN-AE, uses an autoencoder formulation to obtain highly compact mechanisms that remain accurate within the thermochemical regimes used during training. The approaches are demonstrated on detailed mechanisms for methane (53 species, 325 reactions), ethylene (96 species, 1054 reactions), and iso-octane (1034 species, 8453 reactions). GNN-SM achieves reductions comparable to the established graph-based method DRGEP while maintaining accuracy across a wide range of thermochemical states. In contrast, GNN-AE achieves up to 95% reduction in species and reactions and outperforms DRGEP within its target conditions. Overall, the proposed framework provides an automated, machine-learning-based pathway for chemical mechanism reduction that can complement traditional expert-guided analytical approaches.
Executive Summary
This article presents two data-driven chemical mechanism reduction formulations based on graph neural networks (GNNs) for combustion applications. The proposed methods, GNN-SM and GNN-AE, utilize message-passing transformer layers to learn nonlinear dependencies among species and reactions, allowing for significant reductions in computational complexity. The approaches are demonstrated on detailed mechanisms for methane, ethylene, and iso-octane, achieving reductions comparable to or surpassing established graph-based methods. The framework provides an automated, machine-learning-based pathway for chemical mechanism reduction that can complement traditional expert-guided analytical approaches. The authors' work has the potential to significantly enhance the efficiency and accuracy of combustion simulations, making it an important contribution to the field.
Key Points
- ▸ The article presents two GNN-based chemical mechanism reduction formulations: GNN-SM and GNN-AE.
- ▸ GNN-SM uses a pre-trained surrogate model to guide reduction across a broad range of reactor conditions.
- ▸ GNN-AE uses an autoencoder formulation to obtain highly compact mechanisms that remain accurate within the thermochemical regimes used during training.
- ▸ The approaches are demonstrated on detailed mechanisms for methane, ethylene, and iso-octane.
Merits
Improved Efficiency
The proposed methods can significantly reduce computational complexity, making combustion simulations more efficient and accurate.
Automated Approach
The framework provides an automated, machine-learning-based pathway for chemical mechanism reduction that can complement traditional expert-guided analytical approaches.
Scalability
The GNN-based approach can be scaled to handle large and complex chemical mechanisms, making it a valuable tool for the field.
Demerits
Limited Generalizability
The proposed methods may not generalize well to other types of chemical reactions or mechanisms, limiting their applicability.
Dependence on Training Data
The accuracy of the GNN-based approach depends on the quality and diversity of the training data, which can be a limitation in certain scenarios.
Interpretability
The GNN-based approach may lack interpretability, making it challenging to understand the underlying chemical mechanisms and reactions.
Expert Commentary
The article presents a significant contribution to the field of chemical engineering, particularly in the context of combustion simulations. The proposed GNN-based approach has the potential to revolutionize the field by providing an automated, machine-learning-based pathway for chemical mechanism reduction. However, the methods' limitations, such as limited generalizability and dependence on training data, must be carefully considered. Additionally, the interpretability of the GNN-based approach is a concern that must be addressed to ensure the methods' widespread adoption. Overall, the article's findings have significant implications for both practical and policy-related applications, making it an important contribution to the field.
Recommendations
- ✓ Further research is needed to address the limitations of the proposed methods, such as limited generalizability and dependence on training data.
- ✓ The GNN-based approach should be applied to a wider range of chemical reactions and mechanisms to evaluate its scalability and generalizability.
Sources
Original: arXiv - cs.LG
