AI-Powered Surrogate Modelling for Multiscale Combustion: A Critical Review and Opportunities

TL;DR

This paper reviews AI-driven surrogate models in multiscale combustion, highlighting recent advances, challenges, and future directions for reliable, scalable, and physics-informed modeling approaches.

Contribution

It provides a comprehensive assessment of current AI surrogate modeling techniques in combustion, comparing methods and identifying key challenges and future opportunities.

Findings

AI models reduce computational costs in combustion simulations.

Supervised, unsupervised, and hybrid learning approaches are evaluated.

Challenges include transferability, extrapolation errors, and dataset inconsistencies.

Abstract

Recent advances in combustion science have led to the generation of large volumes of data from high-fidelity simulations, detailed chemical-kinetic calculations and engine-relevant measurements and create new opportunities for data-driven modelling across interacting physical and chemical scales. Among these approaches, artificial intelligence has emerged as a promising framework for constructing surrogate models that reduce computational costs, deliver substantial speed-up and support prediction in complex reacting systems. This review provides a state-of-the-art assessment of AI-powered surrogate modelling for multiscale combustion, spanning chemical kinetics, mechanism reduction, turbulent flames, combustors, engines, and emissions prediction. Supervised, unsupervised, and hybrid or physics-guided learning approaches are examined and compared in terms of predictive accuracy, physical consistency, computational efficiency, and generalizability across conditions and scales. The review further discusses key challenges, including limited transferability across fuels and operating regimes, extrapolation errors, inconsistency in datasets and benchmarks, and the difficulty of building robust and trustworthy models for practical combustion workflows. Future opportunities are identified in the development of more reliable, scalable, and physically grounded surrogate frameworks for next-generation combustion research.