Data-driven modelling of spray flows: Current status and future direction

Abstract

Spray flows are crucial in a variety of engineering applications across sectors such as energy and mobility, particularly in enhancing the performance of internal combustion engines, which are integral to the transition to net-zero emissions. However, accurately characterising these flows presents significant challenges due to the complex multiphysics and multiscale phenomena involved, especially when modelling reacting spray flows with turbulence-chemistry interactions. Machine learning (ML) algorithms present promising data-driven solutions that could enhance the accuracy and efficiency of computational fluid dydnamics (CFD) models, uncover underlying physical mechanisms, and optimise spray flow processes. This paper outlines the challenges and opportunities associated with integrating CFD and ML algorithms for spray flow modelling, with a particular focus on spray combustion to improve predictive capabilities. It provides a comprehensive review of existing literature on various CFD models and ML algorithms applied to key aspects of spray dynamics, such as atomisation, droplet transport, and combustion. Despite significant progress, ML applications in spray modelling continue to face challenges, primarily due to the complexity and variability of spray dynamics. These challenges include the need for high-quality, domain-specific data, which is often difficult and costly to obtain, as well as issues related to model generalisation. Furthermore, the wide range of scales inherent in spray flows along with the challenges in quantifying uncertainties present significant difficulties for ML models. The insights provided in this study can contribute to identifying research areas to improve the accuracy of spray modelling.