"Prediction and control of extreme fluid dynamics with artificial intelligence"

这个跨学科的项目涉及以下五个EPSRC研究领域：人工智能技术、燃烧工程、流体动力学和空气动力学、复杂性科学、数值分析。高保真湍流模拟越来越多地被学术界用来了解湍流物理(直接数值模拟，DNS)，并被工业(大涡模拟，LES)作为设计工具，以最大限度地减少昂贵的测试。虽然湍流统计量的计算是准确的，但对极端事件，如猛烈的湍流爆发和自燃混合点，还没有实现时间和空间上的准确预测。无论模拟代码多么精确，极端事件的时空预测都受到反应性和非反应性湍流的混沌性质的限制。该项目将利用人工智能和机器学习算法与动力系统理论相结合，实现对极端事件的预测和物理理解。将设计和训练一种优化算法来控制极端流体动力学，特别是在与能量收集和航空推进相关的多物理问题中。

This cross-disciplinary project falls in the following five EPSRC research areas: Artificial intelligence technologies, Combustion engineering, Fluid Dynamics and Aerodynamics, Complexity Science, Numerical Analysis.High-fidelity simulations of turbulent flows have increasingly been adopted by academia to understand the physics of turbulence (direct numerical simulation, DNS), and by industry (large-eddy simulation, LES) as design tools to minimize costly testing. Whereas the calculation of the statistics of turbulent flows can be accurate, the time and space accurate prediction of extreme events, such as violent turbulent bursts and auto-ignited spots of mixture, has not been achieved yet. No matter how accurate the simulation code is, the time and space prediction of extreme events is limited by the chaotic nature of reacting and non-reacting turbulence. This project will enable the prediction and physical understanding of extreme events by leveraging artificial intelligence and machine learning algorithms with dynamical systems' theory. An optimisation algorithm will be designed and trained to control extreme fluid dynamics, in particular in multi-physical problems that are relevant to energy harvesting and aeronautical propulsion.