Researchers from Aarhus University aim to develop a new way of modelling complex turbulent systems. This is necessary to design, for example, less expensive, more productive wind farms, more efficient pumps, or for making more reliable weather forecasts, to name just a few.
While it is often not fully appreciated, turbulent flows play a central role in an incredible number of both natural and engineering systems. Understanding and predicting the exact details of turbulent flows is of vital importance, but it currently requires simulations at a level of computational cost that we are unable to achieve.
"Even with modern supercomputers, it’s impossible to simulate many of the turbulent systems that are so important to understand. For example, we can still only simulate a tiny section of a large wind farm, and if we can’t accurately simulate the turbulent atmospheric flow through the whole farm, then it’s very difficult to optimise the farm for future energy production needs,” says Associate Professor Mahdi Abkar from Aarhus University, Department of Mechanical and Production Engineering.
He recently received a DKK 6.2 million Sapere Aude grant from the Independent Research Fund Denmark for a new project titled ‘Physics-constrained Learning for Turbulent Flows (PLTF)’.
"The problem with turbulence modelling is about much more than just wind farms. Perhaps we want to design a more fuel-efficient engine, aircrafts that don’t make as much noise, or upgrade the energy efficiency in the shipping industry. Normally, we would simulate the systems and optimise the physics, but we can’t do this here, because turbulent flow is so complex that it requires simulations with impossibly high resolutions. That’s why we model the system, but as soon as we’ve done so the model is inadequate because it’s built on many simplified assumptions under idealised conditions. This is the challenge with turbulence: it’s an extremely complex, multi-scale, unsteady, and non-linear problem," says the associate professor.
Mahdi Abkar’s goal with the PLTF project is to create a paradigm shift within turbulent flow modelling. He wants to develop a new kind of modelling approach, which is coupled with advanced machine learning techniques. The central goal is to tackle the high computational costs while still being able to deliver on accuracy:
"In the PLTF project, we propose a new approach to turbulence modelling using physics-constrained machine learning. The hypothesis is that we’ll be able to capture the underlying physics that is ignored in current models, through data-driven models using state-of-the-art machine learning techniques," says Mahdi Abkar.
If the research team succeeds, it will be a breakthrough in understanding and modelling of complex turbulent systems. Abkar continues:
"The model can then be used to model flow problems in any context. We believe that this project will increase our knowledge of modelling fluid mechanics and turbulent systems. Based on the enormous amount of data that we have access to, and the massive investment that industry is devoting to using data-driven design and optimisation, this project could produce ground-breaking results," he says.