Air flowing over a wing or turbine blade behaves like a string: it starts smooth and layered: that's the laminar regime. But as speed increases or external ripples appear, the string breaks into chaotic vibration — turbulence. Predicting the exact moment of breakdown is crucial: it affects drag, heating, and noise.
The classic T3 series of experiments, conducted on a flat plate in a wind tunnel, became the benchmark for testing models. However, they measured velocity only at discrete points. Modern computational methods need a full three-dimensional picture over time. To get it, the researchers ran a "numerical experiment"—high-fidelity simulation using the implicit large-eddy simulation (iLES) method. They reproduced the conditions of those T3 experiments, but now they could see every tiny eddy in the volume.
The simulation results matched the friction and velocity profile measurements from the old experiments well. But when the same data were fed to two popular engineering models (SA-BCM and k-ω-γ), it turned out that both systematically err on both the onset and the extent of the transition. The predicted breakdown either lags or stretches out.
The resulting database is not just a check of old models. It is a foundation for training new ones based on physics-informed neural networks. These can capture subtle patterns that escape simplified formulas. This approach promises to make calculations more accurate, and wings and blades more reliable.
On a flat plate, turbulence can arise even without a single imperfection — chaotic fluctuations in the incoming flow are enough, like ripples on water from a distant wind.
In Andy Weir's novel The Martian, turbulent flow breakdown in a dust storm becomes a catastrophe — more accurate transition models could have predicted it in advance.