“Ah,” says the wizened old aerodynamics engineer, “colourful fluid dynamics, eh?”
If I has a pound for every time over the last 10 years I’ve heard this, and also had to smile engagingly at the standard put-down of computational fluid dynamics (CFD) I’d probably have about £100 by now.
Not that I disagree entirely with the sentiment. CFD rode the traditional peak of overinflated expectations (“CFD in a bottle” – for those of you who remember one of the earliest and most effective advertising campaigns) through the trough of despair (realising that scale-resolving turbulent simulation was going to take more computing power than available on the entire planet) to the current plateau of industrial acceptance subject to statistical turbulence models with known deficiencies. The constant has been our delight in producing the most aesthetically appealing flow visualisations that can be supported by the available graphics hardware. Colourful indeed – and now 3D movies as well!
It is easy then to scoff at the quantifiable limitations and apparent sartorial vanity of CFD and note that in many instances the (single-digit accurate) turbulence model parameters have been chosen in order to roughly match some measured output of a proper experimental result – one produced in a wind tunnel. I also wonder whether the overinflated expectations of CFD in the 80’s and 90’s have created a tranche of (now very senior) engineering managers with an over-corrected bias away from numerical methods.
Thus we come to the conclusion that CFD is useful as a precursor to a proper experimental analysis – perhaps as a way of downsizing an enormous set of design possibilities so that the finely tuned wind tunnel can inform our final selection – and keep the chief engineer happy. After all, no experienced engineer is about to sign off without a hefty data report from the tunnel.
Strangely, when I describe the general reverence for wind tunnel data with staff members who actually work in them, a certain guilty mist enters the room. “We have to calibrate the tunnel”, the conversation starts, “and of course we can’t match both the Mach number and the Reynolds number at the same time, so we modify the geometry to account for the known differences.” It also turns out that the tunnel model deflects under load; the supporting structure interferes with the flow; the measurement probes are prone to blockage; the flow in the tunnel isn’t uniform, and cannot be made to reflect the appropriate unsteady inlet conditions; the outputs are scattered like the Milky Way on a dark night (a thick pen fit through the points providing the reader with a navigational aid) and the results the next day are completely different (even when the Germans do it).
It is sometimes suggested that CFD can produce data more quickly than wind tunnels, and I wonder if this is because we are used to seeing very condensed reporting from experiments. Admittedly, it can take a long time to make a high quality physical model - but once you have one, a wind tunnel (suitably instrumented) can produce eye-watering amounts of data that is often heavily processed before it is even seen by human operators. An automated system for rotating or moving an experimental object in a wind tunnel can produce in a short time a very large amount of analysis data – in many cases far more than would be expected from a CFD-based system on a very large computer.
Ultimately then we have complementary systems – and the challenge for engineering is to work out how best to combine them. It is still the case that some of the very best experiments (see for example the ERCOFTAC Knowledge Base WIKI at http://qnet-ercoftac.cfms.org.uk/w/index.php/Main_Page) are designed to stress the modelling of underlying flow regimes rather than specific products. Greater industrial use of this model might add significant value to the tunnel data – in its fullest extent.
In its least scientifically constructive form, parties not wishing to be exposed as inferior sometimes withhold background data, and we need to find a healthier and more open partnership model.
I am optimistic that the emerging age of data-centric engineering will provide a technology infrastructure that is up to the task of fusing results from different sources to create a capability greater than the sum of the parts.
We will need to agree on the colour though!