Until quite recently, CFD (Computational Fluid Dynamics) was seen by many as a good technique, but not accurate enough for any useful work.
In fact, some prominent racing car designers believed that if CFD came within 400 per cent of the wind tunnel results, then it was a good day. That was mainly because a few years ago the only way to get results in reasonable time was to simplify the input, whichautomatically undermined the usefulness of the final data.
Four or five years ago the limited capabilities of the software made it unthinkable that only the most experienced could use CFD only small, discrete components or pieces of cars. never complete ones.
It doesn’t seem too long ago that any kind of computationally intensive activity involved large air-conditioned rooms full of very expensive super-cooled hardware – the stuff of sci-fi films. The kinds of problems that these high-powered computers were directed at were geophysical, like climate modelling or seismic analysis – in fact, any kind of ‘domain’ problem that required the solution of millions of simultaneous equations.
Except for the tricky turbulence models, the mathematics is not that complicated – just enormous and highly repetitive (hence the use of these number crunchers in the first place).
CFD is one of engineering science’s final frontiers. If you take the view that finite element analysis is difficult, and non-linear FEA is more difficult still, then CFD is close to ‘rocket science’ in terms of complexity and difficulty.
One of the main popular applications of CFD is in the automotive industry and racing car design. But it can play a significant role anywhere there is fluid flow, from the ventilation of home PCs to the discharge of grain silos.
The main message is that consistent, accurate and repeatable CFD is now possible in a time-frame which allows it to be used for design purposes. In the past, CFD was used to calibrate simple components, such as the 2D wing profiles of racing cars. Today’s accessible hardware is both fast and capable enough to do a full vehicle complex 3D CFD in a reasonable time-frame.
The equipment is still expensive, but you don’t have to work for government agencies or multinational companies to afford it. In fact, with the advent of multi-processor workstations, 3GHz Intel processors and RAM frequently measured in gigabytes, supercomputing capability is available for the cost of an expensive family car.
‘The breakthrough with CFD simulation came with the introduction of relatively ‘affordable’ high-performance computing and the arrival of efficient parallel CFD code,’ said Steve Smith of Fluent Europe. ‘Now you can run sensible jobs overnight and get meaningful output.’
The key to the whole process is parallel hardware – most modern, effective CFD installations use clusters of machines running on Linux. This is popular for two reasons. It is far cheaper than Windows in amulti-processor environment and is relatively stable. When you’re running jobs across 128 processors for eight hours the last thing you need is a blue screen of death when you’re two hours in.
‘We buy and build our own PC arrays from ‘white boxes’ so our hardware solution is relatively inexpensive,’ said Rob Lewis, chief engineer and founder of Advantage CFD, one of the UK’s foremost CFD consultants. ‘The Fluent licence is not linked to a particular processor, so we can swap our processors every 12 months to ensure our performance remains sharp.’
Advantage also uses processors that it can resell easily to recover some of its outlay. ‘I can buy a 24-CPU array from white boxes with gigabit ethernet for about £22,000 and I can probably sell that a year later for maybe £300 a box,’ said Lewis.
‘Our most interesting idea to date is to build PC array supercomputers using laptops. These are fast and reliable and their resale value is much higher because as laptops they’ve hardly been used – the keyboards, for instance, will be in pristine condition. You can buy a laptop now for about £700 with an AMD 64-bit superfast CPU and gigabit ethernet – after a year you can probably get £400 for it.’
CFD is inexpensive when compared to wind tunnel testing. ‘With CFD simulations you get much more data than with tunnels,’ said Smith. ‘You are getting up to eight million points of data and you don’t need to worry about temperature fluctuation in the tunnel.’
The other advantage with CFD is that it can be used to analyse flow trends in areas where it is impossible or undesirable to get a probe in a tunnel. Again, with racing cars it’s very important to know the flow characteristics beneath the cars as they fly very close to the ground and the wind tunnel model is at best a 60 per cent replica.
Consequently, the distance between the model and the rolling road in the tunnel is often too small to insert a probe or pressure tap as the presence of such a device would disturb the flow too much to make the results meaningful – that’s where CFD comes into its own.So not only is the software better than it was, but it also allows you to do things that are difficult or impossible any other way.
But for it to be a success it must be affordable and easy to use which means it has to be automated so that it makes decisions that are correct in the context of the analysis. Essentially what’s going on is that we’re moving out of the lab and into production. This is going to increase the number of users. In order for this to happen the delivery method will change to a ‘pay-per-use’ basis via the web.’
According to Paul Bemis, vice-president of product management at Fluent, the largest potential segment is what he calls the ‘non-consumer’ – the person currently denied access because the financial and technical entry thresholds were so high. The ‘pay-per-use’ model is throwing up some interesting applications.
‘One of the new markets is the prediction of forest fires, said Bemis.
‘Overnight, firefighters run a CFD simulation, which helps them locate where to put the fire breaks and fire barriers for the next morning.’
The simulations provide weather flow data and we run the combustion models, which will predict burn rate, so that we can advise the firefighters, to within a high probability, where the fire will be by the following morning. They use our portal (pay-per-use access) to run the analysis. The models are assembled and run by engineers employed by the fire department rather than CFD specialists.’
Here Lewis sounds a note of caution. ‘CFD still requires specialists, though. the code is certainly getting better and the numbers it produces are more reliable, but you still have to know what’s going on in the background to be able to interpret them properly.
‘At Advantage it takes two years to train a graduate engineer to be a CFD engineer and another two years experience to finish the training. So we look with more than a little scepticism at solutions that introduce CFD into the mainstream.’
‘To some extent this is a little protectionist in that we want to keep our jobs, but we feel that CFD is significantly complex to still require specialist users, maybe not to formulate the problem but certainly to interpret the results.’