applications
➤ However some researchers are taking the bold but also risky step of moving away from traditional airframes with simulation helping them to evaluate new designs. A team of researchers at NASA Langley
Research Centre are employing simulation techniques to develop novel aircraſt designs. Te team used a mixture of Mathworks soſtware tools, Simulink and Matlab, to analyse acoustic data and the Parallel Computing Toolbox to accelerate the processing of this data on Nvidia GPUs. Previously, the engineers were using a legacy
program written in Fortran, which took 20 - 40 minutes to process the approximately 2 gigabytes of data generated during each recording. Mark Walker, principal engineer at
Mathworks, said: ‘One of the real problems today, is just trying to find a Fortran programmer; a lot of what drives the aerospace companies is that they have got these trusted analyse, that may be 30-40 years old, originally developed in Fortran. Te problem is, very few people speak Fortran anymore.’ Walker continued: ‘However because of
Mathworks’ presence in the education sector, there are an awful lot of people, newer graduates in particular, that know exactly how to write the problem in Matlab so that is one of the reasons that you will hear people translating from older Fortran codes into Matlab.’ Te researchers found that once the code
had been implemented into Matlab the same computation could be completed in just 20 minutes. Te addition of GPU computing with Parallel Computing Toolbox cut the time to under a minute. Walker said: ‘We do see some porting of
Fortran into Matlab, it works quite well. Matlab actually grew out of some extensions to some of the things that Matlab could do, so it lends itself to these kinds of applications.’ Walker stated that many of the operations
performed in this example such as FFTs and matrix multiplication are GPU-enabled Matlab functions which lend themselves to parallelisation across GPUs. Another point that Walker stressed was
that the using the Parallel Computing Toolbox takes no additional coding as the application automatically maps the processes across available resources. Once the initial Matlab implementation was completed the team found that it only took 30 minutes to get the algorithm working on the GPUs which got the run time of the application down to one minute – a total of 40 times faster than the original Fortran code. Walker said: ‘As a user I just have to think
right here I want to do an FFT on this data, that is a time-consuming and computationally intensive operation and the parallel computing
34 SCIENTIFIC COMPUTING WORLD
RUAG Space: Antenna bracket simulation driven design process
toolbox will work out the best way to achieve that parallelism behind the scenes.’ Tere are many examples of the work of
soſtware developers who wish to make their soſtware as easy to use as possible while keeping or increasing functionality, a tough challenge as models get more complex and many users call for the integration of different simulations. Mouriaux said: ‘If you really want to benefit
from additive manufacturing then now we need to build up knowledge in all the fields like
ONE OF THE PROBLEMS TODAY IS FINDING A FORTRAN PROGRAMMER
thermal, electronic, and propulsion. If you now want to build using additive manufacturing with integrated systems then you need to have the knowledge of all these things in order to make the best out of it.’ Tis was a thought shared by Robert Yancey
at Altair. He explained that rather than radical change in technology, ‘there is just a lack of good knowledge about how you design for additive manufacturing’. He went on to explain that it would be the convergence of technology and the development of skills that would lead to change in the industry. Yancey said: ‘I think that’s where combining
the additive manufacturing technologies with topology optimisation and then knowledge from companies such as RUAG who have said “Ok, we’re going to make this part out of additive manufacturing; let’s start with a clean sheet of paper and figure out how we’re going to do that”’.
Tis also rings true in acoustic simulation,
Copiello highlighted that MSC soſtware can already integrate different aspects of simulation to deliver a more insightful picture of the real world performance. Copiello said: ‘If you need more complex
acoustics then you need more complex soſtware we have for example Actran VibroAcoustics, which allows you to study acoustic simulation where the both the structure, vibration and the acoustics are solved together.’ Copiello continued: ‘Tis is important because
sometimes you can do first the vibration and then the acoustic and you split the two problems, but sometimes this is not possible. Sometimes you need to have a global overview of what’s going on and always be able to evaluate any change in order to be sure that you will meet the acoustic performance that is required.’ Walker from Mathworks also sees the
integration of different types of simulation as a trend in the aerospace industry. He said: ‘A trend in both civilian and military operations is the number of systems that integrate together is increasing and the coupling between those systems is also increasing.’ As the integration of different types of
simulation and the complexity increases, soſtware developers must abstract some of the unneeded complexity away from the user, delivering increased functionality but not at the cost of making the soſtware more difficult to use. Walker continued: ‘When you do have that level
of complexity is where you will see simulation delivering the most benefit, the number of possible interactions goes up very quickly with the number of times that these systems can connect together, and these can be very complex and subtle.’ l
@scwmagazine l
www.scientific-computing.com
Altair/Airbus
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40