CO2 emissions from the transport sector are increasing, and at this point there are few signs of this trend reversing. Torbjörn Larsson of Creo Dynamics has been running a PRACE project that aimed to iron out some of the issues involved in running HPC simulations that can help improve the design of large vehicles and therefore reduce emissions.
The transport sector contributes around 25% of total CO2 emissions in the EU, and is the only sector where the trend is still increasing. It is therefore of paramount importance to increase the efficiency of freight transport in order to look towards decreasing this trend. The use of numerical methods for time-dependent flow simulations is becoming increasingly important for highly accurate predictions of aerodynamic drag, aero-acoustic sources, vehicle dynamics and handling. This type of work can help to improve the performance of vehicles and reduce the amount of CO2 being emitted by the transport sector.
Torbjörn Larsson of Creo Dynamic has a distinguished career in the field of computational fluid dynamics. Starting in the early 90s at Saab Aerospace in Sweden, developing software for aerodynamic design and analysis, he then moved to the USA where he worked on aerodynamic prediction for road cars in Detroit. He then moved back to Europe where he spent ten years in Switzerland working with the Sauber Formula 1 team developing their nascent CFD team from scratch. This was followed by a similar role at the Ferrari team in Italy.
In 2013, Larsson moved back to Sweden and started work with his current employer, Creo Dynamics. An offshoot of the Saab Aeronautics and Acoustics Division, Creo Dynamics has since grown to incorporate fluid mechanics into its core expertise to become a specialist at aeroacoustics, i.e. modelling fluid flow with noise. Larsson’s role there is as a senior expert of the fluid mechanics and methodologies, applying these techniques to large and complex applications related to the automotive and aerospace industries.
Creo Dynamics recently undertook a PRACE SHAPE project together with application experts from the KTH Royal Institute of Technology. During the course of the project they developed templates and “recipes” for a range of tasks that are involved in the automotive industry – such as automated handling of computer-aided design (CAD), parallel meshing, solving and post processing. These were all tailored towards a particular automotive application, namely the aerodynamics of a heavy-duty semi-trailer.
The project also focused on parallel implementations and executions for large-scale simulations. In addition to monitoring the efficiency of the processes (for meshing and solving run-time performance and scalability) and identifying critical bottlenecks, they gave significant attention to pinpointing performance deficits in the processes so as to give guidance for a further fine-tuning of the overall methodology.
The truck is the generic model used in the SHAPE project. Colours represent static surface pressure (red – high pressure, blue – low pressure). The grey, semitransparent contours, show iso-contours of total pressure. This is to illustrate offsurface flow structures such as vortices and large viscous losses. The truck is here exposed to some amount of side wind conditions (probably 5 deg).
“The models for this study are large and very complex,” says Larsson. “What we wanted to do in the end is not only compute the airflow around the vehicle, but also the acoustic noise that is added into the mix from the mirrors and the wheels and other parts of the vehicle. To do this, you first need a very time-accurate flow simulation, which requires a lot of compute resources.”
The company set out to do this work from the point of a computeraided design (CAD) model all the way to the final data results in post-processing. This kind of work has to be run on high-performance computers in a distributed mode on many cores and many CPUs. “One of the main aims of the PRACE project was to look at the entire process, see where the bottlenecks are and see if we can make it more efficient,” explains Larsson. “With smaller problems, it is quite often the case that everything works smoothly without any problems. However, when models become larger and more unwieldy, all sorts of strange problems that you never envisage start to happen. They could be related to I/O, memory or many other things.”
One of the main bottlenecks often comes at the creation of the computational mesh, as certain parts of this process do not work well in parallel computing modes and uses a lot of memory. “Overall, we were using our varied expertise to debug this whole process from start to finish and looking for ways to improve it at points like this where memory usage was becoming very high. The PRACE project allowed us to monitor this in a way that we normally would not be able to.”
When using a complex CAD model, the ideal situation is to remove the need for any human intervention in the process. “We want the computer-aided side of things to run in a way that we do not have to keep making interventions to make it run smoothly, otherwise it partially defeats the point of it,” says Larsson. “As far as possible, this process should be able to run independently.”
Although the project was not one of the larger ones that Creo Dynamics has been involved in, Larsson views it as a great success as they were able to pinpoint many of the places where they were starting to lose performance with their techniques and where scalablity was dropping off. “This project has allowed us to solve some of these problems by finding unique ways to work around them. In this sense, we have learnt a lot through this work.
“The opportunity to work with application experts at the data centres really helped to guide us in a sense that without them we could have spent huge amounts of time blindly using trial and error to move forward. They were able to look at our runs and provide us with diagnostics as we were doing them, examining our memory usage in a way that as a normal end user you would never be able to.”
Although still a small company that employs around 25 people, Creo Dynamics was recently acquired by Faurecia, one of the world’s leading automotive technology companies. They are now involved in a large Horizon 2020 project called Aeroflex, using many of the same processes that were used and refined in the PRACE project. The aim of Aeroflex is to help create more energy efficient road transportation for the future, and many of the largest truck manufacturers in Europe are involved. Creo Dynamics is responsible for many of the computational fluid dynamics simulations in the project.
Project title: Large scale aero-acoustics applications using open source CFD
The resources awarded were:
This project was awarded 100 000 core hours on MareNostrum hosted by BSC, Spain, and over 50 000 core hours on Beskow hosted by KTH, Sweden.
Research field: Computational fluid dynamics
- For more information see here.