Advanced Computing in the Age of AI | Friday, March 29, 2024

Designing Bicycle Wheels in the Cloud 

While riding a bike may be easy, the aerodynamics involved in wheel design are complex, requiring powerful CFD software running on high-performance computer systems. The simulations generate so much data that smaller manufacturers who could benefit from the technology get overwhelmed or go without. In response, several companies, including Intelligent Light and R-Systems, are offering a combination of solutions aimed at making CFD more accessible.

Sometimes apparently simple things have an underlying complexity that is astounding. Take a bicycle wheel for example. It's round, has a rim, a tire and some spokes and turns in a circle. Simple, right?

Wrong.
bicycle wheel aerodynamicsThe aerodynamics of a bicycle wheel are so complex that powerful CFD (computational fluid dynamics) software running on high performance computer (HPC) systems are required to generate the simulations needed to evaluate different design approaches.

Roger Rintala, strategic marketing manager for Intelligent Light"Bike wheels involve some really tricky physics," says Roger Rintala, strategic marketing manager for Intelligent Light. The company, headquartered in Rutherford, NJ, is the maker of FieldView CFD, sophisticated post-processing CFD software.

For the makers of bicycles and their components, as well as other small- to medium-sized manufacturers (SMMs), the growing sophistication of CFD and FEA solvers is a double-edged sword. On one hand, researchers are able to run multi-physics simulations in much finer detail and investigate unsteady or transient problems that were previously inaccessible. On the other hand, the simulations generate so much data that smaller manufacturers, who do not have a squadron of CFD specialists on staff, are either overwhelmed or simply do without the technology.

"The challenge for CFD in particular is that the volume of data is really big on the results side," says Rintala. He notes that processing the data needed for the simulations is not a problem — there are many excellent solvers on the market. In addition, advances in private and public cloud computing means that companies can run their analyses on a pay-as-you-go basis and not invest in expensive infrastructure. It also makes HPC capabilities more available, which is particularly important due to the growth of time-dependent simulations.

However, compared to finite element analysis (FEA) computations, CFD is an order of magnitude more complicated. If you want to do a structural analysis of pressures on soda can, even if there is some flexion involved, you are still working with surfaces. You don't really care what's happening three inches around from the can.

But when you are dealing with fluid flows, the picture changes radically. For example, consider the bike wheel. The CFD simulation not only needs to capture volumes inside and outside the wheel, but also boundary layer effects, vortices and other unsteady effects both on the wheel and in the space surrounding it. And it has to do this dynamically over time. The solver can't do this; this is a job for post processing software like Intelligent Light's FieldView.

Wheels Within Wheels

Matthew N. Godo, FieldView product manager at Intelligent LightDr. Matthew N. Godo, FieldView product manager, has been leading a four-year study of bicycle wheel aerodynamics that recently has included interactions with the entire bicycle frame. The fact that he is a triathelete and an avid cyclist has lent impetus to his research.

The aerodynamic flow around a rotating bicycle wheel, including its interaction with the front fork and frame components, is incredibly complex. For example, in one recent study, Godo investigated multiple wheels and fork/frame combinations at 10 different yaw angles. About 3.6 gigabytes of data were generated during the steady simulation; the unsteady simulation threw off nearly 1.2 terabytes of data. The most recent work includes high resolution full bike simulations that are generating exponential data growth.

For a recent full bike frame simulation, Godo used a commercial solver — in this case, STAR-CCM+ — and high performance compute capabilities from R Systems, in Champaign, Ill.

R Systems provides technical expertise and optimized HPC cluster resources to the commercial and academic research communities. The company provides fast and flexible HPC implementation, leaving users free to focus on their core research objectives. R Systems takes an agnostic approach to OS, hardware, and software to ensure that its custom solutions will match researcher requirements and expectations. Supported by expert technical staff, users avoid tedious systems management, allowing them to be more productive while saving time and reducing cost.

"We have the software and support already in place to service clients such as Intelligent Light," comments Brian Kucic, business principal/co-founder of R Systems. "This particular project began late on a Friday evening and ran through the weekend. Because we already had the appropriate software licenses installed and were using machines with substantial memory and high speed interconnect, we were able to respond very quickly to Intelligent Light's requirements."

full bicycle simulation
Full bike simulation depicting cross flow speed using modified Q-criterion (a way of measuring fluid flow characteristics).

"Matt had limited experience working with remote HPC services such as that offered by R Systems," Rintala recalls. "That didn't stop him for a second. He uploaded a relatively small input file and ran a simulation that included 8.5 gigabytes per time step running in parallel on 32 cores to create the raw data needed by the FieldView post-processor. By breaking the job up into segments, he was able to begin concurrent batch processing — post processing on eight of the cores — while the other time steps were still running."

Among the post-processing tasks handled by the post-processor were:

  • Resolving the forces on the wheel into drag, side (lift) and vertical components.

  • Determining which resolved forces act on each component including the wheel rim and tire, hub, spokes, and fork.

  • Calculating resolved forces along the wheel circumference.

  • Creating custom visualization of flow structure for standardized quantitative and qualitative evaluation.

  • Supporting future development by building on existing tools.

One of the bicycle manufacturing companies that has expressed keen interest in the Intelligent Light CFD studies is Zipp Wheels, an Indianapolis, Ind., firm that makes high performance wheels for bicycle and wheelchair competitions.

Blowing in the Wind

Zipp has extensive experience using wind tunnels to develop their prototypes. They are also aware that not only is wind tunnel testing expensive, but also it is limited in its ability to fully simulate the conditions that a wheel will be subjected to. Typically wind tunnel tests are conducted without the wheel rotating or, if the wheel is turning, it is not moving across the ground plane.

"This was the first time that Zipp was exposed to CFD studies that dealt with the physics they understood," says Rintala. "It showed them interactions that they suspected were occurring but could not capture in a wind tunnel. The Zipp engineers are very good at evaluating different wheel designs in terms of drag, but Matt's study opened up new avenues for exploration. For example, recent work has resulted in new product designs."

Aerodynamic torque is one of those avenues. FieldView allowed Godo to simulate actions that were impossible to see in wind tunnel tests such as vortex shedding (the unsteady flow generated when a fluid moves past an obstruction), and center of pressure. Godo discovered behaviors in which the vortex would hang on the wheel, creating drag, and then shear off in regular patterns (see image above). He determined that the design of the bicycle wheel rim, not the spokes or hub, made the most difference in minimizing drag.

Rintala says Zipp Wheels represented the classic dilemma facing many SMMs. Although they understand how to conduct physical prototyping for test and analysis, they are missing out on the many benefits conferred by using advanced digital manufacturing techniques such as CFD and FEA running on advanced computer systems. It would be a huge leap of faith, he says, for these firms to make a major outlay for HPC hardware and software and develop the in-house expertise to make full use of these systems.

"Cloud computing presents a great opportunity for the SMMs," he says. "Instead of building their own infrastructure, they can tap into the cloud and use commercially-available software and expert support when they need to run HPC scenarios. And one of the major attributes of on-demand cloud computing is that it is especially suited for batch processing."

When faced with a deluge of data from a typical CFD solver run (for example, a typical analysis of bike wheels might include 120 time steps, six wheel designs, and three different yaw configurations), researchers can execute their scripts in batch mode. They not only get the answers they need, but build a base of intellectual property and standardized routines that will allow them to move ahead with their research.

Scripts can be rerun and old data decks reused while new data and physics are added to the mix. This allows researchers to make comparisons with previous results and move ahead toward new discoveries. For example, recently the same FieldView routine was run 60 times to calculate the drag force for all wheel and fork combinations at two speeds. Just one of the tasks undertaken — calculating circumferential variation, which required 3,600 calculations for each wheel and fork combination — would not have been possible without the capabilities inherent in the Intelligent Light software and batch processing in the cloud on the R Systems servers.

Lightening Up the Cloud

One of the problems with the cloud scenario is that the massive amounts of data generated by the CFD solver resides on the cloud servers and is not easily accessible by a remote client over the Internet.

Intelligent Light has addressed that problem by introducing CFD data management tools enabled by lightweight XDB files within FieldView that significantly reduce data read-in times and storage requirements and speed up batch runs. Rintala says the XDB creates file sizes that can be 45x smaller and allow users to engage in fully interactive post-processing from their desktop systems in client/server mode. Read-in times are dramatically shortened and calculations are far faster. Since FieldView treats the XDB files just like any other dataset, run times that once took weeks now take days, and images and animations that have been generated in batch mode in the cloud can be quickly downloaded to the user's workstation or personal computer for further post-processing.

Users have the option of running multiple instances of FieldView concurrently in batch mode at a fraction of the standard license by purchasing FieldView Batch Packs, sold in packs of 5, 10, 20 and 200. Discounts can range as high as 90 percent. For SMMs this means they have the opportunity to run multiple cases and multiple time steps, create images and animation frames, and allow multiple users to post-process their cases on the server.

CFD for the Masses

In short, the combination of on-demand cloud services, cost effective batch processing, commercially-available solvers, and post-processing software like FieldView, brings the benefits of CFD and high performance computing within the reach of smaller firms.

Says Rintala, "We learned a lot over the past four years working on the bicycle wheel project. For one thing, although complex, the whole process has been remarkably easy. There were no cloud issues — the entire workflow both in and out of the cloud was built around a coherent data management approach. We had all the capacity we needed when we needed it. Concurrent batch processing using a cost-effective licensing model combined with interactive sessions made possible by the XDB-based workflow means that this technology is now within reach of mid-market manufacturers who can benefit from easy to use, affordable high performance computing."

EnterpriseAI