By Ed Pitkin
There is always some degree of uncertainty when moving from engineering simulations to physical prototypes. The real world has more variables and finer granularity than our digital models. Yet recent advances in hardware and software have significantly reduced the gap, enabling engineers to build and run digital simulations faster and extract deeper and more reliable insights into the performance of complex physical systems.
There are few industries where fast, predictive capability is more critical to success than professional motorsports. Race cars are complex, tracks are all different, and the racing season is long and arduous. Yet the difference between winning and losing a 500-mile race can be a matter of inches. Even minor design optimizations can make the difference between defeat and victory.
Roush Yates Engines has risen fast in this uber-competitive world. Since 2004, cars sporting Roush Yates Engines have won more than 280 races in some of the world’s toughest competitions, averaging more than 20 wins per year across NASCAR, IMSA and FIA Series. Speed on the track is only part of the story. The company’s engineering team designs and builds more than 1,000 race engines per year and competes in more than 128 events. The engineering team has to deliver world-class innovation at high volume and with fast turnaround times. Efficient digital simulations that accurately predict real-world performance are key to their success.
Deep insight into engine dynamics
The engineering team uses digital simulations throughout the design cycle. According to Todd English, Vice President of Business Development, “We scrutinize and analyze every component and optimize every engine for a specific race on a specific track. Since we can evaluate most components in a virtual environment quicker than we can manufacture and test them, simulation is one of our most important tools.”
Simulations take time and consume resources, so careful integration into design flows is important. The design team develops and evaluates digital models for key components and sub-systems separately. These parts are then combined into full engine models to examine how they perform together, including how they perform in the vehicle and on the track under realistic race conditions. As ideas solidify, the team transitions gradually toward larger, higher-resolution and more complex simulations to fine-tune their designs.
Software advances for predictive accuracy
Digital models illuminate engine dynamics in ways that are difficult or virtually impossible to explore on a firing engine. Yet the behavior of the models must correlate closely with physical reality to enable effective design decisions. The team at Roush Yates Engines use CONVERGE CFD, which provides a range of modeling options for internal combustion engines. A number of features in the application help to improve predictive accuracy.
- Automatic mesh generation provides an optimized numerical model based on design characteristics and simulation requirements. This eliminates the time, effort, and errors associated with manual approaches.
- Dynamic re-meshing allows the number and size of the cells in the mesh to be optimized automatically at each time step of a simulation. Resolution and accuracy can be increased in the areas of greatest interest, without driving up overall simulation run times.
- Fully-coupled chemistry allows for greater accuracy and efficiency when analyzing how physical and chemical interactions affect each other and impact power delivery.
Together, these capabilities help the engineering team move from design, to simulation, to results more quickly. They also help them shine a brighter light onto the most critical phenomena, such as fuel flow, aeration, dispersion, and ignition.
Hardware advances for speed and efficiency
Dynamic re-meshing and fully-coupled chemistry add to simulation workloads, potentially extending runtimes beyond acceptable limits. Says English, “When developing a component, we might have five designs that we want to evaluate. If each design case takes 10 hours to solve, it could take more than a week to run them all.”
To support increasing workloads without longer runtimes, Roush Yates Engines is introducing more parallelism into its computing platforms using Intel Xeon Phi processors. These processors provide up to 72 cores per socket and four threads per core (more than three times the number of cores and six times the number of threads of current-generation Intel Xeon processors).
Several additional features help to optimize parallel throughput, including support for extra-wide vectors (up to 512-bits) and high-speed, on-die memory. Integrated fabric controllers are another option to support low-latency connections to a high-speed cluster fabric based on Intel Omni-Path Architecture. (Intel Xeon Phi processors and Intel OPA are both part of the Intel Scalable System Framework, which is designed to help simplify the design of balanced, high-performing clusters.)
Firing on all cylinders
Although increasing compute parallelism can be an effective strategy for many HPC applications, it is not appropriate for all workloads. Importantly, the chemistry and flow solvers in CONVERGE CFD parallelize independently, so they are well suited to a highly parallel execution environment. Convergent Science worked with Intel to optimize their code for the new platform. Optimizations for Intel Xeon Phi processors tend to improve performance for Intel Xeon processors as well, so the benefits extend across multiple compute platforms. For Roush Yates Engines, this means that any simulation can be run efficiently on the fastest available computing resource.
Taking it to the track
Roush Yates Engines’ work doesn’t end when an engine is built. The company provides track-side support to help customers get the best performance in the only place where it really matters. Drivers and pit crews make split-second adjustments throughout a race to optimize traction, fuel efficiency, power delivery, and more. Any one of these decisions can mean the difference between success and failure. According to English, “We benefit from these engagements as much as our customers do. We get to monitor engine operations before, after, and in some cases during each race. We use that data to compare real-world performance with our modeling, so we can continue to improve the predictive value of our simulations.”
Accelerating into the next era of reliable speed
Roush Yates Engines isn’t resting on its many laurels. Every engine design must be better than the last, which means taking advantage of technology advances at every level. From materials and physical tools, to simulation software and computing platforms, new innovations must be integrated into the design flow early enough to deliver competitive advantage yet late enough to avoid the costs and risks that are often absorbed by the earliest adopters.
It’s not an easy balance to maintain. However, as the industry continues the march toward fully-predictive simulations, Todd English believes it’s imperative to stay near the front of the pack. “Roush Yates builds world-class engines, which requires world-class computing and design tools. We work with the best to deliver the best. And we have no intention of slowing down.”
Roush Yates Engines
roushyates.com
Intel
www.intel.com
Converge CFD
convergecfd.com