During the process of designing an aircraft, aeronautical engineers must perform predictive simulations to understand how airflow around the plane impacts flight characteristics. However, modeling the complexities and subtleties of air movement is no easy task. In addition to understanding “ideal” airflow scenarios, engineers need detailed insights regarding turbulence and vortices to understand how they interact with an aircraft in flight. Kenneth Jansen, Professor of Aerospace Engineering at the University of Colorado Boulder, seeks to improve the process through his work in the field of computational fluid dynamics. Where existing predictive models are insufficient, Jansen and his research step in.
For several years, Jansen has tapped the supercomputing resources at the Argonne Leadership Computing Facility (ALCF) to improve computational modeling capabilities to provide deeper insight into the problems posed by fluid flow and how their resolution can lead to refined aircraft design. To prepare for Argonne’s future exascale system, Aurora, Jansen is currently leading two ALCF Early Science Program projects focused on advancing simulation, data analytics, and machine learning methods to enable flow simulations of unprecedented scale and complexity.
Jansen’s specialized work involves developing “scale-resolving simulations” to obtain a more detailed analysis of airflow characteristics. His models augment traditional simulation methods by evaluating unsteady, turbulent motions using high-performance computing. Said Jansen, “This approach allows us to resolve the turbulent-scale dynamics to get a much better overall prediction than if we modeled everything at once.” From there, Jansen and his team employ adaptive methods for prediction. When doing any simulation, said Jansen, “We learn where our predictions are right and where they are not as effective. Those predictions that need improvement undergo adaptive methods to hone and refine the simulation for greater accuracy.”
“We call the air that surrounds the airplane a fluid volume. That envelope is exceedingly difficult to analyze holistically, so we break it down into what we call cells. The size of these cells dictates how much of the turbulence detail we can resolve. By adapting the overall mesh of individual cells, we can make the mesh finer in regions where more detail about airflow is needed.”
Safer, more efficient aircraft
Jansen offers an anecdote to describe the nature of the work and reasoning behind it. “In addition to turbulent airflow, we also seek predictions about other things. For instance, how much lift is generated by airplane wings at a certain speed? Simple models describing a typical flight can accomplish this straightforward task relatively easily. However, the models change dramatically in a scenario like an engine failure on a two-engine plane. To fly the plane straight ahead, the pilot must move the rudder to one side to account for the lack of thrust from the failed engine. In aircraft designs, many have rudders sized about 25 percent larger than necessary to handle that type of situation. However, the increased drag caused by oversized rudders means heavier fuel consumption. Smaller rudders alone could save $300 million a year in fuel costs.”
Aerospace is a very competitive, economically-sensitive market. Those purchasing aircraft seek planes which have long ranges and better fuel economy to make flights more profitable. Jansen’s work simulating airflow helps address these needs by suggesting airframe optimizations which can reduce operational costs of each plane as well as its carbon footprint.
Exascale Computing
“Exascale systems will enable new possibilities in our work,” he noted. “First, its computing prowess can resolve more complex turbulent scales, so we can provide engineers a better predictive capacity for complicated flow conditions like when a rudder is compensating for a failed engine. Secondly, exascale computing empowers us to do many lower-fidelity calculations quickly. This process is especially important when we consider things like wing thickness, where to place flow control devices, and more. By doing thousands of these smaller-scale simulations, we can more efficiently impact an aircraft design in positive ways.”
Partners in flight
“In some sense, we blaze a new trail with this research because we can work closely with aircraft designers – and highly advanced compute systems – to help them accomplish work the aircraft industry may not be able to accomplish on its own for many years. Our discoveries can impact new designs today,” Jansen said. He and his colleagues interface with aircraft companies at multiple levels. They work directly with design engineers to increase the accuracy of their simulations, to improve current aircraft designs, and help them plan next-generation airframes. While most major manufacturers have an internal ‘think tank’ group that does research paralleling Jansen’s, the collaborative effort also helps mine deeper for all possible ways to tweak current designs. Together they pursue augmented simulations to assist both today’s and tomorrow’s endeavors.
Advanced simulations using Aurora
Exascale computing[*] facilities, like the forthcoming Aurora system at Argonne National Laboratory, will open the doors to new opportunities in this arena.
Argonne anticipates delivery of Aurora in 2021. Once online, the system will have the capability to perform billion-billion calculations per second. Built by Cray, Aurora’s performance will derive from advanced hardware including the future generations of Intel Xeon processors, Intel Optane DC Persistent Memory, Intel Xe technologies, and more. Commented Jansen, “Aurora would not be possible without the support of companies like Cray and Intel. Aurora will advance many scientific projects, including my own. With a tool that powerful, my team has new opportunities to make meaningful contributions to aircraft manufacturing and the environment too.”
Before high-performance computing (HPC) existed, wind tunnels provided the most accurate data for airframe simulations on a more massive scale. More recently though, Argonne’s Theta supercomputer, Aurora’s petascale predecessor, supported Jansen’s simulations of aircraft flight characteristics. Even with Theta though, barriers in computing speed constrained the resulting simulations. Models simulated an aircraft at one-nineteenth its actual size, flying at a quarter of its real-world velocity. In contrast, said Jansen, “Aurora will help us learn more about the fundamental physics of flow control in a full-sized, full-speed aircraft simulation. From there we can identify where big or small design improvements can make an important difference in flight characteristics.”
Even with exascale systems supporting his work, Jansen recognizes the magnitude of the work ahead, “We want to make the best use of Aurora’s resources, so we must ensure our computational methods are both efficient and effective. Making the best use of hardware means we need to re-shape data structures and algorithms, plus we must develop more accurate numerical methods.”
Overcoming turbulence
“As any airline passenger knows, air turbulence can vary greatly throughout a flight. Sometimes you barely notice it, and other times, well, it’s quite bumpy,” he chuckled. The seemingly infinite variability of turbulence makes it very difficult to simulate an entire aircraft’s interaction with it. At any given second, different parts of a plane experience different impacts from the airflow. Even an exascale computer cannot keep up with storing the enormous volume of data necessary for the job. Added Jansen, “We need to get data insights without the need to write all that information to file. That means we must do co-processing of data real-time as the simulation progresses. We call that process in situ data analytics.” Jansen elaborated, “in situ lets us examine visualizations over time increments, allowing us to see airflow dynamics without writing to file.”
“I’m excited about using Aurora for the first time and performing exascale-level simulations. It will put us at the forefront of predicting and understanding fluid flow around complicated things like airplanes.” Continuing, Jansen added, “We finally have the compute performance to simulate complex airframe components like a full vertical tail and rudder assembly and do it at full scale. That feat has not been accomplished before.”
Rob Johnson spent much of his professional career consulting for a Fortune 25 technology company. Currently, Rob owns Fine Tuning, LLC, a strategic marketing and communications consulting company based in Portland, Oregon. As a technology, audio, and gadget enthusiast his entire life, Rob also writes for TONEAudio Magazine, reviewing high-end home audio equipment.
[*] Editor’s note: Aurora disclosures made in March cited a performance goal of sustained exaflop/s.