When industrial designers plan the design of a new element of a vehicle’s propulsion or exterior, they typically use fluid dynamics to optimize airflow and increase the vehicle’s speed and efficiency. These fluid dynamics simulations, however, are computationally expensive, meaning that researchers typically simulate small parts of the system or a dramatically simplified version of it in order to approximate real-world results. Now, a team at CERFACS (a private laboratory in Toulouse, France) has applied supercomputing resources from the Partnership for Advanced Computing in Europe (PRACE) to complete the first full, high-fidelity simulation of an entire aircraft engine.
The project is named FULLEST, for “First fUlL engine computation with Large Eddy SimulaTion Resources.” Through FULLEST, the research team – led by Carlos Pérez Arroyo and Jérôme Dombard – simulated three key components of the DGEN380 “turbofan” business jet engine: the engine’s fan, its high-pressure compressor and its combustion chamber. During the course of the project, CERFACS worked with PRACE, Safran SA (an airplane and helicopter engine manufacturer) and Akira Technologies (the aircraft engine company that built the engine model being simulated).
The team began with individual component simulations, developed a methodology for coupling the components and, eventually, completed their first multi-component simulation. This also necessitated a 360-degree simulation, as opposed to conventional simulations, which often duplicate components to reduce the computational costs. “In fact, many simulations of single components compute just one of these repetitive sections,” Dombard said in an interview with PRACE.
The initial results demonstrate that this robust multi-component engine simulation is feasible – and beneficial. Running a full-engine simulation allowed the researchers to view the interactions between components, such as the effects of the pressure wave from the compressor on the combustion chamber. “When you couple different engine components, simulations become a lot more extensive and complex,” Dombard said. To wit, the simulation ended up representing the engine with a staggering two billion cells.
To complete the simulation, the project (which is ongoing) was awarded a total of 31.6 million core hours through PRACE. The hours were awarded on the Joliot-Curie system, which is hosted by GENCI (France’s national high-performance computing organization). Joliot-Curie hosts four partitions (an Intel Skylake partition, an Intel Knights Landing partition, an AMD Epyc Rome partition and an Intel Cascade Lake partition), with the most powerful partition – the AMD system – delivering nearly 7 Linpack petaflops and placing 33rd on the most recent Top500 list of the world’s most powerful publicly ranked supercomputers.