A team from Argonne National Laboratory has set out to clean up the combustion engine, using one of the largest computational fluid dynamics (CFD) studies ever undertaken. Running the CONVERGE CFD code from Convergent Science, researchers at Argonne plan to carry out around 10,000 simulations concurrently as part of an investigation into gasoline compression ignition (GCI).
“Improving engine efficiencies by even a few percentage points can take a big chunk out of our carbon footprint,” commented Sibendu Som, technical lead investigator at Argonne’s Virtual Engine Research Institute and Fuels Initiative (VERIFI). “We are working on a proof-of-concept to demonstrate how large studies like this can really help engineers zero in on the optimum engine designs and operating strategies to maximize efficiency while minimizing harmful emissions.”
It’s estimated that the study will require around 60 million processor hours on the laboratory’s IBM Blue Gene/Q supercomputer, known as Mira. At 10 petaflops, this machine is one of the fastest supercomputers on earth, capable of around 10 quadrillion calculations per second.
The sheer numbers involved require a huge amount of computing power, but the purpose of the study is actually to demonstrate how efficiently CFD can be used to evaluate new concepts. It relies on several clever pieces of technology to speed things up.
CONVERGE was developed with combustion simulation firmly in mind and it uses a number of techniques to improve the speed-to-accuracy trade-off normally associated with CFD. Chief among these improvements is Adaptive Mesh Refinement, which constantly regenerates the mesh used for the simulation throughout its runtime. What that means is that areas of specific interest can be captured in improved detail, while reducing the computing time spent on less critical parts of the model.
Engineers from Argonne’s computer science division have also developed a technique for spreading out the computational load across Mira’s mammoth bank of 786,432 processors. Only a fraction of those cores (around 4,000) are actually used for the experiment, but the researchers say this so-called stiffness-based algorithm has improved processing times by a factor of more than three under certain conditions.
Each of the 10,000 or so simulations will investigate a different potential change to a GCI research engine based on a 1.9-liter diesel engine produced by General Motors. Modifying parameters like piston bowl geometry, injection timing, and fuel composition simultaneously helps to pinpoint potential breakthroughs.
“Until recently we couldn’t have run this many simulations in one go, because the computing resources weren’t there and neither was the technology. In this project we have access to both,” explained Som. It’s hoped that the study will lead to specific advances in GCI development, but it also highlights the wider benefit of using large scale CFD studies like this for engine optimization. Thanks to smarter software and ever increasing computing power that’s becoming easier as time goes on.