Vanderbilt researchers’ novel technology that integrates machine learning with traditional fluid simulation methods to enhance the accuracy and efficiency of computational fluid dynamics models will be discussed at the prestigious ACM PASC24 Conference in Zurich, Switzerland, June 3-5.
David Hyde, assistant professor of computer science, will be presenting the paper: “Toward Improving Boussinesq Flow Simulations by Learning with Compressible Flow.” Nurshat Mangnike, a second-year Ph.D. student majoring in computer science, is the paper’s first author.
Researchers say the main problem the paper tries to solve is that simulating complex fluid flows can require solving the compressible Euler equations, which are computationally expensive to solve. To mitigate this, they correct the results of more efficient equations – the Boussinesq approximation – using a neural network they trained to learn the difference between Boussinesq flow simulations and compressible flow simulations.
“It turns out that the inference time for this neural network is negligible,” says Hyde. “So, we get the best of both worlds: a computationally efficient fluid simulation algorithm that is almost as accurate as a traditional compressible flow solver. We only trained our network and tested our method on a small class of problems, but we believe that this methodology could apply to various kinds of fluid simulation challenges.”
Mangnike says he considers it an honor to have the paper presented at the conference.
“The opportunity to have our work featured in such a respected forum validates our innovative approach,” he says. “This accolade is not only a motivator for us to continue our efforts but also an opportunity to share our insights with the global scientific community, fostering further research and collaboration.”
The PASC Conference series is an international and interdisciplinary platform for the exchange of knowledge in scientific computing and computational science with a strong focus on methods, tools, algorithms, workflows, application challenges, and novel techniques in the context of scientific usage of high-performance computing.
Hyde, who will be making the presentation, received a competitive travel grant award sponsored by ACM SIGHPC.
Contact: Lucas Johnson, lucas.l.johnson@vanderbilt.edu