Parallel optimization for large eddy simulations

Abstract

We developed a parallel Bayesian optimization algorithm for large eddy simulations. These simulations challenge optimization methods because they take hours or days to compute, and their objective function contains noise as turbulent statistics that are averaged over a finite time. Surrogate based optimization methods, including Bayesian optimization, have shown promise for noisy and expensive objective functions. Here we adapt Bayesian optimization to minimize drag in a turbulent channel flow and to design the trailing edge of a turbine blade to reduce turbulent heat transfer and pressure loss. Our optimization simultaneously runs several simulations, each parallelized to thousands of cores, in order to utilize additional concurrency offered by today's supercomputers.