Author(s): Elia Merzari; Noah Halford; Paul Fischer; Andrew Siegel
Linked Author(s):
Keywords: Transients; Computational Fluid Dynamics; Surrogate modeling; Proper Orthogonal Decmposition
Abstract: While continued advances in supercomputing are enabling the simulation of flows of increasing size and complexity, the presence of multiple and strongly separated timescales limits their effectiveness in addressing real-world questions of great practical value. Recent results demonstrated that, in order to simulate all relevant time scales in long transients, several years of continuous simulations would be necessary even on the most advanced supercomputers. In fact, when simulating time-dependent multiphysics phenomena, the overall integration time is a function of the slowest time scale: larger problems often imply longer integration times, thus reducing the effectiveness of such machines in simulating transients. This work aims to develop a new time-scale decoupling algorithm, based on the use of a surrogate solution. The physics related to the fastest time scale, which is also responsible for the highest computation cost, is decoupled or partially decoupled. Possible surrogates involve replacing the fastest scale with a reduced-order model (e. g. ,PODProper Orthogonal Decomposition) .The methods are implemented in the DNS/LES code Nek5000 and demonstrated on the solution of the advection-diffusion equation for the temperature where buoyancy is not present. Example cases include the 2D flow past a cylinder and 3D channel flow (for which partial results are available) .The POD methodology has validated and tested on the three-dimensional flow in T-junctions for which Particle Image Velocimetry (PIV) data is available.
Year: 2015