Andrew Duggleby, Ph.D

Using computers to go where experiments cannot: massively-parallel LES of turbulent heat transfer . by Andrew Duggleby, Ph.D, Presentation

For decades, the steady increase in modern computing technology has allowed for faster as well as larger, more complex simulations. With all this computing power, there are still only two areas in which a numerical simulation is better than an experimental data: (1) quick, reliable (10-20% error) simulations for design optimization, and (2) massive resolution, highly accurate (< 1% error) simulations. In both cases the computer is going where experiments cannot. In quick simulation case, the simulations are faster and cheaper than any experiment, yielding results (hopefully trustworthy results) fast enough to be included in a design cycle. In the highly-resolved case, the resolution is far beyond any experimental measurements - in the context of turbulent heat transfer, the entire velocity, pressure, and temperature elds are known everywhere at all times.

In this talk, the current state-of-the-art for both the quick simulations and the highly-resolved simulations will be discussed in the context of turbulent heat transfer. For the quick simulations, recent advances in not just simulation time, but total CAD to analysis time will be discussed: (a) Computer-Aided Design (CAD) model to mesh time, (b) simulation time, (c) accuracy vs time trade-os (models, resolution, etc), (d) analysis time. For computational fluid dynamics and heat transfer, this almost always refers to the steady-state Reynolds-Averaged Navier-Stokes (RANS) models, but an example will be given for a time-dependent Large Eddy Simulation of a venturi nozzle where CAD to analysis was done in under 24 hours. For the highly-resolved simulations, analysis techniques to elucidate useful information out of terabytes of data are discussed, with an example of pin-n heat transfer direct numerical simulation (DNS) where the modes responsible for heat transfer are extracted via Proper Orthogonal De-composition (POD), and then enhanced by endwall contouring resulting in increased convection with minimal drag increase.