SCCS Colloquium - Feb 19, 2020

From Sccswiki
Jump to navigation Jump to search
Date: February 19, 2020
Room: MI 01.09.014
Time: 15:00 - 16:00

Qunsheng Huang: Loose-Coupling of CFD/CSD Rotorcraft Simulations via preCICE

Master's thesis submission talk. Qunsheng is advised by Gerasimos Chourdakis, in collaboration with the Institute of Helicopter Technology (Amine Abdelmoula).

Accurate rotor blade analysis is a complex multiphysics problem. The current paradigm employs multiple single-physics solvers to solve the multi-physics problem. This work develops preCICE adapters for TAU and CAMRAD II. These modular adapters minimize code dependencies between solvers and allow for the easy removal or addition of data processing steps. Additional methods were implemented to support the loosely-coupled rotorcraft simulations. Similarly, an intermediate interpolation step was implemented to allow for the 3D-1D passing of data from TAU to CAMRAD II. Three cases were implemented using the adapter code. A simple, tightly- coupled simulation was used when designing the TAU adapter. The second case was a loosely-coupled rotor blade simulation with only rigid body motion. This simulation was extended to introduce elastic deformation in addition to rigid body motion to create a third case.

Keywords: preCICE, loose-coupling, rotorcraft simulation, 1D-3D mapping

Jan Nguyen: AutoTuning using Bayesian Statistics in AutoPas

IDP submission talk. Jan is advised by Fabio Gratl

In many cases, a program can have many configuration options. The choice can have a large impact on the performance, but may not always be trivial for the layman. It is possible to recommend options that are efficient in most cases. However, if the individual use case leads to significant differences in the optimal choice, automation is preferable. We have analyzed how Bayesian statistics can be applied here. Such an algorithm uses a probabilistic model to generate a good configuration by observing some test runs. For the case of molecular dynamics simulations, we implemented this idea into the C++ library AutoPas. This achieved on average significantly better results than brute force and purely random methods.

Keywords: Molecular Dynamics, AutoPas, AutoTuning