Title: Accelerating CFD-Based Aeroelastic Analysis Using Distributed Processing
Author(s): A.A. Boeckman
Date: May 2003
Pages: 101
Formats: pdf (1905 KB)
Abstract:
Scope and Method of Study: An enhancement to the STARS aeroelastic analysis module has been developed which employs distributed processing over multiple computers to accelerate the prediction of aeroelastic phenomenon. This can be implemented with any CFD solution, so long as the analysis requires multiple independent simulations.
Distributed processing was used in this study because it offers an effective, low cost speed increase. It can be implemented without change to the CFD flow solvers on which it is applied. The distributed batch-processing can be applied to both steady and unsteady solutions, as well time marching and frequency based approaches. The distributed design will correctly predict aeroelastic characteristics assuming the CFD solver is valid within the flow regime.
Distributed processing is applied by first defining a set of initial conditions and/or predetermined motions for the geometry of study. The initial conditions, motions and three-dimensional geometries are distributed to as many independent networked processors as needed. Each processor performs the CFD calculations in isolation from the other computers. The results are collected for study, which may include sensitivity studies and training data for system models of the CFD solver.
The architecture of the distribution is examined. An exact estimation of the time saved is determined. The requirements for a network of personal computers to implement this method are discussed. Several examples of the use of distributed batch processing are developed.
Findings and Conclusions: Results obtained by this method indicate high levels of accuracy coupled with substantial timesavings for aeroelastic simulations over all flow regimes.
Revised: April 22, 2003 [TJC]