## Team for Advanced Flow Simulation and Modeling |

For more information: |
## AHPCRC Bulletin: Summer 1995 - Volume 5 Number 3## Larger Scale Simulation Capability for Ram Air Parafoils is Achieved on the Cray T3D## Shahrouz Aliabadi (AHPCRC-UMN)Realistic numerical simulations of complex flows require very high grid resolutions. Using highly refined meshes in numerical simulations not only improves the accuracy of the solution but also may predict some small-scale physical characteristics such as high frequency modes in turbulence. These physical characteristics might be absent in a solution obtained from a coarser mesh. The major barriers involved in these large-scale computations are memory and CPU time. Efficient algorithms and advanced hardware with sufficient computational power are the keys to removing the barriers involved in large scale simulations.The author of this article, a member of the research team of Tayfun Tezduyar, Professor of Aerospace Engineering and Mechanics, is carrying out very large scale computations to predict the flow field and the dynamics of ram air parafoils. The steady-state performance of these parafoils is simulated using a finite element mesh consisting of 9,741,930 nodes and 9,595,520 hexahedral elements. A coupled nonlinear system of equations with 38,391,715 unknowns is solved at every time step. The Cray T3D with 512 processors and highly optimized finite element flow solvers were utilized to solve this problem. The AHPCRC has, under special arrangements, limited access to this Cray T3D owned and operated by the Minnesota Supercomputer Center, Inc. The finite element flow solvers used are based on state-of-the-art stabilized formulations which maintain their numerical stability and accuracy even at high Mach numbers and high Reynolds numbers. The formulations are also capable of handling problems with moving boundaries and interfaces. The coupled equation systems arising from the finite element discretization are solved using matrix-free iteration techniques. The matrix-free iterations totally eliminate the need to store the element-level matrices corresponding to the left-hand-side matrix. This significantly reduces the memory requirements in finite element computations. Dealing with the large data sets involved in these computations at this scale was a major undertaking. Special techniques were used to display the pressure distribution on the parafoil surface. The visualization of such large data sets is still an open issue in flow simulations involving complex, real-world problems, and needs further research. |