So far we have been concentrating on two types of ships: (1) frigates with helicopter landing pads on the deck behind the hangar and (2) aircraft carriers and LHA's (e.g. U.S.S. Saipan) with several helicopter landing spots on the deck around the control tower. The airwake influences on the helicopter are quite different in these two cases. For frigates the flow separation area behind the hangar cube has a strong effect on a landing helicopter, while on LHA's the deck leading edge vortex and separation are the key flow phenomena .
1. Airwake flow from a cube with frigate hangar dimensions
The CFL3D package from NASA Langley is used here to simulate the mean flow which will be given as a background flow to the unsteady flow computation of NLDE. For preliminary simulations we simplified the frigate hangar to a same dimension box on a hard wall. The incoming uniform flow has a Mach number of 0.05 with 0 yaw angle. Fig. 1 and fig. 2 are two slices, horizontal and vertical planes, showing Mach number contours around the hangar. The flow visualization pattern of (fig. 3) [8], shows four distinct flow regions behind the cube. This three dimensional vortex and reverse flow has very low speed but generally is very unsteady and yaw-dependent. The results for the perturbation quantities will be given in next section.
2. Airwake flow from a box with aircraft carrier dimensions
In this example, the flow around a box of roughly the same dimensions as an LHA is simulated. The incoming uniform flow has a Mach number of 0.06. Two yaw angles are simulated: 0 and 45 degrees. The flow field around the cube is similar to the pattern shown in frigate hangar case. But as mentioned before, in this case we need to consider where the helicopters would be landing and at what height above the flight deck the rotor blades would be located. For a 45 degree yaw angle case, fig. 4 shows the flow velocity vector on this plane which is about 17 feet above the deck for the ship we simulated. The flow is quite complicated, especially near the area just against the wind. If we take an intersection plane along the longitude of the ship, the leading edge vortex is shown in fig. 5. Note that this is a typical flow field around the helicopter landing area tunnel strikes have been reported.
While we have simulated relatively simple geometries here, in the final paper we will include more complex geometries (such as including the 'island' or control tower on the LHA). Preliminary results using unstructured grids for complex ship configurations will also be included in this paper.
For the unstructured grid flow field predictions we will be using the PUMA code [14] from Dr. Christopher Bruner (NAWC). PUMA (Parallel Unstructured Maritime Aerodynamics) is a computer program for the analysis of internal and external non-reacting compressible flows over arbitrary complex geometries. PUMA is written entirely in ANSI C and uses MPI (Message Passing Interface) to ensure high portability and good performance.
PUMA is based on FVM (Finite Volume Method) and supports mixed- topology unstructured grids composed of tetrahedra, wedges, pyramids and hexahedra. The code may be run so as to preserve time accuracy for unsteady problems, or may be run using a pseudo-unsteady formulation to enhance convergence to steady-state. Either explicit or implicit time integration may be used. Primitive flow quantities (density, velocity, and pressure) are computed at the cell centers and saved on exit.
Because PUMA uses DMA (Dynamic Memory Allocation), problem size is limited only by the amount of memory available on the machine. Since PUMA is targeted for distributed-memory parallel computers, which usually have abundance of memory, little effort has been put on reducing PUMA's memory requirements. Currently, with double precision floating point variables used throughout the code, PUMA needs 582 bytes/cell and 624 bytes/face, not including message passing overhead. For tetrahedral grids, this amounts to 2002 bytes/cell (or 250 words/ cell). This requirement can be reduced significantly by compiling the code using single precision floating point variables for which an option is provided.
| Computer | Arcitecture | Nodes | Compile option | Iter | Sec | MFLOPs | |
| grimm | P-II 266 cluster | 8 | mpicc -O4 | 500 | 983 | 105.21 | |
| SP2 | SGI Power II | 8 | mpcc -O3 (pwr2) | 500 | 736 | 140.52 | |
| power | SGI Power Chall. | 8 | cc -O3 (64 bit) | 500 | 577 | 179.24 |
We can run the same cases in PUMA and in NLDE code by converting the structured grids (used in NLDE code) to unstructured tetrahedral grids trivially for PUMA.