ANSYS CFD - Things To Look Out For

ANSYS is a very powerful and extensive simulation software with a capability to conduct virtually any kind of simulation accurately. For a successful simulation all one has to do is to take care that the boundary conditions and the numerical models chosen replicate all conditions of the intended application of the design.

For our design purposes we conducted virtual wind tunnel tests and the conditions were replicated in the Fluent module of ANSYS. For wind tunnel tests we are interested in the region where the air will flow and not in side the pod model, thus the region of air flow must be a solid body which is meshed to solve for given conditions.

Defining the solid region in the Design Modeler of ANSYS is a relatively easy task. The profile of the tube has to be extruded first. The pod design can then be imported as a STEP/IGES file. The next step is to perform a Boolean

subtraction of the pod from the tube which will result in a solid body consisting of only the region of air flow and a void in place of the pod. Furthermore, in these wind tunnel tests the body being tested that is the pod stays stationary and the air is made to flow against it, thus extra care has to be taken to interpret the boundary conditions in order to get a result comparable to the intended application.

For our design we considered the length of the tube track to be infinitely long as the ratio of the pod length to the track length is very high and such an assumption also makes the setup easier. When running in the tube at distances, far away from the front and the back end of the pod, the pressure would be 4000 Pa. The gradients in pressure/density of the air would be observed only in the vicinity of the pod. Thus, the mesh has to be long enough to accommodate the differences and gradients of the air pressure and density up till this finite length. Local refinement is necessary near the pod regions to get accurate results of the lift and drag forces acting on the pod's surface.

ANSYS gives a wide range of turbulence models to choose from. The K – Omega SST turbulence model with compressibility effects enabled replicates the real-world conditions of running a pod inside a tube most closely. The Far Field Pressure inlet and outlet options replicate the infinite pressure of 4000 Pa along the tube. Conducting trial simulations with many of these models along with different viscosity models, we arrived at the following boundary conditions implemented in ANSYS that are usually used for Transient Compressible flows:

Finalizing the boundary conditions was a challenging task for the team. Even with a coarse mesh, each trial simulation took around 6 to 8 hours to converge to a solution, sometimes with positive results and sometimes with garbage results. One of the mistakes we did initially was using a pressure inlet and a velocity outlet at the ends of pod to achieve the required pod velocity. This may work at nominal speeds, but at higher velocities, this condition does not give the accurate results. This assumption was confirmed as the simulated results with this boundary condition did not match the calculated results. At this point we definitely needed help. Our faculty advisor at ASU, Dr. H.P. Huang, gave us valuable insights over the turbulence models to choose from for simulating a high speed compressible flow, and how to reduce the wake formation behind the pod. These insights helped us to greatly improve our result accuracies.

With these conditions implemented, over 15 different shapes were iterated to get a shape which gave a smooth flow at different speeds. Theoretically, we expected to achieve a maximum speed of around 200m/s with our final design. Below shows an image of our first design results. This design was quite large in comparison with the tube as the initial subsystems to be placed inside were not optimized.

Additionally, mesh size also plays an important role. Finer the mesh, more precise are the results. How fine the mesh can be greatly depends on the availability of computational resources. Being slightly limited in computational power, we ran the simulations with an overall mesh size of around 4 million nodes. Reasonably accurate results were obtained with this mesh size. We also have plans to validate these results by running a series of Wind Tunnel tests at ASU.