The technology for creating flow visualizations has advanced tremendously in the past few decades. To determine the velocity at every point, researchers used to place a pitot tube at many locations to measure velocity. The precision of this technique was limited by the number of tests and the minimum distance between test locations. Furthermore, the pitot tube had to be rotated to find velocity in more than one dimension.
A much more accurate way to find the velocity at every point is to use Particle Image Velocimetry. The theory behind this method has been known for some time, but only recently has the software power been sufficient to carry out calculations. In PIV, a smoke generator releases a fog of little hydrocarbon particles into a wind tunnel. Lasers pulsing many times a second are aimed at the car, providing the only source of light (below). On command, a digital camera takes an exposure that covers two of the flashes. The resulting image shows the locations of the smoke particles a fraction of a second apart. The two positions constitute one measurement, and the average velocity over the time interval can be calculated. With such a small change in time, the average velocity approaches the instantaneous velocity. Thus, the approximate velocity distribution can be obtained without solving the complete Navier-Stokes equation.
PIV still has some limitations. The image from a camera is in a plane. This means that a 2-D velocity field is immediately available, a great advantage over the pitot tube technique. But the PIV measurement is also limited to two dimensions, so any motion across the width of the car cannot be detected. However, this isn't much of a problem for the case of automobiles because cars for the most part have axial symmetry. Another problem is that velocities around the edges of the image are wildly inaccurate. All along the leading edge of the shot, there are particles that are seen in the second laser pulse, but were not yet in view by the first. Similarly, there are smoke pieces leaving the trailing edge that "disappear," confusing the computer. This can be remedied by ensuring that there is plenty of room between the car and the edge of the image.
As efficient as PIV is, finding the velocity field over an entire car is still a time-consuming endevour, beyond the level of our lab. Therefore, we did not gather our own data for this part. The measurements are from a summer research project supervised by Professor Anderson.
In our lab, we used StarWorks for CFD. Its flow predictions are based on a SoldWorks model. The model surface is broken up to many triangles, and the program calculates flux through each one.