It's off to track we go

Carrying out test plans and automating data extraction

It's finally track day, and there's little I like more than carrying out a good test plan with equipment we know we can trust. Once we got there, the first step was to set up the sensors so we can record data. The sensor PCBs take the voltage from the differential pressure sensors and convert it to a CAN value somewhere between -32,768 and 32,767, being the range of 16-bit signed integers. We didn't code in any other conversions to keep the recorded data as "raw" as possible to make debugging easier, although we did add an averaging over 10-point intervals to limit the amount of data being sent over the CAN bus. With 25 Hz output frequency, these combined to give an effective 250hz of average readings. Both boards plugged in to breakouts on the dashboard, keeping connections central and thus limiting the required length of both the pneumatic tubing and CAN connection cable.  

We had three tests to get through; the on-track methods are summarised below:

Upon track arrival

Straight airflow Z0.35

Target airspeed: 18 ms-1
Corner radius: N/A

Straight airflow Z-0.11

Target airspeed: 18 ms-1
Corner radius: N/A

Representative corner Z0.35

Target airspeed: 14 ms-1
Corner radius: 13.2 m

Our test plans got more and more precise as the year progressed and we learned what extra detail we wanted and needed from our testing. Additions included required ambient and track conditions, guides/processes for pre-track preparation and post-track disassembly/data extraction, the intended purpose and expected results of each test (to help in creation of on-the-spot contingencies or compromises during testing), required uncertainties and tolerances of installations and vehicle driving scenarios, expected time requirements, etc. They were accompanied by post-test reports containing an overview of results, discussions, required future work, and driver feedback. The aim was to have the test plans detailed enough that the test session could still go ahead and appropriate decisions could be made even if the test coordinator/s and/or relevant vehicle section lead engineers were absent due to unforeseen circumstances. 

Figure 1: Example test plans from initial track-testing of the EV24 side-floors that I've retrospectively included here

Figure 2: Pitot-static and yaw probe in a temporary prototype mount being tested on the nose of the vehicle

Figure 3: Recorded data from the S_Z-0.11 rake after the slow speed out lap, used to assess the quality of data and check for null data before committing to the full test run. In this case, note that one of the sensors is clipping the minimum value; this was fine as we could still get the data we needed once up to speed, but this sensor had a new resistor installed after returning from track to offset the voltage to a more appropriate range. Similarly, we also swapped out the resistor on the sensor with high positive voltage offset to avoid clipping the maximum value at speeds closer to Vmax, for future track days. 

Figure 4: Image of EV23 with the S_Z-0.11 rake attached during an initial low-speed test to confirm structural  integrity. The yaw probe had been temporarily removed from the nose for maintenance while this test took place; unfortunately we had no good photos of the car with both rake and yaw probe attached while on track.

Figure 5: S_Z-0.11 rake (raw) data from its full test plan run. Unlike the S_Z0.35 rake which was positioned with some probes inside the recirculation zone, all probes on the more rearward S_Z-11 observed a positive correlation between total pressure and vehicle airspeed, which reflected expectations.

Before going into data extraction, I'll briefly discuss the new yaw probe as seen in Figure 2. At this point it's still very much a prototype design, and was subject to very imprecise manufacturing methods. As a result, its output was a little noisy but the concept was nonetheless proven at this track test, and was sufficient in fulfilling its role of filtering data that satisfies the test conditions. A more precise design with more rigorous wind tunnel testing will be carried out before the next aerodynamics track tests. Figure 6 contains the wind tunnel test results for this yaw probe. Combined uncertainty of the yaw probe and pressure sensors using a RMS method came to 10 Pascals with a 1.25 factor of safety, which suggests that between approximately 10 and 14 ms-1 the precision is only to the nearest 2 degrees and that the results probably shouldn't be trusted at all below 10 ms-1. 0.5 degree precision is achieved above 22 ms-1. Ideally, 0.5 degree precision would be achieved at all speeds above 10 ms-1 in a future improved yaw probe. 

Figure 6: Yaw probe calibration/mapping results from wind tunnel tests at three airspeeds. Fewer teste were run at negative yaw angles after it was clear that the linear trends were continuing. Further velocities were interpolated/extrapolated. 

Unfortunately due to deteriorating weather, we were unable to complete the representative/median cornering case test. This wasn't a huge loss, as it would have had the least emphasis during validation anyway, due to a fair bit of uncertainty at this point in being able to reproduce the exact driving scenario down to the steering, slip, and roll angles. We therefore didn't want to put significant validation effort into trying to tune our CFD setup to match this case, but had planned to use it as more of a guide to identify any significant discrepancies. More emphasis on cornering, as well as braking scenario validation will come later in the year and into next year. 

Extracting the data was easy enough as there wasn't that much to go through, nor was I interested in any relationship between different systems on the car as we're just trying to replicate 3 steady state cases for now, with an emphasis on quality and quantity of data rather than variety. That said, I'd still rather not have to go through all the values and hand pick the ones we want, so I wrote a MATLAB program to do it for me. Each case has two physical requirements, being freestream airspeed and corner radius/yaw angle, both of which can be obtained from the pitot-static and yaw probes respectively. It was simply a case of filtering out data points where both of these requirements were satisfied. I put a tolerance of 0.5 degrees for yaw and 0.5 ms-1 for airspeed which returned a decent slice of data for the straight airflow cases, with a few missing sections towards the end of the S_Z-0.11 test where wind was starting to pick up and intermittently pushed the yaw probe readings outside this tolerance. The flat section observable at the start of the data in figures 3 and 4 was identified and averaged by the script to give a calibrated 0 reading for each sensor. Finally, the CAN data was converted back to voltages and then to total pressure, after which they were all scaled such that the highest total pressure reading for a free-stream-positioned Kiel probe matched its free-stream value in CFD. This normalised the readings to CFD to account for differing gas properties; this changed the readings by a maximum of only 0.8%. 

Validation for the rake data had two criteria:

The second aspect acknowledges the high velocity gradients between freestream and core wake/recirculation regions across the thin perimeter of the wake "cloud", meaning even if the magnitudes are significantly different, it may only be a 5 mm difference in position of the wake that has caused the otherwise concerning magnitude. 

There were no plots or other automated post-processing done to the data, as it did not reflect how it was collected or to be validated. The total pressure magnitudes were a single value and independent of other points, and analysis of the shape of the wake rests on some assumptions that the real shape between the measured probe points at least somewhat resembles that in CFD. There simply isn't enough spatial resolution of theses rakes to visualise the precise shape using only experimental data. Instead, for each point, it was noted if the experimental total pressure suggested a freestream, perimeter, or core wake reading (Figure 7). This simplified "map" of the wake will guide coarse adjustments to the CFD setup before the precise numerical data is used for fine tuning and assessing overall improvement in CFD/track correlation.  

 Figure 7: Simplified map of measured total pressure values for the straight airflow cases at (left) Z=0.35 and (right) Z=-0.11, based on whether the value was freestream (above 90% of true freestream value), or indicative of a boundary region (above 67%), or of a core wake region (below 67%) 

A final note on surface flow tests as mentioned in the CFD post-processing post: The test runs for the tuft testing on the selected bodywork area were conducted at the same target speed in a straight line, with a camera at 100mm zoom positioned perpendicular to the track to capture the tufts externally with a low field of view to reduce perspective distortion. For this simple test, no optical flow analysis was conducted for precise angles or turbulence analysis. I'll cover these results briefly in the next post.