So, the (Wild) Weasels had a great patch that had a wonderfully descriptive acronym for those moments when there just isn't enough whiskey: YGBSM. It stands for "you gotta' be shitting me..." Of course, we all have YGBSM moments. Some of us, more than others. What are the odds that we would buy a high tech, wireless air data boom and it was accidently miscalibrated at the factory? If you answered 100%, you would be correct.
Interestingly, if you combined the resident stick monkey and a miscalibrated boom with conventional technique for measuring upwash error; you'd actually mask the error and have a pretty good curve that worked anyway. This occurs only if the stick monkey doesn't realize that the upwash plot ended up on the wrong side of the 1:1 1G pitch slope datum (bore site line). That's where an aerodynamicist has a huge tactical advantage over a history major--they notice stuff like that. Figure 1 shows what an upwash plot should look like; and Figure 2 shows our initial upwash curve. The astute reader will note that on our original plot, upwash isn't even upwash, it's downwash...that can't physically happen. This is illustrated in the picture of the smoke stream wind tunnel test in Figure 3.
Figure 3 shows how the upwash deflects the AOA vane nose down. In other words, the vane senses more AOA than is actually present at the wing. During analysis, we need to dial out this error. Unfortunately, a wing tip is about the worst place you can mount an air data boom--a nose location is preferable. Unfortunately, the nose of the mighty RV-4 has one of Craig Catto's nifty props spinning there, making it really difficult to bolt on a boom.
We should have assumed that just like every other instrument you bolt into an airplane, you've got to dial in the vanes on the boom as well. Turns out there is a nifty measurement fixture that allows precisely setting the vane to calibrate, it just never occurred to us that the angles used in the factory calibration were incorrect. The factory plot calculated the alpha and beta curves using 0, 10, 20 and 30 degrees up and down, left and right; but damn if the fixture didn't hold the vane at 0, 15, 30 and 45 degrees! I'm sure it was an honest mistake as they likely went through several iterations of boom design and configuration as well alignment fixtures.
The alpha and beta vanes on the boom are attached to potentiometers that provide raw "counts" to the data stream via wifi. Our software allows the input of boom "curves" to turn raw data into angles. Needless to say, we've been flying with the wrong curves programmed; but as I said, we got lucky by simply calibrating the boom to pitch; and now that we've got correct curves, we have been able to develop corrections that we can apply retroactively to old flight test data. Which is good.
To calibrate these potentiometers, a test fixture or protractor is used to precisely hold the vane angle, then debugging software is run to measure "counts". Our vanes have a 90 degree count range of 2505, so each degree is 27.8 counts. We are challenging these pots to provide 1/10th of degree of resolution, which means we have to be able to accurately measure to within just under 3 counts (2.78).
Figure 4 shows the vane with a protractor fixture fitted to accurately position the vane in 5 degree increments. The string you see in the photo is some kite twine gently wrapped around the vane axle allowing it to be positioned correctly for calibration by increasing friction. Figure 5 shows a test fixture provided by the boom manufacturer that holds the vane at 0, 15, 30 and 45 degrees.
Multiple tests were run on three different occasions with each calibration fixture. Results were averaged for each series of tests to refine data. Although a plot of results appears to be linear, it turns out when you plot data at 5 degree intervals, there is a belly in the curve. And, to accurately describe that curve, a 4th order polynomial is required to get sufficient resolution (i.e., error less than 0.1 degree). Our initial analysis was conducted in Excel, which doesn't have sufficient plotting resolution to show the shape of the curve. Fortunately, MatLab is optimized for that type of visualization and once we ran the numbers there, it became immediately obvious that the calibration wasn't a straight line. Since the protractor method provided more data points, it more accurately captured the calibration curve. Figure 6 shows an excel plot of alpha vane data; and Figure 7 shows a MatLab regression analysis as well as plot that allows you to see the variance of the calibration curve.
The good news is that when we apply the new algorithm, we can measure the vane to an average accuracy better than 1/10th of a degree. This is shown in the table in Figure 8.
So, after all of this "do over" homework, we've now got an upwash correction that looks like, ah, an upwash correction. This is shown in Figure 9. Amen, and pass the whiskey.