testWheel script

This moves the specified wheel between two extreme positions (RAW values 1000 and 430,000) and back again. At the extremes it reads the encoder values (ENC keywords) and then calculates two scale factors: one for motion from low to high RAW numbers, the other for the reverse motion.

2004 April
2005 June

2004 April

I have been running sets of twenty testWheel scripts in all three wheels, then calculating the mean and standard deviation of the resulting scale factors. Also, there was some limited information in some ktest results done before ESI shipped from Lick. Unfortunately, the upper wheel apparently was not tested, or at least the results aren't in the usual place on koki.

Results

Note that for some reason these values are a factor of two higher than the scale calculated in testWheel. However, the fractional standard deviations, in parentheses, are what I am looking at. So preship results showed the stability of the motor-to-encoder scale factor to be 0.0016%–0.003%. Again, no ktests were recorded on koki for the upper wheel.

Table 2. Current tests (testWheel)

So you can see that the upper wheel is much sloppier and less reproducible than the other two wheels. The other two wheels look close to the original condition. Note also that the scales in the two directions are much more different for the upper wheel than for the lower.

When testWheel was started with the IFU installed in the upper wheel, a bigger problem was immediately seen. So the test was stopped immediately so that the movement didn’t further aggravate the problem. Only two measurements of the two scale factors were recorded:

Upper wheel (SLMSK) with IFU installed: scale1 = 0.24279, 0.24008, scale2 = 0.24833, 0.24669

Removal of the IFU and reinstallation of the LowD_6.0 slit did not put us back to the state we were in earlier. The stage was still showing signs of serious slippage, and again testWheel was stopped immediately. Quick examination of the upper wheel showed the drive chain chattering in various positions, becoming worse when slight pressure was put on the wheel or drive belt. It appeared that the harmonic drive is the problem.

The drive belt tension was reduced slightly in an attempt to ameliorate the situation. Rather than run an end-to-end test like testWheel, we chose to simply try to set to the desired slit for the weekend, 1.00_arcsec. This worked smoothly. No problems were seen, and no hunting around the target position was noticed. There was some nonzero settling time, but qualitatively no more than is normal. The upper wheel was then moved to 0.75_arcsec, the adjacent echelle slit, again with no problems. It was moved back to 1.00_arcsec, and left there. Tear down and replacement of the harmonic drive was postponed until after this two-night ESI run.

After the ESI run, the scales were measured again, with the maximum speed and the acceleration of the wheel lowered by a factor of two.

Table 3. Unhealthy wheel, with speed and acceleration lowered by a factor of two.

The speed was raised back to normal, but the acceleration kept a factor of two lower than the original values, and the tests rerun.

Table 4. Unhealthy wheel, acceleration lowered by a factor of two.

After a thorough analysis, replacement of the harmonic drive, and retensioning of the belt, the upper wheel seems to be returned to health. Measurements in this state give the following:

Table 5. Repaired, healthy wheel.

Note that a plot of all of these measurements shows some interesting trends. The unhealthy wheel shows a lower stability, but also a lower scale factor. The "scale2" value in particular was lower than the "healthy" value shown in Table 5. This could provide a sensitive, single-measurement test.

Fig. 1. Upper wheel scales both before (1-63) and after (64-73) fixing the motor and gearbox. Blue and red data are "scale1" and "scale2," respectively. Note that the unhealthy wheel shows a much larger scatter, with a lower mean value than the healthy wheel.
(Click on the figure to toggle between reduced and full size.)

Fig. 2. A blow-up of Figure 1 in the Y-direction. Note the difference in scales. The scatter in scale measurements for a healthy wheel is two orders of magnitude below that of an unhealthy wheel.
(Click on the figure to toggle between reduced and full size.)

2005 June

The problem recurred sporadically, and we only got one scale measurement from testWheel: 0.2502468. This is many (135!) sigma away from the value above: scale1 = 0.2507087 ± 0.0000034. While multiple move attempts of the wheel provided obvious evidence of an ongoing problem, in those cases where an individual move worked (apparently flawlessly), the testWheel scale comparison still provides dramatic evidence of slippage between the motor encoder and the Renishaw encoder.