I used the YE PG documents measuring the 8 YE stations to estimate the Transfer Plate positions, so that I could compare this to Eartly's numbers. The reference dowel pin on the SLM is very close to the Transfer Plate center (6.724mm), and the reference dowel is also very close to one of the PG targets, so I can use that target to estimate the Transfer Plate center. I did, modulo one measurement where I had to use a different PG target since the special one wasn't measured. In what follows those are called New or denoted by the prefix N.
Dave Eartly's numbers are called Dave or denoted by the prefix D.
Estimates from a Cocoa fit to the Transfer Lines at 0T are called Cocoa or denoted by the prefix C.
Some comparisons appear in the figure below. Since Cocoa fits along the axial lines are not sensitive to SLM coordinate Y, I only show X (top row, approximately radial) and Z (bottom row, approximately tangent).
The left hand column compares Dave's PG numbers to my PG numbers. Clearly they agree very well about half the time. (I omit the X values for ME+3/5 in the scatter plot since they wreck the scale) The second column from the left shows the deviations, and the central spike is quite clear--the sigma of the fit (in the range (-2,2)) is 65 microns for X.
Station 3 contributes more than its share of large X deviations, but these occur across the board, ranging from -7 to -1. This suggests something systematic in the calculations, but it isn't obvious which is at fault. For station 3, the average of DaveX-CocoaX is -1.4, while the average of NewX-CocoaX is 1.9.
The Dave vs Cocoa X deviations by station are as follows:
The New vs Cocoa X deviations by station are as follows:
Correlations would be interesting, so below are scatter plots of the difference between New and Dave PG vs Dave minus Cocoa (left column) and New Minus Cocoa (right column). The top row is all entries; the second row is entries for the narrow band where New and Dave agree pretty well, and the bottom row is for the cases where New and Dave don't agree so well. The bottom two cases actually have the same trending: the sign on the dx component is different in the left bottom plot.
The average of the New X is 7176.74, of the Dave X is 7176.38, and of the Cocoa X is 7175.70. Ideal is 7175.83
As may be seen from the following, there is a small core of estimates where Dave and New PG agree quite well. If I restrict comparisons to that small core, the distribution of deviations from Cocoa improves, but not tremendously. (Also requiring the Y PG estimates to disagree by less than 1 does not change anything.) I tentatively conclude that The deviations of the Cocoa fit from PG are not primarily due to the variation in PG.
From the RMS of these distributions I also conclude that for X and Z The New Photogrammetry estimates are better than Eartly's
For completeness, even though I can't compare this to Cocoa, the difference between New and Dave Y estimates (subtracting off 84.3 for the distance from the target to the reference point) follows:
Modified 06-August-2009 at 09:37
http://hep.physics.wisc.edu/~jnb/cms/06Aug2009
| Previous notes | Next notes | Main slide directory |