^y          |
Q0 | Q1        |
---+----->x    |
Q3 | Q2        V g (gravity)

Yefanov, Oleksandr <oleksandr.yefanov@cfel.de> Tue 8/18/2020 1:18 PM
To: Dubrovin, Mikhail; O'Grady, Paul Christopher
Cc: Hart, Philip Adam; Chun Hong Yoon <chuckyoon@stanford.edu>; Batyuk, Alex; Mariani, Valerio; McKelvey, Mark E
2 attachments (5 MB)Download allSave all to OneDrive - SLAC National Accelerator Laboratory

Hi Michail and others,
We’ve found the problem!!!!!


Hi Michail and others,
 
We’ve found the problem!!!!!
Sorry to say, but it’s from your side. :)
Solution at the end of the e-mail as a bonus for those who read the whole e-mail. :) 
Actually I’ve written the e-mail while searching for the answer. But the text might be useful for those who work with geometries.
 
I believe that the optical metrology is right – I was wrong once, by thinking that CS-PAD metrology was not good and I don’t want to repeat my mistake. :) 
But as with CS-PAD I’m afraid the problem is with attributing the optical metrology to the measured data. (here I’m right – see below :)
 
Unfortunately your arguments about influence of z-offset or out-of-plane tilts can be discarded using simple math.
For example at the attached image (…current.png) the middle green ring has radius of 200px = 20mm and the detector is approx. at 120mm. So to have in-plane offset of 2 pixels you’d need to shift a panel by 2mm. That’s too much. Also, as you can see, at left-bottom panel the rings overlap at the top side, but have distance at the left side of the panel. So to correct it you’d need out-of-plane rotation. 2 pixels (at 200px distance) would correspond to 11.5 degrees. Not realistic. :(  And in the attached example error is 4-5 pixels!!! I’m not even mentioning that in case of the tilts and z-offset the behavior is different at different parts of the panel. You can see the effect of these distortions from geoptimizer’s error map.
 
The interesting observation is that I really see z-offset for panels and it is quite big (up to 1mm for corner panel). Fortunately in my algorithm z-offset cannot be confused with x-y (different parameter is considered), but as I mentioned, due to the geometry of the detector z-offset is much more sensitive to noise/error (for example wrong peaks). Therefore to find reliable z-offset of different panels I need more data. I’ve made a very good geometry for AGIPD (with z-offset up to 1mm) using ONLY measured data (optical metrology doesn’t exist for AGIPD), so it is actually possible. I’ll try to optimize z later when I have access to more data.
 
 
Now SOLUTION !!!
 
My student Marina tried to “rotate” the geometry – shifting the ss coordinate by 1408, 2816 and 4224 (1,2,3 quadrants). Apparently shift by 4224 gives correct geometry!!! Have a look at the attached shift4224.png. I’ve also indexed the data and checked with geoptimizer – it looks right (except z-offsets).
So please modify in this way your script that makes the geometry from metrology. Don’t forget to rotate the geometry to have visually correct detector – it doesn’t matter for SFX, but nice for users.
Please share the resulting geometry – I’d like to use it.
And I’ll try to add z-offsets to your resulting geometry when I have more data.
 
And Mikhail, I really want to thank you very much for good optical metrologies! It really saves users a lot of time!
Regards,
Oleksandr.