15 replies to this topic

[Tonk]

l just thought I'd post a story about measuring sensor tilt and how this is affected by flexing - and doing this remotely!

We have been trying to commision our second scope on our remote piggy-back rig in Chile for a while now. The rig uses a 10Micron 2000 mount and the second telescope is a Tak FSQ106 with a FTF3531 focuser, a Starlight Xpress MaxiWheel filter wheel and a QHY600 camera.

Attempts to use ASTAP to measure and correct some very obvious sensor tilt was both confusing and ultimately unsuccessful (ending up in a worse position!). Measurements were never reproducable and changed over time. This was costing us tech support fees and getting nowhere.

After some thought the team decided to use the NINA aberration tool as it makes tilt measurements using a very different method. ASTAP and a few other notable tools uses a single (or stacked) image taken at best focus and tries to estimate tilt by analysing the shapes of stars throughout the image. The main flaw in this approach is the presense of optical aberrations from causes other than tilt. NINA on the otherhand measures where best focus is achieved in all 4 corners and the center of the image. This is largely immune to aberrated star shapes and leads to a far more reliable way to measure tilt.

As the corners of the FSQ image were showing signs of coma it was best for us to use a 80% ROI to avoid perturbed results arising from causes other than sensor tilt.

We quickly worked out that sensor tilt was variable and depended on where in the sky we pointed the scope. So we had tilt AND some sort of flexing/bending going on. We needed to find a way to analyse this remotely and determine if our orginal problems with ASTAP was the due to motions of the pier, mount or the OTA, focuser, flop or maybe a cable snagging.

Measuring Sensor Tilt

To do this we chose to sample the corner and center focus position at a fixed altitude of 45 degrees and walk the scope around the sky at azimuth positions 10, 45, 90, 135, 170, 190, 225, 270, 315 and 350 degrees.  Azimuths 0 and 180 had to be avoided due to pier collision risks.

We then plotted the results of 3 around-the-sky runs to check the results were reproducible with NINA - it was. The plots are of the z axis displacement of the 4 corners in focuser steps with respect to the center of the sensor (were DTL = Displacement Top Left, hence DTR, DBL, DBR should now be obvious).



So we took the average of the displacement values at each sensor corner and, using the dimensions of the sensor and the distance moved by each focuser step, computed and plotted the angle of up-down tilt (XZ plane) and side-to-side tilt (YZ plane). The results below show we have an average up-down tilt of ~0.5 degrees and a side-to-side of around ~0.03 degrees, both with some obvious and reproducable (unwanted) variations due to flexing.



Modeling OTA flexing

Part two of the analysis involved calculating the components of the force of gravity acting on selected planes associated with elements of the rig as the scope positions are varied. The obvious ones to try out were the two planes orthogonal to the optical axis and parallel to the edges of the sensor. If the sensor is placed in the XY plane (with sensor center at (0,0,0) ) the two planes to measure are the XZ and YZ planes. This should be related to bending in the OTA - most likely with the focuser barrel coupling (Occams Razor says lets go for the focuser first!).

What we are now looking for is a correspondance between the shapes of of the plots of the *measured* tilts as azimuth is varied with the shapes of the plots of the *calculated* gravity components as azimuth is varied.

The steps to do this involved converting the altitude/azimuth positions we used to hour-angle and declination using the latitude of the rig (30.5 degrees south) and from the hour-angle, declination and latitude compute the orientation of the sensor at each azimuth position (via a Tait-Bryan rotation matrix) - expressed in terms of the 3D coordinates of the 4 corners of the sensor. From these orientations the cosine of the angle formed between the gravity vector and the vector at the intersection of the XY/XZ and XY/YZ planes gave us the component strengths of the force of gravity acting on the OTA. The plots of these normalised force strengths for various latitudes and azimuths is shown below. Note that the meridian flip of the OTA has been accounted for and why the plots are shown independently for the OTA being on the east and west sides of the pier.



You will note that the shapes of the measured tilt plots and the OTA flexing force plots for latitude 30S correlate very well (you match up the corresponding coloured plots). So it is very likely OTA bending we are observing and we can likely not blame the pier, the mount or cable snagging and instead search for ways to stiffen up the OTA - there is a tightening screw on the FTF3515 that might help.

Why did we go to all this effort??? We'll its a remote rig and we have to pay the tech support by the minute to do investigations and make adjustments on our behalf. Its not cheap so we thought it might be cool if NINA could help us remotely diagnose the issue.

The biggest takeaway for us is to switch from ASTAP to NINA when it comes to measuring sensor tilt reliably. The second takeaway is it is possible in the right circumstances to get some insights into the origins of flexing in the rig. However we dont yet have a model for other potential causes of tilt variation etc - we just got lucky that our problem is very likely only with the OTA and that in turn was easy to model and compare.
 

Edited by Tonk, 15 April 2024 - 10:47 AM.

[lumos]

Great detective work. What are you going to do to get rid of the OTA tilt? 

[Tonk]

 

Great detective work. What are you going to do to get rid of the OTA tilt?

We have a number of things to consider:

1. Is the amplitude of the flex still within the critical focus zone for the optics? If yes then there is nothing to worry about. Fingers crossed that this is so!

2. The FTF3515 focuser has a tension screw that tightens down to reduce lateral motion - however the above analysis is showing that the measured up/down amplitude is greatly reduced already compared to the side-to-side motion wereas the calculated amplitudes are the otherway round. I.e. the  the up-down motion is already constrained wrt to side-to-side. I suspect further tightening likely wont improve things and will just introduce potential binding problems for the focuser motor.

3. Just concentrate on reducing the tilt so that the azimuth regions away from meridian flip regions are optimised to be near flat. Basically we carefully pick a specific azimuth to do the tilt measurements and adjustments

4. The fall back is modern astronomical image processing software can used trained neural nets* to deconvolute star aberrations. We have tried this with the current FSQ tilt in effect and the results are acceptable but could be improved if there was less tilt in the first place.

What it did show us is that the ASTAP approach to measuring tilt was too simplistic and easily fooled. The NINA approach has proven far more robust and allowed us to extract both tilt and flex contributions



(* I refuse to call this A.I. because it is not - these neural nets are highly constrained to solve specific problems for which they are trained and are just working as sophiticated pattern matchers giving a set of ranked (probalistic) matches that directs the software tool to do something specific. This is not "intellegence" - you cannot interogate it to explain its reasoning)

Edited by Tonk, 15 April 2024 - 10:40 AM.

[han.k]

After some thought the team decided to use the NINA aberration tool as it makes tilt measurements using a very different method. ASTAP and a few other notable tools uses a single (or stacked) image taken at best focus and tries to estimate tilt by analysing the shapes of stars throughout the image. The main flaw in this approach is the presense of optical aberrations from causes other than tilt. NINA on the otherhand measures where best focus is achieved in all 4 corners and the center of the image. This is largely immune to aberrated star shapes and leads to a far more reliable way to measure tilt.

 

I assume you have never tried the ASTAP best focus method for nine areas in a series of images:

 

http://www.hnsky.org...p#inspector_tab

 

It is available in stackmenu (ctrl+A), tab "inspector".  Nine areas will work better then four areas.

 

Han

[Tonk]

Yes we used the octagon tool in ASTAP from the start. However NINA gives precise focus displacements of the sensor "corners" from the focus position in the centre and the user can choose the corner position by setting a ROI. NINA moves the focuser to find all focus position of the sample points

[han.k]

I'm not referring to the octogan tool but an other tool introduced in jan 2020 to measure tilt and curvature.  It is in the stack menu tab inspection. It does an hyperbola curve fitting to find the best focus for each of the nine areas using a series of images. The same principle Nina is using.

[Tonk]

Does ASTAP allow you to set an ROI?

That was our biggest issue wrt to getting repeateable results as despite Takahashi' claims that the flat image circle of the FSQ106 is 80mm, it most certainly is not! Modern cameras with very small pixels show that this claim does not stand up - we appear to have astigmatism in just one corner (oval stars switch axis oriention by 90 degrees between inside and outside focus). However the ability to set a smaller ROI helped enormously and gave us repeatable results.

 

Edited by Tonk, 16 April 2024 - 01:00 PM.

[han.k]

I have also noted that the usable image circle could be different then specified.

 

ASTAP does not allows setting a ROI for image inspection. But it should in principle not influence the measurement with a proper star detection and outlier filtering.The whole measurement is time consuming and requires some effort. Nina has the advantage the measurement is automated. But it would be interesting if you could do a comparison test. For each position you just have to save a series of images at different focus positions. The focus position should be in the header.

 

A friend of me has a FSQ106 and mechanical it is very solid. I'm surprised you measure 0.5 degrees tilt drift. How much does that influence the focus of the stars at the corners of the image?

 

The vertical axis "Normalised gravity" is unclear to me.

[Tonk]

 

I'm surprised you measure 0.5 degrees tilt drift.

Its not 0.5 degrees tilt drift - The 0.5 value is the average of the up/down direction of tilt (a tilt we must eliminate by conventional means). The range either side of this value is only around +/- 0.08 degrees if you look at the graph. The most likely cause is the FTF3515 focuser with the very large and heavy SX Maxi filter wheel and QHY600 camera acting on the focuser barrel according to the direction of gravity (as our analysis indicates)

 

 

The vertical axis "Normalised gravity" is unclear to me.

Obviously the values are not the actual SI units of force due to gravity - they are instead normalised numbers from 1 to -1 to indicate the proportional degree and direction the force is acting on the chosen plane. 1 is maximum, 0 is minimum and the sign is the direction.

Edited by Tonk, 16 April 2024 - 07:53 AM.

[han.k]

The 0.5 value is the average of the up/down direction

 

How can a direction been expressed in unit degrees? The graph title tells me Tilt(degrees) is "sensor deviation angle". Is this the tube flexing and how can you calculate it? I assume bending occurs in the couplings and lens mounting and less in the tube itself.

 

 

Obviously the values are not the actual SI units of force due to gravity - they are instead normalised numbers from 1 to -1 to indicate the proportional degree and direction the force is acting on the chosen plane. 1 is maximum, 0 is minimum and the sign is the direction.

The forces causing bending are a function of the azimuth and altitude of the direction the telescope is pointing at. Looking to the gravity graph, why not making the altitude and azimuth the variables and the latitude fixed?

Edited by han.k, 16 April 2024 - 09:02 AM.

[Tonk]

 

How can a direction been expressed in unit degrees?

??? I really dont know what you are asking here. The final graph of the measured tilt shows the two component tilt angles - the tilt angle measured across the top/bottom the sensor (about its intrinsic x axis) and the side-to-side tilt angle (about its intrinsic Y axis). So these are calculated via the angles between the sensor plane (derived from the 3D coords of any 2 adjacent corners of the sensor and the center of the sensor at (0,0,0)) and the orthogonal XZ and YZ planes.
 

The forces causing bending are a function of the azimuth and altitude of the direction the telescope is pointing at

 

Yes - this is experessed by the gravity vector we applied - it points at the center of the Earth. However you also have to take into account the 3D orientation of the sensor (and hence the OTA) on a GEM mount to know how the components of the force are acting to tilt the sensor about both the sensor X axis and Y axis

 

 

Is this the tube flexing and how can you calculate it?

Yes it is something in the OTA bending smoothly. How did we calulate it - well I thought I explained that sufficiently - maybe not, so more details below.

The likely culprit is the FTF3515 focuser OR the SX Maxi filterwheel.

When we got the latter we immediately noticed that attaching a large camera to it caused it to noticable flex with the lid bowing outwards at its center and the edge of the lid lifting off the body between the bolts holding the lid to the body. The front plate/lid of the filter wheel was relatively thin (3mm) compared to the body of the wheel and it had a large diameter. We did a number of modifications to reduce this flex - we added a pillar extension to the filter carousel axis and had the lid bolted to this central pillar so the center of the lid was now directly bolted to the body of the focuser - this increased the stiffness of the lid 8 times (stiffness is inverse cube of the length being bent - we halved that length). The lid is bolted to the body using only 7 bolts. We could observe (using a light) that the lid edge between adjacent bolts would lift off the body depending on orientation of the filter wheel on the mount (i.e. where in the sky the mount was pointed). To reduce this flex we added an additional evenly spaced 14 bolts around the lid rim for a total of 21 bolts. Still after all this it could be possible that we are measuring some residual bending in the giant filter wheel.
 

 

why not making the altitude and azimuth the variables ...

We needed to work out the orientation of the sensor and hence OTA in 3D space on a GEM mount at all the points we measured. To do this we used the appropriate intrinsic Tait-Bryan rotation matrix (https://en.wikipedia...it–Bryan_angles) to rotate a reference plane (containing the sensor) and the reference axes (passing through the center of the sensor) into a final orientation - from which we could apply the gravity vector.. The inputs that enable these rotation calculations are 3 angles derived from the orientation of the mount (Lat, Dec and HA) - hence altitude and azimuth were not appropriate as these are not allied to mount orientation (and hence the OTA/camera sensor orientation).

This diagram might help. The red rectangle represents the camera sensor



Note angle x is latitude, angle z is hour angle and angle y is (90 - declination). Also the reference sensor rectangle was set in the XY plane with its center at (0,0,0) so the applied Tait-Bryan rotations preserve the center at (0,0,0)
 

 

... and the latitude fixed?

The gravity force graphs were done at a selection of various southern latitudes to observe how tilting the GEM mount at various different latitudes varies the relative strength of the bending forces along two orthogonal planes intersecting on the optical axis (and each parallel to one of the side of the sensor - the sensor is assumed to be aligned to PA 0). The only graph relevant to our case is the -30S graph. The rest of the graphs might be useful for others using GEM mounts at different latitudes. For northern latitudes the gravity force graphs just need the 4 traces reflected left/right (but not the X axis).

Edited by Tonk, 16 April 2024 - 01:11 PM.

[han.k]

Nice sketch but this is still unclear:

 

The results below show we have an average up-down tilt of ~0.5 degrees and a side-to-side of around ~0.03 degrees

 

Is this ~0.5 degrees the flexing between the two telescopes solve solutions or between the 10 micro mount position indication and the FSQ106 solve position?

 

 

Also the reference sensor rectangle was set in the XY plane

 

A reference sensor??? This is not clear.

[Tonk]

 

s this ~0.5 degrees the flexing between the two telescopes solve solutions or between the 10 micro mount position indication and the FSQ106 solve position?

No - this is not 0.5 degree of flexing. After adjustments the technician has left the Gerd Neumann camera tilt adjuster with a mean 0.5 degree tilt around the X axis and a mean 0.03 degrees around the Y axis.

As seen on my measurement graph - it shows that the flexing is an additional +/- 0.08 degrees  about the mean angle around the X axis and +/- 0.06 degrees about the mean angle around the Y axis

 

 

A reference sensor??? This is not clear.

Yes - in part 2 I'm modeling an imaginary sensor to calculate its 3D position in space so as to work out the relative strength of component forces on planes related to the XY plane in which I placed that imaginary sensor. To do this the imaginary sensor has to be placed in a starting position before the relevant rotations are computed. This starting position is the (imaginary) reference sensor. You can see the starting position in my diagram - the Z axis points at the zenith and the Y axis is pointed south. From that position I rotate the sensor construct through 3 angles (derived from the original altitude and azimuth positions and latitude) to arrive at its final orientation as dictated by the equatorial mount. Note also that the imaginary sensor starting position is set flat - it has no tilt wrt the (imaginary) optical axis.

Also note that an intrinsic Tait-Bryan rotation not only rotates the object but also the all the axis - which is useful as the latitude, hour angle and declination angles can be used directly. An extrinsic Tait-Bryan rotation rotates the object but leaves all the axis in the original position. The extrinsic form is less useful in this particular excercise as the input angles used need to take this into account.
 

Edited by Tonk, 16 April 2024 - 05:29 PM.

[han.k]

No - this is not 0.5 degree of flexing. After adjustments the technician has left the Gerd Neumann camera tilt adjuster with a mean 0.5 degree tilt around the X axis and a mean 0.03 degrees around the Y axis.

 

Sorry I'm totally lost. How does the technican know he has to set these tilt adapter values? "Mean tilt adapter position", does he adjust the tilt to a measured zero for each azimuth position?   Looking to your graph he seems to adjust the tilt adapter for about 0.15 degrees. If so how can the Nina measurements make any sense? You would expect a fixed/frozen tilt adapter position during the measurements.

[Tonk]

 

Sorry I'm totally lost.

Look we used ASTAP for 2 nights and a total of 5 hours. The technician is well experienced in using ASTAP and has used it on many scopes at the facility with success. With our FSQ106 he was defeated - ASTAP would give one result and then when the tilt was adjusted and best nulled (and even then it was never in the "good" zone) and then tested elsewhere in the sky ASTAP was reporting a poor tilt again. There was no solution to maintaining a flat sensor - the best solution seemed to be specific only to one place in the sky. The current tilt was just were he gave up as there was no sensible result forthcoming, we were all tired of it and frustrated at this point.
 

 

How does the technican know he has to set these tilt adapter values?

He didn't!           As I have explained enlessly now these are values we drew out of our subsequent analysis.

This was last October - its been like that ever since. So before we asked him to have another go and burn up even more of our money we (the team using the scopes) set out to see if we could analyse the problem in detail and do this remotely (to save money ). The technician recently suggested we try NINA as a means to measure the key corner displacement parameters we needed to model the sensor positions wrt to the gravity vector (note he had no idea that ASTAP had other methods for measuring tilt other than the octagon thing - he is the ASTAP user - not us).

For us doing the subsequent analysis NINA was nice to use as it was fully automated - it just takes a long time! It took a whole night to get three all-round-the-sky runs in. What we saw when the results were plotted was clear indications of flex and that this was reproducable - nothing random going on. It explained what was going on with our failure to level the sensor.

We have now done our full analysis - as reported here. I thought it was an interesting problem to recount as it might be useful to others.

We have now shown that we get more meaningful results if we reduce the ROI from 100% to 80% for starters due to the non-tilt optical aberrations in the corners peturbing results - very large sensors with tiny pixels apparently are not the best options for an FSQ! However at 80% ROI NINA shows us that the optical field curvature of the FSQ on the QHY600 sensor is entirely flat which makes things easier in the flex analysis. We have now shown our problems are entirely due to OTA flex and we can now direct the technician to look for the causes and attempt to improve things (tighten up the focuser being the only real option). This was one of the goals of the exercise.

Also we can now see from the analysis were the alt/az sweet spots are to best do the tilt adjustment (either near az 90 or az 270 at altitude 45 are good spots) i.e. to do the adjustments at the mean flex position - and not at one of the extremes of the flex. If we cant reduce the flex significantly then we have to ensure we dont image in the region around az 0 or 180 as this is where the flex is worse. This was the other goal of the exercise.

We are going to ask the technician to do another tilt adjustments at one of the alt/az sweet spots in the coming weeks. We will leave it to him to pick the most appropriate software tool that he is comfortable with to get this done with the least fuss.

Edited by Tonk, 17 April 2024 - 09:20 AM.

[Tonk]

I think I have explained this as best as I can now. If it inspires anyone to try and solve similar problems then I can help with the calculation details. Cheers!