Calibration

[spiyda]

the next step would be to get a grid printed more accurately, ( the one I had was not 100% so wouldn't be a fair comparison of the files produced)

I think Mike suggested a way to get one printed.

Photoplots ( or maybe just get a PCB made ?)
I don't know  if typesetting films are still easily available (I used to use these in teh days when PCB makers would take transparencies), but that would probably be another option - they are typicallt about 1200dpi but not sure what the dimensional stability is like.

This is one possible source - my PCB places uses them and charges me £10 per layer for plots
http://www.phillipsdigital.co.uk/

I don't know much about printing but there may be another way

I don't know if it would be accurate enough, but my new laser could probably do something half the size needed.. 
Its accuracy is only as good as the toothed belts it uses.. 

but it would probably be better than the sheet I had made by a local printer   

My son has a lot of customers with big industrial lasers, some use ball-screws which would be a lot more accurate., I will ask him tomorrow.

I guess I could try different materials on my laser to see what the camera images best, then get a full size one made on the industrial laser.

[Mike]

I don't know if it would be accurate enough, but my new laser could probably do something half the size needed.. 
Its accuracy is only as good as the toothed belts it uses.. 

You'd probably want to make it do everything from the same direction to avoid errors from backlash, though move optimisation in the software may get in the way of that. Also, with a laser you're a bit limited with materials - obviously you need something with good stability.

[alanambrose]

>>> Yes the calibration grid plate keys to the two dowels...

I wonder whether the dowels stick through the grid plate and the grid is directly keyed on them. Maybe the 8mm reamed holes just take bushes to actual locate 6mm dowels and then longer 6mm dowels stick through the 7mm holes in the cal plate to key the film/print on? I wonder how important it is that the grid is located accurately with respect to the bed? Do you think that the matrix results of the cal are the errors between the arm 'thinks' the theoretical spots are calculated from geometry vs. the actual position where the camera is homed in on the spots?

>>> If I remember correctly, it starts a few dots in from the left side..  not far from the position the head waits at after finding the switches..

I just re-read the thread, and your notes suggest around here?

http://www.anagram.net/nuts/Versatronics/RVxS-grid%202.PNG

'spose the machine target accuracy is better than 1 thou (I did a quick check with a dial indicator and each step from the GUI is very close to 1 thou), that suggests a printing accuracy of 10x this would be about the minimum you would need, i.e. ~10K dpi? I know I'm being vague with accuracy, resolution etc here

I was thinking maybe laser-cut black mylar on a white background would do it. Or say this:

http://www.jd-photodata.co.uk/photo-tooling/low-resolution-film/28-x-32-photo-tool.html

28"x32" - they even do 6mm mounting holes! I guess the max size we actually need is 21"x26". Then laser the outline out of the photoplot (doesn't need to be accurate) to fit round the arm etc. Do we need the whole grid imaging as I'm surprised that anything other than the bed needs calibrating - it seems that the positions of the tools, feeders lanes etc are directly calibrated anyway? Also, the arm can't actually reach 1/2 of the grid.

spiyda - how does it walk round the grid - Y front to back (as it seems in your videos) on the spots it can reach then X left to right?

Any ideas why there seems to be two matrices stored in rvdata.dat - head and fid maybe? Lastly, any ideas on how we would we know if the cal is good? I seem to be good with the questions rather than the answers today

A.

p.s. here's a visualisation of the default cal data from blnkdat2.sys: http://www.anagram.net/nuts/Versatronics/RV1S-003%20Paul%20F.Mills.png

[Mike]

I was just contacted by David Clements, who expressed interest in selling his cal kit and providing info on the cal procedure.
No figure was mentioned, but if there's any serious interest in having a whip-round and making him an offer, please PM me.
I'm not interested in offering a cal service or making anything out of it, but I'd be happy to document it here and be custodion of a cal kit for loan to people wanting to cal their machines, or get any low-cost parts copied.
 

[spiyda]

If you are getting a cal grid printed, I would go at least one dot oversize in each direction...   you can always cut a bit off !

@ alan The way it walked around the grid must be programmed in somewhere.. its not as simply as y then x, it follows a preprogrammed path that didn't seem entirely logical but moves around the area getting to most of the extremities.

The start point is critical or it goes "Off piste" at some point in its travels...

but if you look through my thread, I don't think I was a million miles away...

certainly the smaller board camera mounted to the nozzle holder seemed to do the trick, and the 180 camera rotate and repeat is a clever way of correcting for camera alignment.

Having a working calibrated machine took the urgency out of taking the next step.. placing components makes money

It would be nice to know from David, just how close I got , or where my misinterpretation is!

Chris

[basemetal]

After a long absence from the forum I am now back.

We acquired some of the calibration kit from a sale of RV's that we purchased for spares. I also had some good discussions with DC regarding the calibration procedure which was explained to me in some detail.

We were missing the camera that aligned with the centre of the nozzle spigot - it is vital that this camera is aligned 100% coaxially with the nozzle spigot and plumb vertically.

Illumination is also critical. We were told we were missing two sets of three bulb spotlights, which were secured on the left and right side of the RV at the top. However, I see no reason why a circular illuminator could not be used.

Unfortunately, I was asked by someone if they could borrow my camera mount from the calibration set as they had to perform a calibration on a customers machine. The item was never returned - and I have never forgot about it.

In any event the most important item is the screen. Attached are some pictures of my screen as supplied initially by DC.

The base looks like Aluminium plate about 8 mm thick and powder coated white. The film is a photoplot secured with black electricians tape. There are two al bars underneath, one with dowel holes.

Some of the dots are yellow, I don't know why yet. There is a bunch of tape in one location that could signify a start point.

I will measure the sizes accurately and post shortly.

There was a lot of postulating about two sets of data points and if it could relate to the fiducial camera, and to be honest I do not believe it does. This is my understanding of the calibration data and how it relates to the real world, I could be wrong but I don't think I am and would argue the point with anyone:-

1)There are three horizontal 2D co-ordinate planes. The first is the real world plane which is the machine base or table. The second is the calibration grid co-ordinate plane - this is related to the real world plane by two dowel pins. Provided the calibration grid is square and aligned with the real world x and y axis when looking in plan view it should not matter if this is not exactly positioned as to adjust exactly the calibration would be an x and y amount. The third is the arm or nozzle tip plane. This is related to the calibration grid by a 2D set of data points (x' and y'), so for each dot on the cal grid scanned by the calibration camera there will be an offset of x' and y'. This would be related to the real world plane by (x+x',y+y').

2)I do not believe there is a need to calibrate the fiducial camera on the cal grid - even though I was told that it should be calibrated that way. Why do I say this? Well the fiducial camera is fixed to the same plane as the nozzle, so nozzle calibration will automatically calibrate the fiducial camera - the fiducial camera being placed a set distance from the centre of the nozzle, say x" and y". So this would be related to the real world plane by (x+x'+x",y+y'+y"). I have set the fiducial camera using the 'prestick' method as shown to me by DC with no problems whatsoever.

Of course when not calibrating there is another horizontal 2D co-ordinate plane, that being the PCB board.

I will start playing around with the calibration settings again and see what else I can find.

[spiyda]

That looks pretty much like what I ended up with...

regarding the camera, you are welcome to use the board camera setup I built that aligns with the nozzle spigot.

The calibration routine calls for the camera to be rotated 180 degrees about half way through the calibration process and this I think is to compensate for any errors in alignment with the spigot.. 

the circular illuminator around the camera lens certainly seemed to work just fine, the software recognised the dots over the whole screen area and the illumination will be a lot more even than buln spotlights !

The only Thing I was missing was an accurate screen...   

[basemetal]

I will look at fabricating something along the lines of your board camera setup if I cant recover the original.

I have had a little play with one of the RV's and also taken a look at some of the cal data files. I may have to eat my words on the fiducial camera setup grid - but I cant get my head around why you have to calibrate the fiducial camera on the grid. I also remember DC showing me a little trick in setting the fiducial camera with Prestik.

Spiyda, have you tried to run the grid calibration with the fiducial camera and if so how did it work out?

[trev]

Great to see some activity with this. If there is anything I can do to help, let me know.

Trev

[spiyda]

Spiyda, have you tried to run the grid calibration with the fiducial camera and if so how did it work out?

to be honest, I can't remember...  if I did, I will have mentioned it in the thread somewhere

[trev]

Anyone made any progress?

[ColinBarron]

Why does the software need intermediate plots and not just 3 or 4 accurate end points?   

[Mike]

RV camera cal info from alanambrose - added here to keep cal info together

[alanambrose]

Hi,

>>> We acquired some of the calibration kit from a sale of RV's that we purchased for spares.

Ah, I hadn't noticed that there was some progress with this. This is very encouraging

>>> We were missing the camera that aligned with the centre of the nozzle spigot - it is vital that this camera is aligned 100% coaxially with the nozzle spigot and plumb vertically.

if that helps, I created some cad and a 3d print to hold, what I assumed was another fid-style camera.

>>> In any event the most important item is the screen.

I was hung up on where the dot screen was located relative to the positioning dowels. Is it possible to measure that or post an image with the grid in position and a ruler showing the distances? There are a few logical positions and I couldn't figure out which made the most sense.

>>> I have set the fiducial camera using the 'prestick' method as shown to me by DC with no problems whatsoever.

Can you describe that in more detail, I would love to setup my fid camera properly?

>>> There was a lot of postulating about two sets of data points and if it could relate to the fiducial camera, and to be honest I do not believe it does.

This is my interpretation ...

Firstly, there are two clear matrices of positioning cal data in rvdata.dat. Secondly, looking at some general information about scara arms and algorithms - it seems to be a general strategy to (a) use geometry to get from desired Cartesian nozzle x,y on the bed to rotational a,b and then use (b) adjustments based on measured data to get a more accurate position. I'll see if I can locate again some articles I found on that. That points to a tactic of figuring out which square you're in on the bed and then using the cal values and 2d interpolation to get the accurate fix. Hence, you need some data showing how far off the real numbers are from the calculated ones. That points to using an accurate physical grid (i.e. the photoplot) and a cal routine to calculate the adjustments...

Yeah, since the fid camera is located a fixed distance from the nozzle it wouldn't be strictly necessary to have two adjustment matrices. However, if you've already written the cal routine code based on a camera mounted on the nozzle holder, it's perhaps the easiest course of action to just use the same code for the fid camera - particularly if they're the same camera model

Alan

[Mickyblueeyes]

Hi I am trying to make a camera attachment to fit onto the nozzle for calibration, I need a miniature camera but so far I haven't found one that is compatible with the RV video input, I tried one from Ebay listed as a 12V 380TVL PAL Black and white but although I can capture images to a pc with a composite video interface, the video card on the RV just shows a black screen. any Ideas.
Ebay listing No was 223439036512

Micky