Machsupport Forum
Mach Discussion => General Mach Discussion => Topic started by: xxtoni on January 06, 2014, 06:38:09 PM
-
Hello everyone,
This is only my second post on here but this is something that has been on my mind for over a year now and I've finally decided to see if something can be done about it or not so here goes.
First a little background.
I am in the signage industry and also do the majority of my CNC work as part of that. Having a CNC machine in this work really isn't an option anymore and we can't really imagine doing most of our work without one.
There is a problem though - of course.
A big part of the work that we do is made up of printing a graphic, laminating (applying) it to a hard substrate like foamex (PVC foam board) and then contour cutting it out to a shape, that usually looks something like this - (http://www.signsexpress.co.uk/media/img/product-service/7-197-point-of-sale.full.jpg)
The problem here is that we usually do the cut with a jigsaw manually because it is much faster and safer (for the workpiece) than trying to do it our CNC.
Obviously doing complex contour cuts on the CNC router would save a lot of time and just be all around much easier than doing it by hand but right now that usually entails that you make some kind of registration mark in the artwork during preparation for print. Then you would have to export the outline of the artwork, as well as the registration mark, DFX/DWG/whatever and use that to generate your g-code for cutting later on. Then you apply the print to the board substrate and put it on the CNC for cutting, and here's the rub. At this point your work piece has to be perfectly aligned with the machine so that when you set your zeros on the registration mark, which serves as a reference to where the artwork is, that it cuts exactly along the contour. Doing this once or twice isn't that challenging but having to do it all day long quickly becomes tedious.
Now if it could be somehow accomplished that a camera reads those registration marks and that then software adjusts the angle of the registration marks in the g-code accordingly to the ones that the camera read then this whole problem would be eliminated. Initially I just assumed that doing this was far too complex to be attempted by a hobbyist but lately I noticed that Mach3 seems to have video capabilities and based on Tweakie's replies on a thread that someone else made to this question it seemed to me that it was implied that this wouldn't be too hard to do.
If all of this was too convoluted to understand I apologise but this video should explain in a few seconds what I tried to explain in 500 words here
http://www.youtube.com/watch?v=s3MeHLluzC0
(tried embedding this but didn't work)
Systems like this certainly exist but they cost thousands to tens of thousands of dollars which is not viable for me.
I do hope that someone with deeper understanding of these things can shed some light if doing this is possible or not.
In any case I'm looking forward to the replies.
-
The angle of rotation and linear scale aspects are no problem for Mach3 to resolve (as I think I demonstrated in my short video) but, as yet, there is no software to make the camera recognition of the registration marks automatic, it has to be done manually.
As I always say; everything is possible it just requires time and effort.
Tweakie.
-
FWIW,
Maybe consider changing the sequence or manor in which your do your work.
IE. Mount the board with print attached anyway you wish on the router table. Then use CopyCat along with a cheap camera to trace out the outline and generate the gcode pathing
for the cutting. Using an Xbox 360 controller to poistion the camera cross hairs will make tracing out the outline quicker and easier. After a little practice you will find that you can do this quite easily and rapidly, certainly not as fast as what you show in the video. Just depends on how accurately you want the cutting to be ( it can be very accurate).
I would say to trace out the top of one panel you show in the pic wouldn't take more than 15 minutes. For the other duplicate panels, you already have the gcode, so use CopyCat to
find the work offset required or even rotation, minor gcode change to the first program. No DXF, etc!
For a business may not not be the cat's meow, but it is economical, only you can do the math for what makes sense.
May be worth investigating. You will find a manual for CopyCat in Member's Doc's.
http://www.machsupport.com/forum/index.php?action=dlattach;topic=10498.0;attach=13710
RICH
-
First of all thank you both for your replies.
I think that the simplest and fastest solution may be to add a certain amount of bleed to the print so that you have some more wiggle room but even then you don't want to go too deep into the bleed cause you want as little excess color around the print itself.
Anyway I think that having a pair of laser crosshairs on the CNC head would be the fastest solution. In that case I would just export the registration marker and an outline of the work piece to g-code and then when adding the workpiece to the CNC table I could use the laser crosshairs to position the workpiece at an perfect angle then I could just place the bit onto the middle of the registration marker, set that as the zeros and start routing.
Still having a camera system would be easier and faster.
I don't have a programing background but how hard would it be to make a piece of software that would take a video, look for the registration markers based on the g-code (or dxf or whatever) and then compare their angle to the g-code and align to the g-code to what it saw ?
-
I don't have a programing background but how hard would it be to make a piece of software that would take a video, look for the registration markers based on the g-code (or dxf or whatever) and then compare their angle to the g-code and align to the g-code to what it saw ?
Because the machine’s camera is unlikely to be able to view the whole job in one frame and it would be a waste of machine hours scanning the total work area just looking for the markers my suggestion would be…
A software program (plugin) to view an on-screen display of the original artwork where pixel size versus steps per unit is known – using the mouse, the registration marker positions are located and stored, their position is then translated to the necessary X-Y co-ordinates for the machine to move the camera / machine head to each selected marker. The camera then centres each registration mark in turn or is used to determine the difference in the pixels position (not sure just how to do this yet but there must be a solution) and the difference between actual position and original image position of each marker is calculated. This data is then used to set the Mach3 X and Y scale factors (if necessary) and ‘local system rotate’.
It would be challenging project to do but, in my opinion, it is entirely possible.
Tweakie.
-
I wrote a plugin years ago, (among other things) I use for alignment calculation. Gives me lots of data.
Setting the first location to 0,0 and then going to the second location I can see the angle to set rotation axis G68 to.
I first did it with a macro pump then went on to make it a plugin to add all 6 axis.
So yes anything (for the most part) is possible with Mach. Its just what are you willing to live with.
Crude but might be effective for you;
You could use the camera find the first reg mark, take a small pin to pivot on that reg mark, set axis's to 0,0 and then rotate the whole image manually. Moving in only one axis use the camera to align the second reg mark. Would work good with Vacuum tables.
-
Did you post the plugin somewhere?
me wants, me wants!! ;D
I look for it on your website.
Dude,
That's some serious wood carving you been doing!
Sent you an email.
-
I don't have a programing background but how hard would it be to make a piece of software that would take a video, look for the registration markers based on the g-code (or dxf or whatever) and then compare their angle to the g-code and align to the g-code to what it saw ?
Because the machine’s camera is unlikely to be able to view the whole job in one frame and it would be a waste of machine hours scanning the total work area just looking for the markers my suggestion would be…
A software program (plugin) to view an on-screen display of the original artwork where pixel size versus steps per unit is known – using the mouse, the registration marker positions are located and stored, their position is then translated to the necessary X-Y co-ordinates for the machine to move the camera / machine head to each selected marker. The camera then centres each registration mark in turn or is used to determine the difference in the pixels position (not sure just how to do this yet but there must be a solution) and the difference between actual position and original image position of each marker is calculated. This data is then used to set the Mach3 X and Y scale factors (if necessary) and ‘local system rotate’.
It would be challenging project to do but, in my opinion, it is entirely possible.
Tweakie.
I'll do some research to see if there is anything similar in the "open" out there but I don't code so the chances of pulling this off at this point are slim.
Besides this isn't something that needs doing right now, we plan on using it on a new machine which we don't even have yet but I thought if I could figure it all out by the time we get the new machine I could have it implemented shortly afterwards and we could start using the system relatively quickly.
At this point I think it would be easier to just create the design with a + shaped reg mark, a decent amount of bleed, do the print and export that reg mark and an outline of the print to a dxf, handle it in the cam software, set the machine zero of the work in the middle of the + mark and then, possibly, use a pair of lasers to perfectly align the substrate so that the X and Y are straight. Then simply set the zeros in Mach3 to the middle of the + mark and that should be precise enough to get it to cut out properly.
The camera setup is certainly something to keep thinking about though, not giving up on it just yet but this is more realistic for now.
-
Not trying to discourage your efforts to find a programming solution, but if I were doing mass production of something like this, here is what I do:
Mount a spoil board onto the table, and use a drill to locate your registration marks on the substrate. Install alignment pins into this first piece of substrate, and into the spoil board. Run your gcode to cut out your profile of this first piece. Remove the piece, and mount the second (and each subsequent piece) onto the alignment pins of your spoil board (after drilling the holes at your registration marks of the piece), and run the gcode again. Sometimes, the simplest solution to overcoming technical limitations can be achieved by just looking at the actual task at hand. (NOTE: locate these registration marks in an area of the substrate that is to be removed, when cutting the profile- this will eliminate the drilled holes from the finished product!) Hope this helps.
-
This is how I would contour cut something. You place the print on cnc table anyway you want. Then you jog to first registration mark (bottom left on computer) and set X0Y0. Then jog to next registration mark and write down XY coordinates. After that you calculate the angle (arctan(X/Y)) and use g code G68 for coordinate system rotation.
I hope you can understand, I'm not good in English :)
-
Pi=4*Atn(1)
X = GetDro (0)
Y = GetDro (1)
V=X/Y
BaseAngle = Atn(V)*(180/Pi)
Angle = 90-BaseAngle 'G68 angle for Quadrant 1
-
So I've been thinking.
Optics may be too big a bite for now but how about something like this:
http://www.youtube.com/watch?v=bSPeiQl2k10
One would print 4 (or so) cross-marks at each corner of the print to be cut, manually go to #1 and in the software sort of zero there, then go to #2 and "zero" there as well and doing so for all 4 of them. Now the software has a reference points compared to the original toolpaths and can rotate and scale it accordingly, just like in the Zund example above.
How hard would this be to implement ? It's a bit more tedious to do manually but it's alright.
On a different front I've also found a company in the Netherlands that offers a vision system:
http://cnccamera.nl/src-gen/products.html
I've sent them an email and hopefully their system will be in an acceptable price-range.
-
A visual location method and local system rotate was discussed in a earlier thread http://www.machsupport.com/forum/index.php/topic,22932.0.html
It is not difficult nor expensive to implement.
Tweakie.
-
A visual location method and local system rotate was discussed in a earlier thread http://www.machsupport.com/forum/index.php/topic,22932.0.html
It is not difficult nor expensive to implement.
Tweakie.
Thanks for the quick reply.
I'll explore it.
-
Guess it's time to start learning VB and figure out a way to implement this.
This is not a bad solution for now, in the future doing it with a visual camera system may be possible but for now this will do and is more than enough to keep my busy.
If anyone can point me in the right direction on how I would go about implementing this I would appreciate it.
-
Here is a thread relating to the laser cross hair pointer http://www.machsupport.com/forum/index.php/topic,10841.0.html
Tweakie.
-
Here is a thread relating to the laser cross hair pointer http://www.machsupport.com/forum/index.php/topic,10841.0.html
Tweakie.
Yea I already read that and it's certainly something I want to add.
A question though, would I absolutely need the laser pointer ?
I'm asking cause if I make a cross mark and zero the tool in the dead middle of it it should be accurate enough, is there a need for the pointer for this talk alone ?
I mean the zund has a dot laser pointed and I don't really need that much either, I can just lower the bit itself into the middle of the cross and that should do it or have I missed something ?
-
Another question, I'm trying to wrap this around my head...if the distortion is non-linear how would one go about performing the scaling so that the modified toolpaths match the drawing.
Zund iCut and others somehow do this perfectly based on the registration markers alone but I'm having a hard time figuring out how if the distortion is non-linear.
-
Another question, I'm trying to wrap this around my head...if the distortion is non-linear how would one go about performing the scaling so that the modified toolpaths match the drawing.
Zund iCut and others somehow do this perfectly based on the registration markers alone but I'm having a hard time figuring out how if the distortion is non-linear.
We did this before (http://www.machsupport.com/forum/index.php/topic,23758.0.html) and the whole thing hinged on "non-linear". Unfortunately the OP didn't respond so we kinda dropped it.
-
You could certainly do it without the laser pointer and just by using the tool point if you wish.
Only linear distortion (scale) is easily catered for at run-time.
Tweakie.
-
Am I correct in thinking that the non-linear distortion is introduced when the print is affixed to the substrate? i.e. every piece will be "differently non-linear"?
OR
Is the non-linearity introduced at the print stage i.e. A batch will be non-linearly distorted with respect to the original but the individuals within a batch will be identically distorted?
-
Ian,
I would very much like to pick up where you left it at that conversation and I promise I won't disappear any day soon.
To address the question of the distortion here's a more in-depth view.
The technology that we use, and most others do is solvent inkjet printing. What basically happens in the printing process is that a piezo-electronic printhead shoots tiny droplets of ink pigment in an acetone solvent. The solvent "burns" into the material and leaves sort of a relief into which the pigment gets into and gets stuck there.
The printing process itself doesn't necessarily stretch the material, we are talking about self-adhesive vinyl here since it will be applied to a rigid substrate, but it leaves it susceptible to the stretching during application since the vinyl is much softer now.
If you leave the vinyl to gas off for 24 hours or so the ink will dry, the solvent gases will fume out and the stretching effect from the solvent is mostly gone.
The other stretching though comes from the mechanical forces that it endures during application to the rigid substrate.
If you have never seen solvent vinyl without the liner on it is almost rubery.
This is what the application looks like - http://www.youtube.com/watch?v=-u4TxTBU80g#t=51
As you can see the vinyl is quite stretchy so the distortion really depends on the direction you are applying the vinyl in. There may be some distortion in other directions as well but I think that it would mostly go in the direction of the application.
Now in this industry when you are doing serial cutting either via punching as is done in offset or via hand you leave some bleed. Bleed is basically some extra space or color around the image which is a tolerance so that if the cutting misses a bit that you don't get a white part there.
This image explains it perfectly:
(http://www.print-print.co.uk/blog/wp-content/uploads/2009/05/print-bleed.jpg)
To see just how much of a real-world problem the distortion really is tomorrow I'm gonna do a few prints, apply them straight out of the printer, ungassed, to a substrate and try cutting them based on a single cross-marker and then we can proceed from there.
Am I correct in thinking that the non-linear distortion is introduced when the print is affixed to the substrate? i.e. every piece will be "differently non-linear"?
OR
Is the non-linearity introduced at the print stage i.e. A batch will be non-linearly distorted with respect to the original but the individuals within a batch will be identically distorted?
The majority of the distortion would happen during application but I have reason to believe that either each or at least some of the pieces would be differently distorted. The reason for this assumption is that the distortion (stretching) is a result of mechanical forces applied on the vinyl during application. So for example you could apply some parts with less force and you could maybe pull a bit harder on a certain part and that could cause the distortion to increase on that part but not on the others.
Here is one thing I don't understand though, optical registration of print marks and cutting based on them is nothing novel in our industry. Pretty much any plotter cutter you buy has this option, even those shabby Chinese ones so it can't be that hard...I mean we are talking about a $500 or so and they cut the printed vinyl without fail every single time.
Here is an example of how that works - http://www.youtube.com/watch?v=jV1v8dRdsn4#t=158
It prints 4 of these marks (the black dots) total, one on each corter of the print then the cut head scans each dot and based on those it starts contour cutting the print based on the vectors you defined in your design software.
Same goes for the Zund and iCut system. You have to print these reg marks out and the camera scans these and based on them it cuts the print, no matter the distortion.
What I can't understand is how do they account for the distortion based on those registration marks ?
Demonstration of how the iCut system works - http://www.youtube.com/watch?v=s3MeHLluzC0
An even more impressive example, an iCut system on a laser cutter - http://www.youtube.com/watch?v=F4OLzJsakag
That must be less than a millimeter in precision and from the description:
"Printed signs, displays or POS materials can be cut on an individual basis using laser technology and i-cut®. The cutting line therefore follows the exact printed specification. With other methods, slight distortions of the printed design create unsatisfactory results. However, i-cut® is an intelligent, easy to use system which recognizes any distortions in the printed design. No matter whether it is a linear or non-linear distortion or a rotation, the cutting path is adjusted automatically and dynamically. The cutting lines therefore always perfectly match the printed design on both flexible and rigid materials."
EDIT: Added the iCut demo
-
I didn't have time to perform the cut tests today but will do so early tomorrow so it doesn't get away.
Tweakie (hopefully you're still following this) the 2 laser setup to get a crosshair seems a bit daunting to me because of the need to angle them and well...use 2 lasers.
Would a single laser like this work or am I missing something important ? - http://www.ebay.com/itm/New-650nm-5mW-Laser-Cross-Module-Diode-w-driver-/260839502067?pt=LH_DefaultDomain_0&hash=item3cbb3ef0f3
-
Nevermind I mixed something up.
Just re-read Tweakie's post. I thought that you used two lasers for getting the cross instead of one. Mea Culpa.
-
Why don't you use microscope camera, it is much easier to pin-point marks?
-
Look at the sheet cam forum. It looks like Les is working on an idea that will do this. It uses a camera to trace edges and the exports that.
Steve
Sent from my iPhone using Tapatalk
-
Resurrecting a dead thread here...
We have several newer CNC machines and one old multicam which I converted to mach 3 when the factory controller electronics crapped out. As we use it less and less, we have decided to make it a print/cut only machine. We already do this successfully on routed substrates in conjunction with our flatbed printer with very accurate results. We have print/cut thousands of dibond boards by just dropping them on the bed and using the stops we machined in as fences to align the board. Fast, easy and accurate.
I just purchased a Donek drag knife and want to start cutting our roll substrates on the machine. When un-mounted, the distortion is fairly linear over a distance of less than 12 feet (our bed is 5'x10' anyways). So the stretch and rotate work arounds that already exist will work just fine. What we really want to do is automate the registration mark detection vs using a web cam or pointer.
Like the OP, we have a Mimaki vinyl cutter with registration mark detection which works very well. The system works like this: It has a light/dark edge detecting camera/sensor and two "L" shaped marks of between 10mm and 40mm long and 1-3mm wide are printed at the corners of the print. Mind you, all this system does is address skew (rotation) and stretch. The operator positions the sensor in the corner of the L and begins the automatic routine. The camera scans approximately 80mm left and upon hitting the edge of the line, returns to its starting point and then travels 80mm up to find the horizontal line. This gives the machine an origin corner so to speak. It then travels along the X axis until it hits the edge of the next horizontal line for the upper corner mark. It relies on the marks length and the operator loading the media vaguely straight to insure that it intercepts it. If the media is too skewed, it will miss the mark and fail, but you have to be incredibly lazy to load it that far off. Upon finding the horizontal edge, it advances a few mm forward and then finds the vertical edge and thus is able to divine the top corner inline with its new found origin corner (confusing I know, but when you see it it is simple). It returns to the first corner and runs the same routine on the y axis. Ultimately, by finding the coordinates of the 3 outer corners of the file, it can calculate linear stretch and skew and the job begins. This system is pretty bullet proof.
I think many of us could benefit from a collaborative effort to get a system like this working in mach 3. The commercial turn key options are obnoxiously expensive for the quality of the machines. Compare what you get with an Esko or Zund vs a Biesse for the same money.
Here are my thoughts on potential solutions:
Create a stand alone arduino based edge detecting camera. When edges are located, have it simply go +5 feeding an input to your mach 3 interface (we use one the CNC 4 pc boards and its very easy to IO from, I think it's the C32?). There is a healthy repository of open source machine vision type projects in the arduino community and it would entirely divorce the edge finding logic from the PC piece and mach 3. Thus, all it would take is a simple macro that responded to the input from the arduino based edge finding sensor.
My other thought is to run in parallel on the PC a basic piece of machine vision software and use a sort of virtual IO to do the same as above.
Am I just crazy or is this imminently doable?
-
I think it is a brilliant suggestion Eforer and it could become a very useful plugin for Mach3.
Tweakie.
-
I should have the sensor mounted up and on the machine early next week. Does anyone want to help me with the macro?
-
I have the sensor board working well, meaningful output difference as it passes over a reg mark and I am lighting up an output based on some simple evaluation of the incoming data. I'm ready to start playing with a Macro. Is there a way to make a sort of development environment so that I don't have to mess about on my mach 3 installation out in the machine shop?
-
Doh! I just found this sensor which completely eliminates the need for the Arduino. This provides a stand alone solution for registration mark detection. Now we can just work on the VB! This should be easy to wire in as a probe. Does anyone have some examples of calculating an arc tan in VB and setting the G68 coordinate system rotation from a script?
-
ATN (arc tangent) is covered in the "Cypress Enable for Mach" manual , page 44
-
How would i then set the g68?
-
I'd provided how to set the g68 in the file that was posted to your other thread.
http://www.machsupport.com/forum/index.php/topic,26758.msg189574.html#msg189574
Call SetOemDro (118,NewAngle)
OEMDRO was added to screen with machscreen. To show me what it was set too. As if it is anything other than zero the Axis X Y DRO's values turn red.
The sensor you used before does not need the Arduino to work, it just needed a 12 volt DC supply, as it incorporated a SPDT relay.
You could then provided the 5 volt source at the relay.
Though it does a poor job of detecting a thin line and lighting the area has to be extremely controlled to be accurate and repeatable.
Using a wide line to get it to work well, makes me suspect its precision at the ability to find the true corner of the scanned probing routine.
-
i made a camera that finds the registration marks and adjusts for distortion and rotation
-
Hi guys! Any news on this? Does anyone have a registration system up and running?
-
because it was for my work and i spent work time i can not give a lot of information.
attached is the script i used for finding the dots and rotating and stretching the file.
hope it is a help
-
Thank you! I really appreciate it! :-)
-
Judging from your post, this solution that my team and I have been working on probably fits your criteria: https://www.youtube.com/watch?v=dqFpI4t21Z8
Effectively, we use registration marks to re-align the generated G-code, allowing for a proper cut despite the material being rotated or distorted. Let me know if you have any questions, I'd be happy to chat about this feature.