[Nails]

alphatroniqu - Nice work.

I'd like to see some close-ups of the tool changer mechanism, and details of the operation, as well as the tools themselves. Can you help me with this?

[Eclipze]

Single sided X rail, single Y rail, X480 Y90 Z20 A360, 2 steppers, 2 RC servos, ATC planned, 30 8mm feeder banks (can be changed for trays etc...), step resolution 0.0125mm. Planning to be controlled by Mach3.

;-)

[alphatroniqu]

Hi Nails

i will try to explain ATC working whit my limited English

the bit hold on machine whit metal ball and slip rig whit spring it hold bit
by push ball into a grove on the bit

the atc was cylinder in 2 part whit spring and air cylinder it also
use 3 metal ball for grip bit and remove it from machine head

for attach a bit cylinder raise and head go down the ATC metal cylinder push
on machine head ring the the machine head was raised so ring slip back on position and grip bit

for resale it drop the ATC cylinder air pressure drop spiring hold ball on atc bit
then exterior tune lower whit grabber bit

this was one of most complex part of that machine and surely the harder to duplicate

[RomanLini]

I put your details in the list Eclipze.

Everybody, please copy the whole list and put your own details in, that way you can update it anytime if your build changes. It's got quite a few people on it now!
--------------------------------
Eclipze; Single sided X rail, single Y rail, X480 Y90 Z20 A360, 2 steppers, 2 RC servos, ATC planned, 30 8mm feeder banks (can be changed for trays etc...), step resolution 0.0125mm. Planning to be controlled by Mach3.
RobinHewitt;
Cnczoner;
RomanLini; Single sided X rail, raised Y gantry and support rail, X420 Y180 Z28 A180', 2 steppers+screws, 3 RC servos, 5 tool ATC, head pin tape feed, step resolution 0.008 mm, custom PIC controller custom PC software.
Neilw20;
Nails;
Nisma?
Alphatroniqu;
--------------------------------

Are you really going for only 90mm Y axis travel Eclipze? That seems small and limits all your PCBs to 90mm width unless you turn them around...

That Zevantech is a good photo Alphatroniqu, it's good to see the insides of one.

Nails- I think they used a red laser de-focused to put a lot of light on a small area, and out of red or green visible lasers the red was better as it reflects less from the green coloured PCB. These days there are a lot of options for superbright LED colours.

It's nice to see photos too!

[Eclipze]

Oops... Made a mistake with the Y-Axis, it's ~170mm. I might increase it to 210mm before I start cutting material, however I want to get a better idea on the limits of using a single X-axis. If it will do 210mm easily, I'II increase. If it's not doing well with 170mm, then I'II add a supporting rail and go 210mm anyway.


--------------------------------
Eclipze; Single sided X rail, single Y rail, X480 Y170 Z20 A360, 2 steppers, 2 RC servos, ATC planned, 30 8mm feeder banks (can be changed for trays etc...), step resolution 0.0125mm. Planning to be controlled by Mach3.
RobinHewitt;
Cnczoner;
RomanLini; Single sided X rail, raised Y gantry and support rail, X420 Y180 Z28 A180', 2 steppers+screws, 3 RC servos, 5 tool ATC, head pin tape feed, step resolution 0.008 mm, custom PIC controller custom PC software.
Neilw20;
Nails;
Nisma?
Alphatroniqu;
--------------------------------

[RomanLini]

Cool, 170-210 sounds a lot more sensible.

I get a little bit of skew (0.3mm?) with the head at the far end of the Y travel from both friction (on the support rail) and inertia. The X linear bearing has not been properly shimmed yet so that will fix maybe half of that or better.

Then being a software guy I can just compensate in software with a Y-graded X axis backlash if that makes sense.

Either way I'm not worried as all the small precise parts are on one side of the machine (420mm of small tapes) and the far side fo the machine where the most error is found will have the large components, things like electro caps and large FETs etc, they will be ok to place within a precision of 0.5mm anyway even if all the backlash compensation etc wasn't there.

Hey! I think you might have 3 stepper motors too?

[Eclipze]

Dammit... yeah, there are 3 stepper hahaaa.... Forgot about the tiny one for the A-axis.

What I've done with the new model is to use two carriages (butted close together) on the rail instead of one. I believe this will further increase the rigidity, as well as help minimize the movement that a single carriage could allow. Particularly important for a single rail design approach.

[nisma]

.

[RomanLini]

Eclipze- Yep I considered that option too, running 2 carriage bearings joined together to reduce the slop and skew forces. But it would have cost 40mm travel as I already had the rail, and there are a couple of differences like on my machine most of the movement will be Y axis as the tapes are at each side of the machine so I can live with the lower speeds etc of leadscrews and run things a bit more snug.

If I was to build a "perfect" machine later it would probably be with belt drives for speed (like you did) but with reducer belt/pulley too to increase the stepper torque and reduce motor resonance. But (as always) the perfect system starts to get more complex and cost a lot more.

Most of my parts placement will be traveling less than 80mm so I can live the leadscrews.

Did you see that Zevantech photo in post #320? It looks like it just uses a round bronze bush on a steel shaft, that's about as nasty as it gets.

[Eclipze]

My small CNC, Widgetmaster, runs a bronze bushing for X and Z rails. It was nasty and use to bind... until I started using the INOX lubricant. That stuff turned nasty into reliable :-O

If I wasn't using a 1/10 stepper driver, I'd be using a pulley reducer. Using the 1/10 driver and 10 tooth pully gives a good level of precision and probably more speed than will be practical. But if the step resonance is a problem, I'II be looking at a reducer and possibly a damper.

[dandumit]

Only alphatronique said what kind of vision system will use.
No one else .
I have a x-y gantry and I would like to transform it in a pick and place machine. But i have no ideea how should I implement shape recognition for components.
For what it's used second camera - vision from below.
Would someone explain more what options are available ?

[dandumit]

A guy from Lithuania seems to sell some machines. However those have a look like home built
???? (????????? ??????????????? ??????? ?????????????? ??? -400)
May be inspiring since pictures are pretty clear/big

[alphatroniqu]

Hi

this above machine look like\similar to a old MDC ECM-93

ok just for clarify the zevatech use Bering not bushing

i also add some detailed photo

also i use mena34/24 5 phase stepper motor(PH566HA) and (PH564A) 500 step/turn
5phase was for better torque less resonace
pulley was made for that each step = 0.02500mm on the machine so this ok
for 0402 and all ic

Best regard
Marc L.
Alphatronique inc.

[nisma]

Vision, it´s depends. Until yet i have used the usual basic algorithm, histogram and rotoation inside a ROI. Time for displacement and rotation detection is under 1-2ms.
Now i´m thinking of switching alghoritm, or at least using different algorithm for different
things. The downside is, that it needs 53-96ms for detecting one thing, and probably i
need detecting two or three things, so time goes up from 300ms to 800ms allowing feature
detection like reading the orientation mark and other things without doing patten matching
with a golden image. Being more flexible, the downside it needs a factor of at least 100 in time. It always need a database and that´s is the major limit of universal usage because everyone want it different. Annother thing is, you need 2-3 different light types
for illumination and yes, illumination is the key that make the solution win or fail.
Annother thing is, that usual things in image processing, like building a histogram, making
normalisation, computing mean value, making segmentation as example, all this thing
are removed for speed reasons, because most things remains constant and so you can
reduce make most of the required operation with constants eliminating the steps for
gathering info about the arguments like threashold level, median color and so on.
This time saving options and machine (illumination) dependant algorithms makes a solution non portable.

[dandumit]

Hello Nisma,
Thank you for reply ! I understand that it's no solution that could be universal but I need / request a little guidance.
I have played until now with an usb microscope (probably it's poor quality) and OpenCV.
I have requested a trial for Halcon.
Unfortunately I didn't managed to detect a smd capacitor and find it's center and rotation.
If you can tell what library have you used and what methods would be great.
Kind Regards,
DAniel

[RomanLini]

For vision, I just planned to place blind. Mainly it will be 0805 parts and SOIC so they can just be picked from the correct coords and placed. 0805 sit in the tape within reasonable tolerance (maybe 20% of the component size) so I figure for production it is better to place 1 part per second blind then 1 part per 2 seconds with a fraction more precision.

And for the occasional TQFP or QFN I will just use a straightening square or even a cone shaped square hole to mechanically auto straighten the part and position it on the nozzle. The bulk of the placement will be discretes like 0805 and SOT23. I'm confident enough after using my home made CNC milling machine with plastic leadnuts that it can put itself every time within 0.02mm of where it needs to be and that is blind.

Alphatroniqu- Is that actually a bearing with balls? It still looks to me like a plain bushing with grease seals on the ends. And wouldn't it be better to join the 2 (Y and J) belts with one elevated shaft instead of all those other belts?

Having said all that about placing blind, my $7 USB webcam arrived today so I will play with that a bit...

Nisma- Why is that image recognition system so slow? Surely it only needs one frame and then a little bit of image processing on the PC?

[atlab]

A guy from Lithuania seems to sell some machines...

From Belarus, not from Lithuania. Very old design.
Precision X,Y - 0.15mm, rotation (angle) step - 45 deg.

[nisma]

For Vision, as example this is a example of locating a component.
The yellow points are hand draw for the first image and the sw should check them for
component orientation. With this lines, you know the component location
(centeroid data), the rotation and dimension. If the component don´t feature
the half circle cutout for the top location, you could use the hough circular trasformation in order to detect the circle mark. That´s the reason why i have
used 50 and 200 for the canny, because in this case, the line and circle extraction can share the same canny trasformation.
PS. This use the high precision routines, the lower precision are faster.
The other system is histrogram based and more faster.

Resuming, vision need
1) Knowledge of what to search (eliminate classification) = database
2) Good source image (reduce effort to extract wanted feature , ie fast) = illumination + background (color)
3) vision algorithm (histogram + rotation <10ms or hough transformation <100ms as example)

[dandumit]

Thank you for teaching. Indeed illumination has a huge impact.
I feel myself recharged : I will play again with OpenCV.

Image of your pcb it's very nice. I have tried a lot of webcams and cheap lenses => Result it's very poor. Would you please tell us what camera and what lenses are you using ?

Daniel

[alphatroniqu]

Hi RomanLini

it join belt on the base for avoid to have shaft on front of machine that
will conflict whit feeder and operator vision ,it not easy to see bit,part during
feeder pick point teaching

also i place 0603 blind on dense FPGA board whit no problem 20% off on a 2 pin passive
part was not a problem, solder surface tension will fix. vision was mandatory for 2 thing
1-) fix machine lack of repeatability this was become common on low end commercial machine 2-)QFN,TQFP,BGA or anything that have pin on the 4 side since it need 0.18 deg precession on rotation and ~0.1mm x,y accuracy