Low Cost Pick and Place Machine

[eneuro]

You will need a way to rotate these parts if you want any kind of reliability.

It is much faster make small adjustments while we have part prepared to PnP, I mean very close to perfect fit, than do all this job in the fly, so yep small adjustment might be needed for final placement, but time needed for those tiny adjustment can be much smaller when very small angle need to be chamged, which can lead to huge speed improvements in the case fast PnP assembly head is used like those 3axis ultra fast pick & place robots:
Ultra Fast Pick & Place Robot - FANUC's New Three-Axis Delta Robot Packs Small Batteries


BTW: It weights less than your PnP machine, but for sure costs much more, I guess.

Anyway, I've ordered steell balls today from  reliable local bearings supplier, to create axis similar to those used in this ultra fast robot and we'll spot weld them on friday, I hope, to see how it works  and if it coulbe usable, so theres is some time to create 3D njumeric model of this thing to simulate complex 3D space positioning, based on dimensions of ordered steel balls and robot arms axis dimensions
It is a lot of fun trying to create such fast robot in DIY project 

Good luck, I want hit another PnP market with much cheaper PnP available for more electronics and CNC hobbysts from my neighbourhood and without any Kick(off)starter and similar company which earns close to 10% or more on someones ideas, instead people donate us from time to time via Bitcoin (cheap and safe moneys transfers quickly exchanged to real $es), so we are very motivated to deliver as fast as possible something usable and everybody are happy- we can save those 10% which had to else benefit doggy Indiegogo, etc, promoting sometimes stupid, useless ideas, which owners  claims they are smarter than... Nobel price winners  scientists

[mikeselectricstuff]

It is much faster make small adjustments while we have part prepared to PnP, I mean very close to perfect fit, than do all this job in the fly,

You can do the rotation during the time the part is moving from the vision location to the placement location, it shouldn't add any time for most placement positions.
I can't see how it can make sense to do it anywhere other than on the head.

[eneuro]

It is much faster make small adjustments while we have part prepared to PnP, I mean very close to perfect fit, than do all this job in the fly,

You can do the rotation during the time the part is moving from the vision location to the placement location, it shouldn't add any time for most placement positions.

I see more challenge making huge rotation during the time the part is moving, unless we have stored in CV memory perfect image when part was lying in the feeder or some kind of additional artificial background below taken part, since when part is moving fast there will unpredictable crappy background, so very difficult to make another correction photo while moving and during placement we have our PCB in the bacground, so not perfect conditions for image segmentation (we have to deal with many different PCB colours, printed text silk descriptions, etc.

I didn't tested this idea of feeder machine providing exactly rotated and aligned parts which are scheduled for PnP by assembly head unit, but there i no question about it, that those rotation could be made using another CV in parallel with assembly head doing mainly only pure PnP job.
I do not know if it works, but I will go this way in my PnP and we'll see how it works in practice... 

[mikeselectricstuff]

You misunderstand. You pick it, image it as soon as you can, then as you move to the place location, you do the rotation in time that you need to spend anyway moving to the place location. 

My Versatronics does this very efficiently - when the head moves fully up, it flips a mirror under the part at 45 degrees to a horizontal camera so it can image while moving.

[Kjelt]

My Versatronics does this very efficiently - when the head moves fully up, it flips a mirror under the part at 45 degrees to a horizontal camera so it can image while moving.

Wow do you have some video of this Mike?

[matseng]

My Versatronics does this very efficiently - when the head moves fully up, it flips a mirror under the part at 45 degrees to a horizontal camera so it can image while moving.

Wow do you have some video of this Mike?

He does :-) https://youtu.be/ZaqvEftw7aI?t=5m47s

[Kjelt]

He does :-) https://youtu.be/ZaqvEftw7aI?t=5m47s

Thanks! That is some big device that head wow.

[mikeselectricstuff]

He does :-) https://youtu.be/ZaqvEftw7aI?t=5m47s

Thanks! That is some big device that head wow.

The machine has origins in an earlier version that was designed to do CNC, so it's probably more heavily built than it needs to be.

[ProtoVoltaics]

That's a great video! Cool looking machine!

[eneuro]

You pick it, image it as soon as you can, then as you move to the place location, you do the rotation in time that you need to spend anyway moving to the place location. 

Anyway, it looks like there is no top view of part with PCB on the bottom and 45 degs mirror is used only to rotate part to final position?
I mean, when part is placed on PCB this machine has a chance to see how part touches PCB and part pins matches PCB pads or do it at blind based on memorized position (coordinates)?

Quite tricky mirror flipping-thanks for this tip, we'll see maybe it will be the only way to make final high precision rotation adjustment 

[mikeselectricstuff]

You pick it, image it as soon as you can, then as you move to the place location, you do the rotation in time that you need to spend anyway moving to the place location. 

Anyway, it looks like there is no top view of part with PCB on the bottom and 45 degs mirror is used only to rotate part to final position?
I mean, when part is placed on PCB this machine has a chance to see how part touches PCB and part pins matches PCB pads or do it at blind based on memorized position (coordinates)?

Quite tricky mirror flipping-thanks for this tip, we'll see maybe it will be the only way to make final high precision rotation adjustment 

Why would you want to downwards image the part or the PCB?
You can't image the PCB for alignment as the pads are covered in solder paste.
The only use for a downward camera is fiducials.
You could probably adapt the mirror arrangement to flip one of 2 ways to look down for fids, but cameras are cheap so you might as well have a dedicated one for that. 

[ProtoVoltaics]

Why would you want to downwards image the part or the PCB?
You can't image the PCB for alignment as the pads are covered in solder paste.
The only use for a downward camera is fiducials.
You could probably adapt the mirror arrangement to flip one of 2 ways to look down for fids, but cameras are cheap so you might as well have a dedicated one for that.

We are look downward looking vision to see the pads. Our vision sees pads with or without paste applied.

[eneuro]

The only use for a downward camera is fiducials.

When you design PCB eg. silk layout in the mind of to help PnP machine do its job faster, than pad covered by solder paste is not such a big deal 

Especially, when you can see invisible with.... thermal image cameras, with a little help of laser... or induction heater hidden in assebly head, even track covered by solder mask can be unveiled 
I'm designing my PnP not only for PCB assembly, while I'd like to use it too for automatic sorting and packaging pieces of wood so considering different options available.

Flipping mirror over picked part fits many requirements, since it reflects not only visible light but IR too 

[mikeselectricstuff]

Why would you want to downwards image the part or the PCB?
You can't image the PCB for alignment as the pads are covered in solder paste.
The only use for a downward camera is fiducials.
You could probably adapt the mirror arrangement to flip one of 2 ways to look down for fids, but cameras are cheap so you might as well have a dedicated one for that.

We are look downward looking vision to see the pads. Our vision sees pads with or without paste applied.

So how does it see through the paste to find the pad edge?

[ProtoVoltaics]

So how does it see through the paste to find the pad edge?

We'll unfortunately we aren't an open source project, but you can see us aligning over the placement areas in our videos. You will see the head move to the place location, and then move slightly again. This slight movement is the CV correcting for offsets.

[rx8pilot]

So how does it see through the paste to find the pad edge?

We'll unfortunately we aren't an open source project, but you can see us aligning over the placement areas in our videos. You will see the head move to the place location, and then move slightly again. This slight movement is the CV correcting for offsets.

Uhhhhhh....open source or not, there is no practical way to see through solder paste. Looking at bare pads is useless and looking at paste as an alignment aid is also useless. This is why any/all pro machines use fiducial marks instead of the component pads.

[mikeselectricstuff]

So how does it see through the paste to find the pad edge?

We'll unfortunately we aren't an open source project, but you can see us aligning over the placement areas in our videos. You will see the head move to the place location, and then move slightly again. This slight movement is the CV correcting for offsets.

Using vision to compensate for poorer accuracy of low-cost mechanics seems like a nice idea, but the paste is going to be a major limitation on the ultimate performance. If the paste print is offset, you still want to place on the pad centre, not the paste centre. And as for trying to do it with dispensed paste, good luck with that....
It is also likely to be a decreasingly viable solution as pad sizes decrease, which is exactly when you need better accuracy.
IMO a better approach to replacing expensive precision mechanics with cheaper electronics & software  would be to find a way to very accurately measure the head position, and use that in a closed-loop system. something like laser interferometry or structured light could perhaps be viable.
Maybe even an upward-looking camera imaging an overhead grid
 
 

[mikeselectricstuff]

We're also trying a NEMA 23 motor on the z-axis to see if there's any speed difference versus the old NEMA 17.

I'm not convinced steppers are the optimum answer for z and theta, as they are large and heavy, just what you don't want on something you want to move quickly & accurately.
Certainly for theta, the torque required is negligible, and a small gearmotor+encoder servo ought to be viable. The AMS magnetic rotary encoders might be worth a look. these look capable of better than 0.1 degree accuracy.

Similarly the Z axis, provided the weight is low, may be amenable to something a lot smaller an lighter than a stepper. Maybe rack & pinion or captive toothed belt, or band like they used to use on hard disks, plus a linear strip encoder.

[ProtoVoltaics]

We're also trying a NEMA 23 motor on the z-axis to see if there's any speed difference versus the old NEMA 17.

I'm not convinced steppers are the optimum answer for z and theta, as they are large and heavy, just what you don't want on something you want to move quickly & accurately.
Certainly for theta, the torque required is negligible, and a small gearmotor+encoder servo ought to be viable. The AMS magnetic rotary encoders might be worth a look. these look capable of better than 0.1 degree accuracy.

Similarly the Z axis, provided the weight is low, may be amenable to something a lot smaller an lighter than a stepper. Maybe rack & pinion or captive toothed belt, or band like they used to use on hard disks, plus a linear strip encoder.

You'll see some of these changes in our new video actually 

[ProtoVoltaics]

Yes, you will have to use a stencil for the paste. These are actually very accurate. Also, some of the errors the paste will correct in reflow.

We are testing the limitations of the machine week after next week, and tomorrow we are testing more with 01005's. We successfully placed our first 01005 today using our vision system.

I'll see if I can throw some solder paste down on it and place one. More videos to come!

[eneuro]

This is why any/all pro machines use fiducial marks instead of the component pads.

In this Fiducial Mark Design Guidelines we can find that not only global, but local fiducials marks can be used, but it will depend on density of parts on PCB, etc, so when I'll have at home my own PnP, than I can use for example silk layer to create local fiducials marks, to do paint them rather on solder mask, than recomended exposed copper, because of simply there can be no space between tracks or to close.

1.2.2. Local Fiducials
Fiducial marks used to locate the position of an individual component that
may require more precise location, such as a .020” pitch QFP.

1.5. All Fine Pitch components should have two local fiducials designed into the land
pattern of the component.

1.9. A minimum of two local fiducial marks are required for correction of translational
offsets (x and y position) and rotational offsets (theta position). This can be two
marks located diagonally opposed within the per
imeter of the land pattern.

So, we can have on PCB some hints printed for PnP machine and it can be local fiducial marks, but in practice I'll have them as mentioned rather not as standard plated copper at least 1mm, etc, but hidden on white text silk layer, instead of useless text descriptions of PCB parts, since this information is useless in automated assembly, but painting local fiducial marks instead can help assembly head make faster moves on long distances (less precision) and make high precision final slower movement based on those local fiducial marks, which could be eg..... part of lets say 1mm x 1mm grid printed in silk text layer on PCB 

Of course expensive PnP machines do not need this and for them probably global fiducial marks are enougth, but when I'm looking for custom cheap PnP machine, than I can take advantage of oadditional marks on PCB which will create pattern easy to recognize in assembly head CV which will help make more precise parts placements at decent speed 

Who cares how beautifull PCB is-the only thing matters for electrons to flow is copper layer, especially when such PCB is later hidden in epoxy enclosure, etc.

[ProtoVoltaics]

Who cares how beautifull PCB is-the only thing matters for electrons to flow is copper layer, especially when such PCB is later hidden in epoxy enclosure, etc.

As a supplier of such items, I care. 

[eneuro]

Who cares how beautifull PCB is-the only thing matters for electrons to flow is copper layer, especially when such PCB is later hidden in epoxy enclosure, etc.

As a supplier of such items, I care. 

Maybe I'm more interested in overall quality, rather than silk text descriptions, which are useless without circuit schematics.
In the age of augmented reality, by encoding in those many local fiducials marks PCB board hash (unique ID) one can download this PCB schematics on the fly and apply to image from visual camera in his phone/tablet, so no need to print anything more than copper layer and those marks...

I want my PCBs (product of my innovation- not items) without "Made in China" labels, but "Made in Europe" including assembly, so I do not care too much, if there will be any paint art on those PCBs....I will use mentioned augumented reality instead...and end customers will get enougth information without messing with China characters on their PCBs 

[rx8pilot]

I am not a seasoned expert, but on my machine the local fiducials are not necessary. So far, just using panel fiducials is enough to get very precise placement on fine pitch parts. If I was placing super high value FPGA's on a really big PCB, I may use them.

I cannot figure out why anyone would bother trying to use pads, especially if they have paste on them. Fids are so easy to place and reference and they offer all the geometric corrections you will need.

[mikeselectricstuff]

I am not a seasoned expert, but on my machine the local fiducials are not necessary. So far, just using panel fiducials is enough to get very precise placement on fine pitch parts. If I was placing super high value FPGA's on a really big PCB, I may use them.

I cannot figure out why anyone would bother trying to use pads, especially if they have paste on them. Fids are so easy to place and reference and they offer all the geometric corrections you will need.

The only reason would be if your mechanics weren't sufficiently repeatable to get enough accuracy for the placement position.
If it weren't for the paste, visioning the pad as part of a closed-loop system would be a neat way to allow the use of much cheaper and less accurate mechanics.