Bringing PCB assembly in-house with Dima Optimat?

[mrpackethead]

Is the machine actually out of calibration?

[Styno]

The PnP was probably setup while not level before transport so when we set it up level in the shop the pcb transport rails were slightly twisted which locked the axles. We noticed this only when we tried to load a pcb, it burned out the pcb belt drive motor.

See the quote above. Also, the machine was originally configured specifically for wide pcb's so only the front of the machine could hold feeders. The supplier has added a 'nest' (cassette) mounting beam at the back of the machine for us. So I'm not sure if the machine is really out of calibration but I reckon it won't hurt to do a full calibration when it's in working order again.

The calibration manual certainly appears to be written for the end-user. It's detailed and often talks about 'you' and 'the user', nowhere does it say that the process should be performed by a trained technician.
E.g. "This manual is written to help you calibrate the Optimat." and "To support the user in this calibration, a software program is provided, which guides the user through a menu, so that all items are calibrated in the correct order".

And some of the calibration tools as mentioned in the calibration manual are permanently fixed to the machine (e.g. some fiducials on the pcb transport rails and feeders). I suspect every machine came with a calibration set.

[Styno]

Today was not particularly a good day, we discovered more problems with the PnP.
- A segment in the fiducial camera lighting ring isn’t working. I suspect this to be an easy fix.
- It looks like one or two power transistors on the I/O VME card for the pcb conveyor burned out, as well as a darlington array ic. The card will be repaired or replaced.
- The new cassette base plate is unfortunately not made very accurately. The cassettes do not line up with the registration holes in the coveyor beam and, when installed with some force anyway, bend the conveyor beam. This bending is also the likely cause of the locked belt drive shaft earlier and the subsequently burned out motor/IO-card.

So more delays unfortunately, but tomorrow I’ll try to backup the software (DOS) and to start building a package and component library.

[mrpackethead]

^^ lesson for everyone whos buying a PNP ( new or otherwise ).. Dont' commit to a job for it, until you have it all up and running and commissioned.     It will bite you .

[Styno]

Heh, lucky we don’t have a big deadline close and our manual PnP is still operational as well.

Today we made copies of the computers. Working with diskettes again brings back old memories, like I still have Norton Commander musle memory. Also added DOS USB drivers so we can use USB sticks (with FAT16 filesystem) from now on. These work great and are free.
- Just unzip and copy the files on the harddisk
- Run “usbuhcil.com DisableLegacySupport”
- Run “usbdisk /DriveLetter:G”
If it works you may want to add the last two commands to autoexec.bat

I removed app pcb- and component configurations from the previous owner and started adding our own, that felt really good.

Also sorted a whole bunch of feeders/cassettes into ‘good enough’ vs ‘something wrong’ groups. Next is to refurbish some until we have the amount of feeders we actually ordered. All feeders will need some oil on various joints/hinges, but together with the new springs we’re getting should render them ‘like new’.

[mrpackethead]

I put USB-floopy disk emulators in my machines!

[Styno]

Yes, I could have done that as well, but using the usb on the compaq sff desktop inside the PnP does not require changing the hardware, I only needed to add two small programs.

[Styno]

The Optimat has a few basic package types and you have to pick one for each device that you add. E.g sot, soic, qfp, melf etc. The package types determine what package information you need to enter and configures the vision system.

I added a bunch of devices to the machine config. Many went without a problem even when I dind’t expect it, like the Microchip RN2483 which is a pcb subassembly with castellated ‘pins’. I chose Leadless Chip Carrier as the package type and within 5 minutes I had the correct lighting conditions set and a good vision recognition.

A small vqfn chip gave me much more trouble. It’s a 4 by 4 mm 24 pin package with a large ground pad and it was impossible to find a lighting condition where the package outline was visible without under-exposing the pads. Eventually I found that using the soic package type for configuring the vision the position and rotation of the part on the nozzle was correctly identified.

What can you do? I wonder if qfn packages even existed when this machine was built. The component camera uses purely IR illumination. I’ll try adding some visible light, perhaps it improves things.

[Styno]

Because the rear base plate was made inaccurately and the blown motor controller and the led ring problem (which appeared to be a bad solder joint somewhere, because at one point it worked correctly for a short while), the supplier offered to arrange a replacement machine on short notice. It arrived this morning. It’s 5 years older but these machines are old anyway, so okay, we want to move forward.

We started setting up when it became apparent various cassette guides were worn enough to give the cassette so much play that I was worrying about mispicks which I can’t accept and the supplier offered to replace them. Good.

Then the fiducial camera went dark. Luckily we traced this to a loose connector.

Then the component camera lost it’s h-sync signal. This was traced down to a broken wire in a connector which we fixed.

Then it was apparent that the component camera needed refocussing. So we followed the calibration manual and did that.

Finally I could start adding packages and during the fourth component I heard a cracking noise from the machine followed by a burning electronics smell. I rushed to switch the machine off, opened the electronics compartment door and was welcomed with a smoke cloud. So I was clearly done for today. 

The supplier will be back on monday to fix the problem with the old machine for spares. Hopefully it’s just a power supply...  I really hope this streak of bad luck ends soon!

[Styno]

So, after the problems with the electronics (we haven't checked but the supplier suspects a blown tantalum capacitor) we now have an Optima P&P that is built from two donors. The mechanics (frame, gantry and head) from an older machine and the electronics from a newer machine. In the past two weeks I've been calibrating the machine, refurbishing feeders and adding components.

I'm positively surprised by how 'easy' it is to calibrate the Optimat. There is a separate application that guides through a number of calibration step in a very detailed and userfriendly way. I.e. it tells you to "Open the hood, put the calibration nozzle in tool position 1, close the hood and press 'OK'" then the nozzle is picked up, brought towards the user and asks "Open the hood, place the plastic calibration fiducial on the nozzle, close the hood and press 'OK'" etc. The entire calibration procedure can be performed by the end-user without the need for a specially trained technician, though we needed to make a calibration tool ourselves (yay to still having some overhead sheets!). Of course there were some quirks, e.g. at some point the calibration program had difficulty rembering where to find the temperature calibration fiducial which was just tought in the previous step, but -as always- the 'Ein Reboot tut immer gut!"-rule applies.

Unfortunately the last calibration step (x/y/phi relation between fiducial camera and nozzle) is time-consuming and prone to error so I ended up manually editing the Optimat settings file to correct the x/y offset. There is still a phi error but I was finally able to place a 0603 resistor on a panel of 12 pcb's and I'm very happy with how repeatable the placement is as I couldn't spot any difference in the placement.

I only have four questions left:
- Figure out which of the two settings that (indirectly) influence the rotational calibration needs tweaking.
- Figure out how to prevent the machine spitting out the assembled pcb. There is no in-line oven behind it and having the pcb drop on the floor is not very helpful :p
- The component camera processing has difficulty identifying the corner pins of a big castellated RF-module we use a lot. But it appears to correcly calculate the rotation on the nozzle so hopefullly it will place correctly.
- Some 8mm feeders feed 4mm per actuation, some feeders of the same type feed 8mm or 12mm per actuation. I haven't yet figured out how to keep 0603 components covered by the feeder mechanism (to prevent them jumping out) and feed only 4mm per actuation.

[Styno]

It's been a while but we've been busy in the mean time. More than 100 single boards and panels have gone through the p&p and we're quite happy with the machine. The fastest panel to date was 30 minutes for 590 components, about 50% it's maximum rated speed, matching the expectations we had before we bought the machine and there is still room for improvement. A few things we learned:

The feeding problems mentioned in the previous post were caused by the cover tape spool tension. If the spool slip is too heavy then the spool pulls on the tape more than the advance pin's movement. The result is either an advance more than one hole or the tape advance ends up in between holes. The spool slip tension is regulated by loosening or tightning an M17 nylon insert lock nut (DIN 1805) pushing the spool against a spring washer and nylon sliding bearing. Some of these feeders are 15+ years old and the nylon of the lock nuts has dried out so the feeder won't stay tuned (often loosening after a while causing the cover tape to stay attached to the tape). Luckily these nuts are still readily available so we ordered some from Conrad. We found some feeders with the nut glued in place with Loctite, not useful when a cover tape needs a different setting and a pain to undo ...

Increasing the feeder actuator air pressure fixed a problem we had where the retracting lip (that holds the component steady in the pocket until pickup) was not moving back soon enough when the nozzle came down to pick up a component. Increasing the pressure on the actuator means that the actuator moves faster and the lip is pulled back earlier. Too much pressure on the actuator leads to more components jumping around so just fast enough is the trick. Another tip is to have a camera with slowmotion video function at hand, reviewing problematic pickups in slo-mo or even frame-by-frame is very enlightening.

A very light 0603 schottky diode in a plastic tape had a tendency to jump around at whichever actuator pressure and reducing the z-axis movement fixed this, though I'm not quite sure how. It's not the nozzle that makes the part jump, it's the actuator moving the retaining lip backwards. I haven't figured out why reducing the z-axis speed works, it seems like there is a second electronically controlled air valve somewhere that reduces pressure (or volume) when the z-axis movement is reduced? This needs more investigating, however, changing the setting works and that's what counts.

Some small problems left:
On one board we use two different 4-pin DIP devices , the molding of one of the devices is slightly different than the other and I can't get the vision system to reliably detect the outline position of the device and the orientation on the nozzle while the other device is detected flawlessly. It has something to do with the smoothness and molding artefacts of the plastic on the bottom and perhaps also the 'shinyness' of the feet. For now we place it by hand as the machine rejects about half of the devices.

Sometimes a random feeder (with resistors or capacitors in paper tape) refuses to pick a component a few times in a row, the machine will complain, we press Enter (try again) and everything works just fine again. Not sure if this is caused by differences in required cover tape peel strength (all tapes are still quite fresh), something else that needs tweaking or just random noise. It's not a big deal anyway.

A bit more worrysome is that the alignment of the components is not the same between panels. On one panel all components are placed dead center but on another they are all shifted a few tens of a mm to one side (perhaps up to 0.5mm in extreme cases!), always to the rear of the pcb on the x-axis (along the conveyor direction) and way outside the accuracy that the machine is capable of. The boards still reflow fine but sometimes we precautionary manually realign an SOIC-8. The cause is yet unknown. The panels have tooling fiucials that are used for the vision alignment and performing an extra temperature calibrarion does not help. Any tips are welcome.

[jmelson]

On our Philips CSM84, the small feeders (8 and 12mm) are operated mechanically by the head.  The larger feeders are air operated, and there are adjusters on EACH feeder for speed of the air cylinder stroke.  The air cylinder operates the cover blade, sprocket advance and cover tape pull.  So, you can set the stroke speed for each part based on how easy it is to flip.  If the embossed tape is properly made for the part, it should be impossible for the part to flip.  So, maybe these parts were re-reeled onto the wrong tape that has too big/deep a pocket.

As for fiducial registration, the sensor needs to be tuned to accurately pick up the centroid of the fiducial spot.  On our CSM, it seems to either find the fiducial properly or not at all.  One issue is some of the fiducials have an uneven solder plate, and the algorithm could incorrectly computer the centroid based on differing reflectivity toward the camera.

Jon

[Styno]

The part could have been re-reeled, how can one tell? Anyway it sometimes jumped half way up the nozzle, I even saw the nozzle ‘grabbing” a part mid-air once then continued down into the pocket and lifting two components after that 

The Optimat feeders are purely mechanical and all, from 8 to 32 mm, are actuated by the air cilinder mounted on the head.

I never experienced the machine not recognising a fiducial at all but the manual states that for best accuracy a double cross fiducial (looks like this:#) should be used.

Perhaps as an experiment I should ask the board house to skip HASL on the fiducials, just bare copper. Or only use ENIG.

[mrpackethead]

I dont' think i've used a HASL board in a long itme.  Its all ENIG, as its so much flatter.   If your doing QFN its mandatory.

[jmelson]

The part could have been re-reeled, how can one tell?

Hmm, not always easy.  I once got some salvaged parts that 50% had been powered up with the chips backward!  The giveaway was the date codes were all over the place on the same reel.  Diodes are not big enough to have date codes, though.

Anyway it sometimes jumped half way up the nozzle, I even saw the nozzle ‘grabbing” a part mid-air once then continued down into the pocket and lifting two components after that 

Yes, if the nozzle is not centered on the flat top of the part, it can flip the part sideways or fully upside down.

The Optimat feeders are purely mechanical and all, from 8 to 32 mm, are actuated by the air cilinder mounted on the head.

So, then the problem is the cover blade is not keeping the part from jumping in the pocket.  There should be very little space above the part until the cover blade is retracted.  This has worked VERY well on my CSM with Yamaha-style feeders, and flipped parts are quite rare.  Mis-picks are more common the smaller the parts are.

I never experienced the machine not recognising a fiducial at all but the manual states that for best accuracy a double cross fiducial (looks like this:#) should be used.

I use round fiducials only.  The CSM can use plated through holes for fiducials, or anything that is solid copper, but they recommend round or octagon.  You have the option of XY or XYX fiducial measurement.  The XYX uses the result of the X scan centroid for the Y scan, and then after the Y centroid is obtained, goes back and re-does the X scan.  I do that on boards with fine pitch parts, not needed on ordinary boards with 0805 and SOIC parts.

I'm not sure what your # fiducial is.  I have seen single cross designs used on some boards.

Perhaps as an experiment I should ask the board house to skip HASL on the fiducials, just bare copper. Or only use ENIG.

Bare copper should be fine until it tarnishes.  Most (like 99+%) of my HASL fiducials are just FINE.  More common is a solder paste fingerprint over part of the fiducial.  But, I have had just a FEW fiducials that had a kind of barrle-shaped blob on them, and my "beam sensor" needs a really flat, reflective surface to bounce light right back into it.

Watch out for gold finishes, great for long-term solderability, but if not done right, you can get HORRIBLE black pad issues.  If it doesn't solder properly on the first try, rework is some kind of nightmare from hell!  I tried it once (I think it was just gold flash on bare copper, NOT ENIG) and NEVER AGAIN!

Jon

[Styno]

I dont' think i've used a HASL board in a long itme.  Its all ENIG, as its so much flatter.   If your doing QFN its mandatory.

On this particular board the smallest package is a 0.65mm pitch SOIC-8, so (lead-free) HASL is fine and a bit cheaper than ENIG. On the RF-board with 0.5mm pitch QFN ic's I use ENIG indeed.

Yes, if the nozzle is not centered on the flat top of the part, it can flip the part sideways or fully upside down.

Yes, but in this case the device already jumped before the nozzle came down, just by the feeder action.

So, then the problem is the cover blade is not keeping the part from jumping in the pocket.  There should be very little space above the part until the cover blade is retracted.  This has worked VERY well on my CSM with Yamaha-style feeders, and flipped parts are quite rare.  Mis-picks are more common the smaller the parts are.

That's how it works on the Optimat as well but the part jumped constantly after the blade retracted and before the nozzle could enter the pocket.

I use round fiducials only.  The CSM can use plated through holes for fiducials, or anything that is solid copper, but they recommend round or octagon.  You have the option of XY or XYX fiducial measurement.  The XYX uses the result of the X scan centroid for the Y scan, and then after the Y centroid is obtained, goes back and re-does the X scan.  I do that on boards with fine pitch parts, not needed on ordinary boards with 0805 and SOIC parts.

I'm not sure what your # fiducial is.  I have seen single cross designs used on some boards.

The double cross fiducial is mentioned in IPC-SM-782A fig. 3-13.

[jmelson]

Yes, but in this case the device already jumped before the nozzle came down, just by the feeder action.

Well, there must be a problem with the feeder.  There just shoudln't be much vibration when the cover blade is retracting to make the part jump.  On my Philips machine, this does not happen, but then that machine is 700 Kg and the feeders mount to a massive steel plate that keeps them from bouncing.

Jon

[Styno]

Time for an update.
The problems with the parts jumping around is primarily solved by reducing the pressure on the feeder actuator cylinder. This means that on some other feeders the z-axis movement needs to be reduced because a slower actuator sometimes means that the component retainer lip is still located above the component when the nozzle tries to grab it. This turned out to be a finicky thing but the cost of a slightly reduced CPH does not weigh up against more reliable picks.

Then there were problems with properly recognizing some -otherwise rather straightforward- components. E.g. we had one dip-4 package that was recognized without a problem and then another that turned out to be a total bitch for the vision system. The solution here turned out to be a to reduce lighting from the 6 led-sections to just one. For other problematic packages falling back to the simple "SMALL" package setting (normally used for chip resistors/capacitors) often did the trick. In that setting the vision system does not attempt to recognize legs/pads but only looks at dimensions.

We assembled some double sided panels with a large pcb package (Microchip RN2483, roughly 18 by 27mm) and 0.5mm pitch QFN mcu, hundreds of pcb's and all of them placed accurately without needing attendance except for exchanging RN2483 trays every 10 minutes. I sure wish those would be available on reel and those fancy P&P's with tray stackers suddenly look really appealing!

Unfortunately the M17-1.0 nylon insert lock nuts (DIN 1805) we ordered as spares for the feeders do not fit. They have the correct outer diameter and threads per mm but perhaps the angle of the thread is different. Anyway, the nuts don't fit and I haven't yet found a supplier that is able to supply the correct lock nut.

[Nauris]

A

Unfortunately the M17-1.0 nylon insert lock nuts (DIN 1805) we ordered as spares for the feeders do not fit. They have the correct outer diameter and threads per mm but perhaps the angle of the thread is different. Anyway, the nuts don't fit and I haven't yet found a supplier that is able to supply the correct lock nut.

A drop of wood tar (or something else sticky) may stop it loosening itself and still allows adjustment unlike loctite.

[Styno]

A drop of wood tar (or something else sticky) may stop it loosening itself and still allows adjustment unlike loctite.

Thanks for the tip! Until now we've used some pipe-fitters teflon tape.

[Styno]

Unfortunately the M17-1.0 nylon insert lock nuts (DIN 1805) we ordered as spares for the feeders do not fit. They have the correct outer diameter and threads per mm but perhaps the angle of the thread is different. Anyway, the nuts don't fit and I haven't yet found a supplier that is able to supply the correct lock nut.

Well, that turned out to be a wild goose chase...
TL;DR Apparently the threading on the feeder axle we used for our nut-replacement experiment was damaged, we tried another feeder and this time the replacement nuts fit as you would expect. Aargh! Went to visit several businesses and scour the internet for hours for nothing.

I contacted Nordson Dima for help on this topic and while they officially do not provide support anymore their service was excellent! They were able to tell what type of nut they used and how they modified it for use on the feeder. The nuts need modification because the locking action would otherwise be too strong, so they use a tap to machine a (slightly different) thread in the plastic to give some more play. We used this information to mill one of the DIN 1805 M17/1.0 nuts, removing all but the last 0.1mm plastic. This seems to work fine too.

The support technician from Nordson found a few original nuts and offered then for a fair price so we ordered them as well.

Excellent service, should have contacted them earlier 

[Styno]

Help! I'm specifically looking for 44-mm feeders for our Dima Optimat SMPM-2000, these have partnumber SMPM-6440.

But while I'm at it:
- extra feeders (8mm/2mm pitch: SMPM-6082, 16mm: SMPM-6160)
- cassettes (SMPM-5000)
- large-reel holders (SMPM-5051)
- cut-tape holders (SMCS-400x)

Are also welcome! Please send me a PM 

And to continue toe user experience reports:
- Last week we had an issue where component pickup failed. The nozzle would move down at the pickup position retracting halfway down and continue to place the component only to find out that the camera rejects the missing component. This turned out to be caused by a constipated nozzle and the pressure sensor sensing a vacuum when it starts to measure half way on it's way down. Poking a strand of copper wire through the nozzle fixed this.
- We learned that exporting a misformatted placement file from CAD can cause the pick and place software to crash halfway a job. This is prevented for the future by adding proper formatting in the output stage of the CAD-export and extra checking of e.g. partnumber lengths.

[jmelson]

- Last week we had an issue where component pickup failed. The nozzle would move down at the pickup position retracting halfway down and continue to place the component only to find out that the camera rejects the missing component. This turned out to be caused by a constipated nozzle and the pressure sensor sensing a vacuum when it starts to measure half way on it's way down. Poking a strand of copper wire through the nozzle fixed this.

Our old Philips CSM-84 checks the vacuum sensor before every pick-up.  if it shows vacuum, it goes to the dump bin and tries to clear it with up-down strokes and blow-off pressure.  if that doesn't work, it stops and asks for help.

The nozzles need to be cleaned every once in a while.  They do get dirt, solder paste and etc. on them.

Jon

[Styno]

It’s been a while since the last update.

The machines have helped us a lot. Sales volume is slowly accelerating so we use them more often and produced several thousand pcb’s. They gave us the flexibility to sail cleanly through the parts crunch and shutdowns. However, operating the pnp machine remains a very hands-on experience, mainly requiring often retuning of feeders and fixing breakdowns.

At one point the machine went rampant and first crashed it’s head full speed into the left side of the machine gantry housing and then again into right side before I could hit the emergency stop switch. This was either a fluke event or a power dip caused by corroding contacts, since reseating all the connectors in the VME control box this hasn’t happened again. Besides a misalignment of the head, nothing appeared to be broken except my trust in the machine. A few hundred pcb’s have been built afterwards.

Some wearables need replacing like the air cilinders on the head that drive the feeders. These will move ever more irregular (not smooth) which causes bouncing in the feeders and pick timing issues. Unfortunately Festo stopped making them in 2018, so the spare-parts issue starts to get serious now.

A nearby company needed to sell their Optimat as they were ‘forced’ by their new parent company to buy a shiny new Mycronic for $$$ (even the single piece of machined aluminium for an 8mm feeder costs $500, eek!). They found this thread, contacted me and we were able to take over their old machine for a very friendly price. It provides us with spare parts and a lot of feeders.

Tomorrow I’ll be replacing the x-axis drive belt because it starts disintegrating. Bits of plastic break off, the inner steel wire core becomes visible and strands are poking out.

All in all, this was a great learning experience on all aspects of operating a small SMT assembly line on the cheap (excluding the labour). The business model is validated and so we’ll be looking for an upgrade soon.

I’m still charmed by the Dima MP-200 for which there appears to be a pretty good second hand market here. With much improved (electric) feeders, four heads, 0402 (and 0201) capability and large feeder bank, it has the important traits I’m looking for. Negatives are the small brand and the spate parts limitation this entails. Finally, the question everyone has: I wonder how this compares to e.g a Samsung 230, Yamaha XY or Juki. Any suggestion for a user friendly machine for high mix, 0402/0201/0.4mm pitch machine? Size/weight is not an issue.

[jmelson]

The supplier will be back on monday to fix the problem with the old machine for spares. Hopefully it’s just a power supply...  I really hope this streak of bad luck ends soon!

Well, I bought a 20+ year old Quad/Samsung machine at auction.  Apparently, it had sat for 6 years or so before the facility closed.
It fired up when I got it here, but then stuff started breaking.  A board that interfaces the PC to motion computer and E-stop failed, the camera failed twice, some other connections were flaky, I had to change out one of the heads, but then it all started to settle down.  I have fired it up and just started making boards, although I am still getting the programming procedures down.
Jon