Setting fiducial positions i.e. board alignment

Started by alanambrose, August 06, 2018, 05:41:49 PM

Previous topic - Next topic



I'm doing the procedure described by Mike below i.e. fudging the fid positions to make the placement work accurately. Everything is working good, except for the 2nd (i.e. top right) fid, which seems to have a mind of it own. These are 1206s btw.

So - bottom left is good:

...but top right is bad:

And no amount of shifting and 'use current position' seems to help. Anyone come across this or throw some light?

Bonus question - surely all these machines are not so out of alignment that the fid cameras are 0.5mm / 20 mils or so out? Does anyone have a machine where you can align the fids directly with the camera without adding these offsets?


Without recent calibration (which is still under investigation) there will probably be offsets between the fid cam and nozzle, and these will vary with fid position, however these are consistent for a job, so if doing multiple panels then the fid cam is definitely worth using. For a 1-off panel, less so.
In either case start at the "nominal" fid posiitons, do a test place of the top-rightmost and bottom-leftmost parts (use  tag/omit), and tweak the fid positions by looking at the placement errors. I can generally get these bang on after 2-3 attempts, judging the offsets by eye & correcting.
after setting up fids to nominal positions with nozzle in Setup PCB, omit all, un-omit tl/br corner parts.
Use Fid correct, manually move to centre fids in camera, and test place the two parts with DS sticky tape.
Eyeball the offsets and add to the nominal postions when re-fidding.
e.g. if BL part is 10 thou high, and TR one is 25 to left, start with fid at centre of cam, and then move it  up 10 for the , 25 to the left for the second. Repeat until test placements are OK. Note offsets on a post-it & use the same offsets for each subsequent panel.

Don't bother with the auto fid correct unless you find you never have any offsets - just move it manually and "use current position"


I see offsets as much as 50mils, though I suspect when it gets this big, there may be other factors like errors in the PCB size itself.
Bear in mind you can get offsets if the CDF parameters don't give accurate centres - look at the vision result, and also remember parts can move on the nozzle.
You should be using tool 2 for 1206. turning off "merge moves" can sometimes help.
Your "top-righ" isn't as bad as I've seen my machine do sometimes, even when set up well


I recently did some panels with quite a lot of 0603s and considered myself lucky that only a few QFPs needs some nudging after placement - it's sometimes a lot worse. Bit of a "phase of the moon" thing - sometimes I can place a matrix of 1024 0402 LEDs perfectly, sometimes even 0805s are marginal.


One thing you could check is the condition of the belts, if there is a worn area that could show up as inconsistent placement in some parts of the placement area. Probably not something you want to tinker with unless you have some scheduled not that busy time tho'.
Back in the days when we had an RV and it was still new, accuracy really wasn't all that bad, we build boards with biggish QFPs and lots of 0603 fairly reliably, you'd still need to inspect and possibly tweak the QFP before reflow but you could be pretty confident everything else was exactly where you wanted it to be. That was less true 10 years later when we encountered things like 0.5mm pitch 5mm QFNs (almost all needed tweaking), something the next machine found very easy, as indeed should any machine with linear encoders & decent drive control from the last 10 years.


The other, easier, approach to possible inconsistency due to worn belts, is simply to try moving the PCB to a different place on the bed. I've seen this reduce big fid offsets in the past


Thanks guys for the responses:

>>> You should be using tool 2 for 1206. turning off "merge moves" can sometimes help.

Yes I am using Tool 2 and I'm running on slow speed and have turned off merge moves. Is there a way to figure out whether the parts move on the nozzle - I've tried the slo-mo camera on my phone but without much success so far as I can't manually pan the camera fast enough to capture the nozzle action?

>>> Your "top-right" isn't as bad as I've seen my machine do sometimes, even when set up well

Interesting - I don't quite understand why as when I did a quick and dirty test with a dial indicator it looked like the machine could step 1 mil fairly reliably. I know that doesn't make for overall accuracy but still.

>>> One thing you could check is the condition of the belts

Ah thanks - didn't think of that - quite possible.

>>> something the next machine found very easy

OK now I have to ask, what was the 'next machine' :)

>>> is simply to try moving the PCB to a different place on the bed. I've seen this reduce big fid offsets in the past

Hmmm, oh dear, does this mean we really have to figure the calibration at some point?

FYI I was thinking about the problem and came to the conclusion that maybe the machine is only figuring an X, Y, R adjustment from the two fids. That would be consistent with some of the GUI and much simpler than figuring out a general transformation from theoretical position to actual position. To make two fids fit exactly it would need to include some magnification factor. If my theory is right, that means that the 1st fid gives say an X, Y adjustment and the 2nd just contributes the R. It would also mean both fids could not be fit exactly.

With this though in mind, I kind of averaged the fid adjustments, rather than fitting the 1st fid exactly - if that makes any sense. That made the overall adjustment a bit more sensible.



Quote from: alanambrose on August 14, 2018, 10:27:49 AM

OK now I have to ask, what was the 'next machine' :)

We stayed non-conveyored and moved up to an Essemtec Paraquda  (for feeder capacity over the smaller models they had at the time), quite a significant step, skipping over machines from Autotronik/Novastar because we didn't like their machines or answers to what support and maintenance looked like. Also skipped over Dima which was a similar speed machine, however seeing as Dima later became part of Nordson and ditched placement entirely.. bullet dodged methinks. At the time however, had we discovered the Intelligent drives machine without Vivo written on it.. could have gone that way.

Currently lazily planning the next one... which way it will go depends a bit on the type of work we see coming out way, 6kcph is as fast as I would go without a conveyored line when doing any kind of volume. Current trends were seeing indicate a fondness among engineers for very high BOM line count requiring many feeders, which potentially pushes us towards 2 conventional machines from Yamaha/Samsung/etc or something vast like a europlacer iineo (which we almost chose last time).

From memory the least accurate area of our machine in later life was the top right corner where you would typically locate the PCB, which makes sense.


Ah interesting re Intelligent Drives - I bought some Versatronics feeders from them recently and didn't even clock that they were a P&P vendor. I see ID is run by 'Neil Buck, Software Developer' rather than David Clements. I spy 'VivoSMT' on some of their images. Maybe the ghost of Versatronics still lives :)


Who can say, I think quite a few of the RasPi support industry startups went ID, at least one of which went Europlacer after that, no idea what those people did with their ID machines - in a corner for NPI/Dev perhaps? The only existing Vivo customer I was put in contact with at the time, was having a less than stellar experience which I thought made them a somewhat interesting choice but there we go.


I think nowadays anyone looking to upgrade from an RV for in-house work ( as opposed to full-time subcontract work) should probably be looking seriously at the various Chinese machines, some of which look close to being pretty good, albeit with generally terrible software. Most of them seem to have given up trying to design feeders and are using Yamaha/clones. Apart from Neoden, who by all reports really should do as theirs are pretty poor.
There is work being done on supporting some Chinese machines with OpenPnP.


I would agree, unless they stumble on a cheap TWS/DDM these Chinese machines are getting there. I don't understand the people who were happy with the early versions ~5-6 years ago which appeared to be very poor and just expensive toys rather than useful tools. For in-house work in micro-businesses you likely only need the machine to do the donkey work and maybe a few of the complicated parts. Start scaling above that tho' and it is important to realize how quickly placement requirements can grow in both complexity and quantity, at that point always buy more than you think you need.


Sorry to chip in late about a comment that was already slightly off topic! but I wanted to make a point or two very clear...

Quote from: alanambrose on August 14, 2018, 01:34:19 PM
Ah interesting re Intelligent Drives - I see ID is run by 'Neil Buck, Software Developer' rather than David Clements. I spy 'VivoSMT' on some of their images.

DC was never employed by, nor had any stake in ID. The company was founded by NB. DC was just an independent distributor for a couple of years at the beginning. DC owned the VivoSMT brand and the machines were branded with stickers as such at the time. They were however ID machines behind the odd sticker - designed and 100% built in house, in the UK, by ID.

Aaahh, just wanted to be clear about that!