The Reality of a Pick and Place Lifestyle

[rx8pilot]

I wanted to start a thread about the general process of PCB assembly. There are quite a few topics about specific machines, solder paste, ovens, etc.....but none that discuss the big picture reality of owning and effectively operating a pick and place line.

I started my pick and place experience on eBay a few years back and the details of the story are thick and rich. The learning curve has been intense. The overall benefit is hard to put on paper. Sometimes I love it and sometimes I want to sink the whole system in the middle of the Pacific Ocean. It occurred to me that my personal experience is full of problems and challenges that are likely self-inflicted. I purchased a used/broken machine with zero documentation. I had no previous experience.

The earliest lessons were that the pick and place machine is only a fraction of the process. The process starts in the PCB design phase - assigning part numbers, creating an accurate BOM, ordering the right parts, organizing the right parts, making sure the parts are packaged in a way appropriate for the limitations of the machine.

Then I had the learning experience of solder paste, stencils, ovens, hand placed parts, etc. Solder paste is a non-trivial decision, stencil design is important, stencil printers can wreck your day, printing small single PCB's is easy - large fine pitch panels is another story.

All that babble to say this: I have no idea if I am doing it well. I have no idea if I am efficient. I have no idea if all of it is worth it. At any given moment - I feel like the system is never good enough. I am always short something like feeders, or something needs adjustment, or whatever. My board designs keep getting harder and harder to assemble. Smaller parts, higher density, double sided = zero process forgiveness.

I am very curious if anyone is willing to share successes or failures in the realm of PCB assembly.

[rx8pilot]

Today's party:

Just need to run a quick batch of 16pcs of a single design.
2 of the coveted precision feeders that can handle 0402/0201 parts were acting up. Each needed about an hour for fiddle and fix.
4 parts are in 4 different size tubes,, but my vibe feeder does not have the right plates - so I have to hand place those fine pitch parts by hand.
2 electrolytic caps would not fit in the feeders and I did not want to figure out why - so I hand placed those as well.
I had a bunch of cut tapes of 0402, diodes, smd fuses. The 0402 parts made a mess when I had the nozzle bumping them during pickup. Took a while to get that sorted.
This could go on for quite a while.....but in the end.....these PCB's ended up taking a very long time. It took all day when I planned for 3 hours. Hell - the 1st one took 3hrs all by itself. The last one took about 10 minutes.

A gazillon mispicks, errors, etc, etc. I would swear that my machine has a personality and it can sense my fear and frustration.

Perhaps a total of 500 curse words screamed at the top of my lungs and all is well - PCB's assembled on a holiday when all assembly houses are closed.

Sure hope this is worth it, because it appears to be taking years off my life.

I need to improve part organization, solder paste printing, feeder maintenance, machine maintenance, make or buy a vibe feeder that can handle any size tubes, and spend a whole day just doing experiments to understand how to reduce the number of missed pickups. The lack of consistent performance is deadly for a small operation. I never know what to expect.

[EEVblog]

We have discussed this a lot on the AmpHour and also many places scattered across the forum.
The general consensus is that there is a quite narrow windows of usefulness for a personal PnP machine.

Mike's Electric Stuff is a classic example. He does short-ish runs of boards in short lead times that very often use thousands of LED's and other parts repeated on the board, so he derived great benefit from having his own PnP for those uses. Most people will not fit into that category though.

[RoGeorge]

All ears and wanting to hear more, please go on.

[mikeselectricstuff]

4 parts are in 4 different size tubes,, but my vibe feeder does not have the right plates - so I have to hand place those fine pitch parts by hand.

Cut the top of the tube away for 1 part length, so the parts are picked from the end of the tube - no need for any part-specificness on the vib feeder.
 

[mikeselectricstuff]

My number one tip is to take the time to write software to make the process of getting from your PCB software to the P&P machine as quick and error-free as possible. This would include such elements as :

Have fiducial components in your library, so fid positions get set automatically
Have all your PCB library parts with consistent orientation with respect to taping, so rotations just work correctly without needing to check.
Physical differences (height and vision parameters) mean you need to have different PCB library parts for the same nominal package, e.g. for 0603 you may have resistor, thin cap, thick cap, ferrite bead, LED etc. and each may have different parameters for vision, speed,height etc.
 
For parts that can come in tape or tube (Mostly SO ICs), you need seperate library parts as the orientations are different.

Dealing with panelisation - step/repeat, intelligent handling of fids ( PCB may have fids per circuit or per panel, so detect the outermost and only pass these to the P&P system)

Other things :
By far the biggest factor in getting good assemblies is stencil print quality. You need a tensioned stainless stencil which is accurately flat against the whole PCB surface.

Have as many feeders as possible - loading parts into feeders is takes the  longest time in setup ( if other things take longer you're doing it wrong).
Buy jellybean parts in full reels and keep them in the feeders.
 
It is almost never worthwhile to machine place a single board, unless it has more than about 100 parts, or most parts are the same.
With a foot operated vacuum pen and tape holders you can typically manually place at 10-30cpm

Unless it really matters and someone is actually going to check, use leaded solder. It has a much bigger margin between "reflow" and "incinerate"

[mikeselectricstuff]

If you think you can build a useful pick and  place machine you are almost certainly wrong.
If you think you can do it for less than $1000 you are definitely wrong
More so if you start by making an XY gantry. The hardest part is the picking, not the placing. Feeders are everything. A P&P without reel feeders is a toy.

A reliable pick & place process ( the machine is only part of the process) is not one big problem but a a huge collection of small problems, any of which can result in lost productivity, wastes parts & trashed boards
 
If you want a fun project, go for it but treat it as an educational experience which is unlikely to result in a useful tool. If you need a cheap machine, go buy a used dinosaur or Chinese machine and be aware it will take time and tweaking to make it useful.

Any pick and place machine, at any price, will take you a few weeks to get a process and running well. If you need some boards made quickly, go to a subcontractor
 
One example of the sort of issue you can hit  that immediately comes to mind : I had a crazy timescale project - ten boards with about 2500 LEDs, plus firmware etc. to play videos from a MicroSD card. Timescale : about a week and a half, from nothing.
 
This didn't particularly faze me - done this sort of thing loads of times before - thought I had at least a day contingency.

However the only LEDs that anyone had enough stock of at a reasonable price came on really thin tape. When this went through my feeder, about half the time, the LED would jump out of the tape when it was advanced.
It took most of my contingency day to modify the feeder to the point that I only lost about 5%, fortunately Digikey had enough extra stock to cover this and there was enough  time to get another reel to make up the wasted parts

[mikeselectricstuff]

And another one - LEDs with soft silicone tops, sticking to the nozzle ( my machine doesn't have blow-off)
https://twitter.com/mikelectricstuf/status/994172467232018432

[DerekG]

I wanted to start a thread ..... but none that discuss the big picture reality of owning and effectively operating a pick and place line.

Excellent idea.

At any given moment - I feel like the system is never good enough. I am always short something like feeders, or something needs adjustment, or whatever. My board designs keep getting harder and harder to assemble. Smaller parts, higher density, double sided = zero process forgiveness.

For all these reasons, I have decided to outsource our SMD assembly.

But, I'm always wanting to know if I have made the right decision.

[Smallsmt]

Have all your PCB library parts with consistent orientation with respect to taping, so rotations just work correctly without needing to check.

Funny thing but the feeder setup solve this problem!
Because if you have feeders on different side of the machine this can't help, especially if you use feeder cassettes.
You need to define the angle deviation inside feeder setup.

[Smallsmt]

And another one - LEDs with soft silicone tops, sticking to the nozzle ( my machine doesn't have blow-off)
https://twitter.com/mikelectricstuf/status/994172467232018432

Blow off is normal on our pnp machines but sometimes a special teflon nozzle solve this problem too.
We use another parameter to close vacuum in advance on Z axis to push the part to paste surface using the blow off.

[Smallsmt]

I wanted to start a thread ..... but none that discuss the big picture reality of owning and effectively operating a pick and place line.

Excellent idea.

At any given moment - I feel like the system is never good enough. I am always short something like feeders, or something needs adjustment, or whatever. My board designs keep getting harder and harder to assemble. Smaller parts, higher density, double sided = zero process forgiveness.

For all these reasons, I have decided to outsource our SMD assembly.

But, I'm always wanting to know if I have made the right decision.

If you always want to cover 100% of part placement it is a hard job.

In our prototype assembly I found out it it helps a lot if machine place the standard parts already on feeder so manual assembly for some special parts took not too much time.

Definitely the stencil printing and paste type are most important for good results.
And cheap bend PCB's cause a lot of problems too.
We use mostly gold plated pads on our pcb's because it's more reliable if you store the PCB over longer time.

One of the biggest problems is to care the right feeder setup! If you put a part in wrong orientation  or wrong part in feeder you receive a lot of rework.

Mostly I do a first PCB PNP run and check everything after finished. Then start the remaining pcb's.

In one PNP job I received 68 wrong placed TQFP100 because of 180 degree turned JEDEC tray. It took 2 days to repair the boards.

[Kjelt]

Excellent stories and such bad experiences. 

Just need to run a quick batch of 16pcs of a single design.

Is this not the kind of job where you and Mike each and everytime in a topic say you do NOT use an P&P for ? 

It took all day when I planned for 3 hours. Hell - the 1st one took 3hrs all by itself. The last one took about 10 minutes. 

Exactly and the last 100000 will take each a couple of minutes, you have the quircks ironed out and are ready for massproduction.
Oh wait you only had 16 boards. Perhaps it was quicker to do it manually as you say so many times yourself.

For these kind of jobs you don't need a P&P machine made for massproduction that runs ultrafast.
You were better off with an intelligent machine that can visually check it picked the component and orientation, then checks the placement after.
Takes time yes but who cares with <50 boards.

However the only LEDs that anyone had enough stock of at a reasonable price came on really thin tape.
When this went through my feeder, about half the time, the LED would jump out of the tape when it was advanced.

For modern P&P machines there are probably feeders suited for this kind of tape ?
You were lucky you could modify the feeder in time, I think no other party would have accepted this contract in that limited amount of time.

[mikeselectricstuff]

Have all your PCB library parts with consistent orientation with respect to taping, so rotations just work correctly without needing to check.

Funny thing but the feeder setup solve this problem!
Because if you have feeders on different side of the machine this can't help, especially if you use feeder cassettes.

Only if the p&p software is so dumb that it doesn't know where the feeders are, so it can automatically apply an appropriate rotation.
if you move a feeder, the only thing you should need to do is tell the machine where it is - the machine should know to do any rotation to compensate.

[mikeselectricstuff]

And another one - LEDs with soft silicone tops, sticking to the nozzle ( my machine doesn't have blow-off)
https://twitter.com/mikelectricstuf/status/994172467232018432

Blow off is normal on our pnp machines but sometimes a special teflon nozzle solve this problem too.
We use another parameter to close vacuum in advance on Z axis to push the part to paste surface using the blow off.

My machine went obsolete long before silicone LEDs were a thing. if it was a major issue I'd mod it to blow, or make teflon nozzles, buf for now a spray of PTFE lube spray on the nozzle every so often fixes it.

[mikeselectricstuff]

Excellent stories and such bad experiences. 

Just need to run a quick batch of 16pcs of a single design.

Is this not the kind of job where you and Mike each and everytime in a topic say you do NOT use an P&P for ? 

Depends entirely on the job. Once you have a good process, a single panel can be worthwhile but you'd typically hand place anything there was only ohne or two of per circuit, unless it was already in a feeder.
Subbing a single panel gets expensive due to their setup, and the fact that (IME) subcontractors always insist on using full-sized stencils that fit their machine, despite the fact that a small, cheap frameless stencil is just fine for 1-off jobs

However the only LEDs that anyone had enough stock of at a reasonable price came on really thin tape.
When this went through my feeder, about half the time, the LED would jump out of the tape when it was advanced.

For modern P&P machines there are probably feeders suited for this kind of tape ?

Quite probably. The feeder design on mine is maybe 25+ years old and far from ideal. It mostly works, most of the time, and doesn;t need too much tweaking. if I used my machine mopre than a few times a month I'm sure I'd have moddied it to use some Yamaha or simlar feeders

[rx8pilot]

Just need to run a quick batch of 16pcs of a single design.

Is this not the kind of job where you and Mike each and everytime in a topic say you do NOT use an P&P for ? 

[/quote]

Like everything in the P&P world...it depends.
In this particular case the PCB has about 125 placements, about 30 unique 0402 passives among other parts. In addition, this is the third time I have run these boards so the setup was almost complete. I think I had to swap just a few parts in feeders to get started.
Even after the programming was done and proven on previous runs and the feeders all loaded.....all the stupid problems still ate my lunch. One feeder was getting stuck, another feeder had an alignment issue. The Z-pickup heights all needed some adjustment that I was not expecting. The laser scanner that measures and orients the parts on the nozzle had a smudge on it which increased the reject rate even though the parts were perfectly positioned. On that one, I was chasing it down in software thinking the part profile was messed up. The lens of the scanner is very hard to see and it was the last thing that came to mind - not even sure how it would get dirty. Perhaps I manually changed a nozzle at some point - who knows.

I did the initial prototype of this design by hand which took about 4 hours or. That included all the time needed to double check part values, re-calculate things that did not 'seem right' just to make sure, etc. My guess is that it would by about 90 minutes each on average if I manually assembled these. After all the delays and difficulties with the P&P - the average time was about 30min/each. Considering that 30min is the disaster scenario for P&P - that is not the end of the world.

While this setup is a sad story, I do have some good ones for my other designs. The designs I do the most of (and keep all the feeders loaded for) are easy to run just a handful of boards without much trouble. Boards that took 60 minutes on the best of days with my manual setup are done in about 5-6 minutes on the P&P. That is pretty good, but in fairness, it took many hours to get it all working like that.

For these kind of jobs you don't need a P&P machine made for massproduction that runs ultrafast.

Totally true. My machine has an IPC rating of 3600CPH but I never run it that fast. I estimate that when I am at 'full speed', the machine is only placing 1,000CPH. The whole thing is limited by the processes before and after the machine and I am the only one doing it. If I had an 8 head machine doing 30,000CPH it would not make much difference unless I also got a fully automatic printer, a production oven, an AOI machine. A LOT of my time is burned dealing with cut tapes and only having a few spare parts available. One of my designs has 3 chips that are about $20-$30/each. I cannot just let those get dumped in the bin and keep guessing. I have go fetch the part, put it back in place until the adjustments are made. Ceramic capacitors are getting into this range as well - many of the caps in my power stuff are $.75-$1 each so I have only a small number of spares and have chase down any bad picks. A small issue will put $100 of parts in the dump bin pretty quick.

And another one - LEDs with soft silicone tops, sticking to the nozzle ( my machine doesn't have blow-off)

For the first month - I had no idea my machine had blow-off. Many of the parts were sticking to the nozzles just enough to shift them out of position during the release. It took me a while to figure out that blow-off was built in and I only needed to connect compressed air.

Is this not the kind of job where you and Mike each and everytime in a topic say you do NOT use an P&P for ? 

Depends entirely on the job. Once you have a good process, a single panel can be worthwhile but you'd typically hand place anything there was only ohne or two of per circuit, unless it was already in a feeder.
Subbing a single panel gets expensive due to their setup, and the fact that (IME) subcontractors always insist on using full-sized stencils that fit their machine, despite the fact that a small, cheap frameless stencil is just fine for 1-off jobs

My decision to get into P&P was dominated by deadlines that CM's or other assembly options could not accommodate. For the times that the schedule technically permits - it is still a rush job and priced accordingly. Even if I was willing to pay the price, I still have to organize the project data precisely and kit all the parts and get them delivered to the assembly house. That is, in many cases, half the work of making a P&P run if not more. On top of that....I still have the stress worrying about component rotation or some other simple error that will throw the project into a tailspin. At least when I run my own PCB's, I can take the first one to the bench, inspect and power it up before I run the rest of them.

[mikeselectricstuff]

My decision to get into P&P was dominated by deadlines that CM's or other assembly options could not accommodate.

Me too. And it is not just simply timescales. Doing it in-house means a lot less time documenting things - I can sometimes have a whole job assembled in-house in less time than it takes prepping everything for an assembly house, waiting for quotes, dealing with queries etc.

 

[rx8pilot]

4 parts are in 4 different size tubes,, but my vibe feeder does not have the right plates - so I have to hand place those fine pitch parts by hand.

Cut the top of the tube away for 1 part length, so the parts are picked from the end of the tube - no need for any part-specificness on the vib feeder.

I sorta hacked something like this together a number of months ago - and it sort-of worked. It seems I am fighting a poor design overall. The parts were getting stuck or they were vibrating right out of the feeder. There is an adjustment for frequency and amplitude, but it is tricky to set it. My guess is that a fairly simple machining effort should allow some improvement. My latest effort was laser cutting some thin plastic sheets to guide the parts. That was helpful, but still not quite reliable enough for prime time. SOIC-8 parts are pretty good, but DFN-10 are not. More fiddling ahead.......

My number one tip is to take the time to write software to make the process of getting from your PCB software to the P&P machine as quick and error-free as possible.

This was, bar none, the biggest single process improvement I have made! It wasn't even all that difficult relative to other areas of improvement.

What this forced me to do is create and enforce a part numbering scheme that covers the whole process from design to delivery. I built a spreadsheet (all I had at the time) to create the part numbers based on pre-defined criteria. That number is used in the CAD software, purchasing, storage bins, feeders, P&P software, etc. When I have completed a design, I can one-click the output with the part number X-Y and Theta to the P&P and that is it. I have it setup to do top and bottom of the PCB in separate files so there is no programming penalty for double-sided designs. HUGE time saver.

The part numbering scheme was designed so that when it is sorted - it naturally groups part families together and in sequence of values.

Example:
E20-040-0960
E='ELECTRONIC COMPONENT'
20= Resistor
-040= SIZE 0402
-0960 = 10k (part number has no relation to the real value)

during prototyping or maintenance, I can just scan through the bins quickly to get any part I need. At the moment, I think I have about 500 unique parts in use and I can find them very fast. I can look at a feeder and know what is in it. I can look at a DigiKey invoice and know what the part is and what products it is used in. It seems so simple, but without a logic part numbering and organizing system, I would spend a lot of time looking and guessing.

The next major upgrade would be developing a database - Filemaker is what I am looking at. The excel spreadsheet has long been outgrown and is very difficult to maintain.

Have fiducial components in your library, so fid positions get set automatically

I forgot to put fiducials on my last design!!! What a pain. The latest Eagle version allows user-defined design templates which I think I will use to avoid forgetting in the future.
Initially I thought I could use some vias or pads - but that did not work very well. All the vias had connected traces or they were stitching vias. The connected vias are no longer a circle. The stitching vias had no copper edges - just the solder mask which is not nearly accurate enough.

I had to manually align each board and verify it before any placements. Probably only a minute per PCB, but that seems like an eternity when it is usually totally automatic and takes 1 second.

In one PNP job I received 68 wrong placed TQFP100 because of 180 degree turned JEDEC tray. It took 2 days to repair the boards.

Oooof! That is my nightmare scenario.

Re-work can undo the efficiency of the fastest pick and place system in the world. The faster your machine is, the faster it can break things.

[mikeselectricstuff]

The part numbering scheme was designed so that when it is sorted - it naturally groups part families together and in sequence of values.

Example:
E20-040-0960
E='ELECTRONIC COMPONENT'
20= Resistor
-040= SIZE 0402
-0960 = 10k (part number has no relation to the real value)

This may not work for everyone, but I always treat part types and values seperately - e.g. SMALL0603R is the PCB footprint and P&P machine package type for vision & height, "1K" is the part value. ("SMALL" is a historical artifact)
My machine also treats them seperately, so you don't end up with ridiculous numbres of part types. The part type determines dimensions and vision params, the value is the label it gives the feeder you loaded it into

[rx8pilot]

The part numbering scheme was designed so that when it is sorted - it naturally groups part families together and in sequence of values.

Example:
E20-040-0960
E='ELECTRONIC COMPONENT'
20= Resistor
-040= SIZE 0402
-0960 = 10k (part number has no relation to the real value)

This may not work for everyone, but I always treat part types and values seperately - e.g. SMALL0603R is the PCB footprint and P&P machine package type for vision & height, "1K" is the part value. ("SMALL" is a historical artifact)
My machine also treats them seperately, so you don't end up with ridiculous numbres of part types. The part type determines dimensions and vision params, the value is the label it gives the feeder you loaded it into

Without a doubt.....part number schemes are like underwear. You don't want someone else's underwear. :-)

I made of list of personal goals for my part numbering system (not important to everyone)
1. Fixed length if possible
2. Sorting the list automatically groups similar parts
3. Easy to human read with modest amount of memorization
4. Syntactically easy to use in the various systems at play
5. Scalable beyond any current expectation

My machine is kind of similar - but requires a unique part number for each part. You start with a defined package. When you add a part, you have to pick the package which determines the feeder, nozzles, dimensions, etc. In the end....I now have a very long list of parts in the P&P that match the part numbers in the rest of the system. The length of the list is not much of a challenge since I never really directly use the list. When I import a new design - it queries the list and imports the data. For new parts, it gives me a list and I add them as needed. I can also add them with an offline utility as well.

I have 'generic' parts too, but so far have not used them. By the time I am assembling the PCB, I have been forced into creating the new part numbers already for purchasing and storage.


On the topic of solder paste......

I used leaded solder for quite a while. As Mike pointed out, it is easier to use and a lot less likely to melt stuff. As I started using smaller parts, I moved to a Type 5 solder paste Loctite GC10. It is un-leaded and I have to be much more precise with the oven profile. Does anyone know of a Type 5 leaded paste? I have no requirements to use lead-free at the moment, and would love to have Type 5 leaded. Easy to print and easy to re-flow.

[Kjelt]

Type5 is called AGS or AF5  from Multicore usually sold in syringes AFAIK because the syringe needles require the smallest particle size to pass.
But they should be sold in jars as well, can not find the leaded type 1,2,3 seems like the entire industry went Rohs.

[jmelson]

I have a Philips CSM84 (made by Yamaha).  It is quite old-school, no vision, chuck alignment jaws to center small parts, and a mechanical alignment station that centers large parts on the nozzle by bringing the part over to that station.

It is built like a tank, though.  I've had a few problems with it, but it mostly just keeps on running.  But, without vision, the centering is not so great for fine-pitch parts.
It will do SOIC parts with perfect accuracy, but .65 mm quad packs are a little iffy, and .5 mm definitely have to be adjusted by hand.

I recently did just TWO boards for a customer, but they had about 400 components each, and I did have the whole setup already saved.

I wrote a little C program that takes the placement file from my CAD package and converts it for the P&P machine.

Jon

[jmelson]

I forgot to put fiducials on my last design!!! What a pain. The latest Eagle version allows user-defined design templates which I think I will use to avoid forgetting in the future.
Initially I thought I could use some vias or pads

My Philips CSM84 can use pads as fiducials.  I'm not sure they are as accurate as a fiducial pad, but it works pretty well.  There is a setting for normal or reverse fiducial.

Jon

[rx8pilot]

I forgot to put fiducials on my last design!!! What a pain. The latest Eagle version allows user-defined design templates which I think I will use to avoid forgetting in the future.
Initially I thought I could use some vias or pads

My Philips CSM84 can use pads as fiducials.  I'm not sure they are as accurate as a fiducial pad, but it works pretty well.  There is a setting for normal or reverse fiducial.

Jon

My machine is supposed to do it, but rarely does. The worst part is that it will get a false position and not throw an error - then 'correct' all the positions into oblivion. It becomes a big mess pretty quick.
Lesson learned - fiducials are just as important as food and water, lol.