Neoden 4 pick and place

[TheSteve]

We have never had it image more then a single part at a time with the camera - there is an option you need to find and change.

Are your X values on the screen for your PCB the same or different? I can't recall what the software calls things any more specifically as I don't have the machine in front of me and we didn't even turn it on last week so its been a while since we used it. I do know the X values should be the same but if you do things in the wrong order or click the wrong button they will be skewed a small amount and so will all of your parts.

[elmood]

It images parts individually if there are a bunch of parts from the same nozzle in a row. But in my case it used nozzle 3 for a large IC and the next chip in the list was nozzle 2. I'm not sure how to make it do each one separately.

Can you explain this thing in the wrong order problem again? I set the board edge align, and then I set the first chip align and create panelized list. I need to do more testing to see if all the parts are off or just ones from some nozzles.

Also, after doing a careful check of the alignment of camera and nozzles in the setup screen I compared new and old values and nothing was more than 0.1mm differently than the numbers from the factory. So that definitely wasn't the issue.

So close! Yet not close enough.

[mikeselectricstuff]

The first run went somewhat okay, but not great. We had a few placement problems because the PCB was actually not located far enough into the machine. It's important to check that the entire PCB is within the movement area for all nozzles or else there will be an error during the job. Simple, but wasn't obvious until about half way through the job.

Once again, let down by poor software - it should check this before it tries to place anything.

Another problem (since we restarted after a few errors) seems to be that the software lets you restart part way through the job (cool feature!)

Cool ? Essential more like. A partly-placed job is likely to be scrap unless it can be completed.

We also placed a TQFP part and it was in a completely wrong spot. I'm not sure if it didn't image correctly but it was about 0.25" or more off. It was coming from a tray, so I'm not sure I'm using the trays correctly. The user manual has a lot of nice screen shots but the text is more or less unusable.

It has vision - the pick position should have no effect in placement accuracy

Also, when stopping a job the machine appears to drop the part randomly, instead of putting into the trash box.

sigh.....
Further evidence that they have a lot more work to do on the software - such a shame they just don't seem to "get it"....

[mikeselectricstuff]

TheSteve: Is there a way you've found to prevent multiple parts from being picked up at the same time? I found that the TQFP worked perfectly when it was the only part on the head. Previously my job had two different nozzles in the chiplist one after the other. The TQFP got picked up first, followed by a SOIC on a different nozzle. When it went to image the TQFP the other chip messed up the detection and then it placed it in the wrong location.

Yes, you select if you want the parts picked/placed "jointly" or not. Pretty sure you can also select the camera to operate per nozzle or "jointly" as well. We do all of our camera parts one at a time.

..but presumably it;s not sensible enough to have a programmable clip area for the camera image to avoid imaging adjacent parts..?

[elmood]

I'm still not sure if there is a willingness to fix the software or not. If so, I would definitely recommend the Neoden 4. If not, then I would recommend against it. I'll have a better idea soon, as I've been in touch with Neoden about the issues. It's hard to tell if they care about the feedback. A few times they've said they'll have the engineers look into various issues. I'll believe it when I see a new release with at least one of the things I've requested fixed or added.

I've done more testing today and found that for some reason all parts on nozzle 1 are placed at bit "off" the mark. This is despite being imaged and despite having the nozzles all aligned according to the system calibration.  I'm sure most of these parts will correct during reflow but I really just want it to be right. I'm trying to figure out if there is some kind of pattern to the error, but so far have to run more tests and analyze some macro photos of the boards.

Placing ICs on the larger nozzles is bang on! I was able to get my TQFP placing correctly by selecting "individual" vision instead of "large component"... apparently it fits within the range for the individual size. But it still picks up multiple parts since I had three parts in a row with nozzles 1, 2 and 3. Apparently there is no way to "block" it from optimizing this way. Maybe it doesn't matter, but I'm not sure for an even larger component how this would ever be able to work.

If I can work out the offset problem for small parts I'll probably be able to make some of my actual production boards!

[uncle_bob]

Hi

Ok, so here's a few questions:

How many alternatives are there at the same price, with the same features?

How well are they supported outside their country of origin?

How good is their software?

Yes, there is stuff that is way better, for a higher price. If one spends more, they get more. Something that does not work is useless at any price. The main issue is -- can you get something that works at this price point or not?

Obviously it is never ever a works / does not work thing. As long as these machines have been around, there have been issues. Every machine I have ever owned (.. company owned ...) needed work arounds of some sort. The learning curve always was there. At the end of the struggle, they either placed the parts we needed or they didn't. If it was a fail, we either changed parts or changed the machine.

Bob

[elmood]

I don't expect any industrial machine to be completely smooth... these are low-run products with limited user bases. I definitely don't feel "ripped off" yet, provided that Neoden can help me sort out these remaining issues. A public announcement of new software with a changelog would go a long way toward showing goodwill for current and potential customers.

Speaking about price... every other system I looked at in mid-late 2015 was either in progress, or looked completely horrible / DIYish from a hardware perspective. Comparing the build quality of the Neoden 4 with the Liteplacer kit that I was working on prior to buying this machine is like night and day. The Liteplacer looks like it's made out of popsicle sticks compared with the solid frame of the Neoden. The motors are WAY bigger too, and the entire frame of the machine feels really solid. All of the other low-cost solutions looked like crappy 3D printer construction... one of them was even made out of wood!

mikeselectricstuff: I get the feeling like you don't want these low-cost machines to work for some reason. Yes there are a few wrinkles but I'm confident that the hardware is capable, and the software feels like it's at about 90% of what I would have expected for a first release. If Neoden fails to make any fixes then I might join your camp, but until then I'm hopeful.

For small shops such as mine having a machine that is physically small and affordable definitely opens up the doors to profitability. At least in Canada these days the PCBA landscape is the worst thing ever. The service and quality is so terrible, and the price is insane. We've been sending PCBA jobs to Thailand and they do amazing work, but the freight costs more than doubles the price!

BTW if you want to read a story about a "high end" piece of equipment and how awesome it was at first release, there is a famous article here that basically doesn't make me feel so bad about having a few issues with the Neoden 4: https://www.cs.princeton.edu/~bwk/202/summer.scanned.pdf

[uncle_bob]

Hi

Ok, so simply put, there really isn't any competition at this price point.

Next up ... Is the software actually written by Neoden or by a third party?

I have no idea how you would be able to tell. It is a pretty critical point though. Often this sort of stuff gets jobbed out. An almost adequate job of multi-lingual software gets done. The software guy(s) now realize what they are into. Up goes the price for future work. The other side of the deal factored a fixed one time cost into the pricing. Instant impasse, both sides are at fault. I've seen this happen a *lot* of times. You can see traces of it in the interesting "we had to re-write the firmware from scratch" announcements. (= we brought it in house / found a different outsource location).

Bob

[elmood]

uncle_bob: I sure hope they make the code in house. But yes, at least as of December it's fair to say that nothing seems to beat the landed price of this machine. A lot of parts of it are very cool, and I feel that apart from some "nice to have" feature requests, it's nearly okay.

I know thesteve had success at his shop making real boards. Indeed during my testing I was having better luck but something must have changed that is now causing me some trouble. I'm hoping it's my fault and that a simple change to my setup or input file can correct it. As long as the machine is clunky yet consistent I have no problem with that. This is still to be confirmed.

[uncle_bob]

Hi

At least based on the companies I've visited over there (I know *nothig* about Neoden) it is rare to find an outfit that is really good at hardware (as these guys seem to be) *and* that does software in house. The smaller outfits tend to get very narrowly focused. They sell the hardware to somebody else and that guy tacks on the software (or they job it out). The first involves a markup the second a potentially difficult relationship.

Further obscuring it, twin brothers Bob and Mike may be involved. Mike's company makes the great hardware. beer swilling Bob's company makes the software. Is it really one company or two ...

Bob

[mikeselectricstuff]

mikeselectricstuff: I get the feeling like you don't want these low-cost machines to work for some reason. Yes there are a few wrinkles but I'm confident that the hardware is capable, and the software feels like it's at about 90% of what I would have expected for a first release. If Neoden fails to make any fixes then I might join your camp, but until then I'm hopeful.

Quite the opposite - I'd love to see a low-end pick & place done well. If the Neoden4 had good software and could handle tall parts I'd very seriously consider getting one.
I just find it very frustrating that the N4 could be a very good machine, but like most Chinese products is badly let down by an inadequate attention to software design and implementation - something that would not be hard to fix if the manufacturers had the right attitude. IMO the utter balls-up they made with the video says it all.

[uncle_bob]

mikeselectricstuff: I get the feeling like you don't want these low-cost machines to work for some reason. Yes there are a few wrinkles but I'm confident that the hardware is capable, and the software feels like it's at about 90% of what I would have expected for a first release. If Neoden fails to make any fixes then I might join your camp, but until then I'm hopeful.

Quite the opposite - I'd love to see a low-end pick & place done well. If the Neoden4 had good software and could handle tall parts I'd very seriously consider getting one.
I just find it very frustrating that the N4 could be a very good machine, but like most Chinese products is badly let down by an inadequate attention to software design and implementation - something that would not be hard to fix if the manufacturers had the right attitude. IMO the utter balls-up they made with the video says it all.

Hi

You might be surprised at just how small an operation many of these outfits are staff wise. It is *very* common to meed "the engineer" in a lot of these places (with sales in the > $30 M range !!!). They have a half dozen to a dozen guys who work for "the engineer". Once you start talking to them over a few days, it would be an exaggeration to call even 2 of them tech's.  The rest are what we would call assemblers / prototype wiring guys or inspectors (he may also be the QA manager). In that environment, adding a full blown software development and support group is a really big deal.

Just so nobody gets offended here. The exact company details above are *not* for a Chinese pick and place company. The same business model exists in multiple countries in that region and covers a lot of electronic manufacturing / design / production. My intent is not to slam anyone. The model works amazingly well. The product is top quality. Often ....errr ... some people ..errr ... team up with these companies. The result is a product that has ... errr ... support .. and a higher price tag. Details hopefully well enough concealed to let me keep my job.

Bob

[MTNELECTRONICS]

Just a quick update:

I finally got my replacement nozzles and after setting up the machine with them I'm happy to report that it is now working well with the 0805 caps.  I have run the machine all the way up to 100% speed now and the placement accuracy doesn't appear to be visually degraded between 100% and 60% speed, but so far I have not done any fine pitch ICs or components smaller than 0603 so I am definitely not testing the machine's limits yet. 

When I have turned on the "Jointly" vision recognition I get an obviously incorrect offset on the parts, even with fairly large 0805 parts.  The calibration is obviously off, because the machine places much more accurately with the vision off, and very well with the vision set to 'Individual'.  I have not needed the extra speed yet so I haven't bothered to try and calibrate it but it is on my "to do" list. 

Make sure that your component library value actually matches what you are putting in the machine.  If you are feeding the machine bad information, it cannot give you good results.  I took a caliper and measured each of my parts then added that part to the component library for vision recognition.  Check and make sure that the outline the vision is recognizing actually matches up with the part's outline. 

The software does have some idiosyncrasies that could be fixed, although I am doubtful that they will be fixed.  Those that I mentioned earlier (manual board offset being the biggest one), along with some that I just read (such as the machine just dropping the part where it sits if you press "stop" during placement).  I also just began using the vibratory tube feeder, and while it works good enough once set up, it is disappointing that you have to manually go into the test menu to start and stop the vibrator both before and after a placement job, so if you forget to turn it on before a job you will start placing then run into feed errors, then have to exit out and start the job over again.  In my mind, if one of the vibratory feeders is called for during placement then it should turn on automatically when you start the placement routine.

This is just like most of the Chinese flashlights I deal with---they go about 90%-95% of the way to making a truly great product, but then fall short on some of the small details that could elevate the product from 'acceptable' to 'great'.  I am happy enough with the machine as it sits as long as it keeps running reliably, but if they had a more polished interface and a better sales pitch can you imagine how many more of these they could sell? 

[mikeselectricstuff]

but if they had a more polished interface and a better sales pitch can you imagine how many more of these they could sell?

Unfortunately it appears they don't have sufficient imagination....

[thommo]

I've done more testing today and found that for some reason all parts on nozzle 1 are placed at bit "off" the mark. This is despite being imaged and despite having the nozzles all aligned according to the system calibration.  I'm sure most of these parts will correct during reflow but I really just want it to be right. I'm trying to figure out if there is some kind of pattern to the error, but so far have to run more tests and analyze some macro photos of the boards.

Placing ICs on the larger nozzles is bang on! I was able to get my TQFP placing correctly by selecting "individual" vision instead of "large component"... apparently it fits within the range for the individual size. But it still picks up multiple parts since I had three parts in a row with nozzles 1, 2 and 3. Apparently there is no way to "block" it from optimizing this way. Maybe it doesn't matter, but I'm not sure for an even larger component how this would ever be able to work.

If I can work out the offset problem for small parts I'll probably be able to make some of my actual production boards!

Sounds like you are running just a few 'test' components currently [which is understandable Elmood].

Do the components you are selecting at Nozzle #1 require 'rotation' at the PCB placement position by any chance?
If that were the case, and you're assured that it is collecting the component correctly from the tape, then the 'seating' of the nozzle could be slightly out of alignment [not concentric with the shaft], causing the displacement error you are referring to. Just a thought.

[elmood]

thommo: I thought that might be the case, but in the setup screen I was able to do a rotation test where the machine will overlay images from multiple rotation angles, and it seems that it's not deviating.

My gut feeling is that somehow it's failing to calculate an offset or correct for it. I have to confirm this by checking the screen for how the part images each time. It seems like parts of both rotations are off. Obviously the imaging system works because it picks up parts from the trays and are able to perfectly align them on the pads. (TQFP parts) I do recall that previous tests worked better, but there are a lot of variables and things that have changed since I first tried it.

There was no response from Neoden last night so I'm hoping to hear from them tonight. I really hope it's something silly that I'm doing wrong. One feature that I think should be included is the ability to show the alignment of a component on the board. You can align with the camera the feeders, which is a great way to double check that the machine is set up correctly for a job. But I wish you could see each component with the downlooking camera to see where it thinks it's going to place the part. That would help to rule out errors in the input data.

[TheSteve]

But I wish you could see each component with the downlooking camera to see where it thinks it's going to place the part. That would help to rule out errors in the input data.

This isn't possible with larger parts as the camera has a fixed field of view - so while you can see smaller parts you'll only be able to see a small portion of larger components. I believe with all larger parts in the future we will place cross-hairs on the PCB to mark the center of the part. That will make it easy to verify position with the downward facing camera.

[elmood]

TheSteve: How do you go about actually pointing to the part locations with the downfacing camera? I imported my coordinates from a file and there is no option to do this manually it seems.

[MTNELECTRONICS]

TheSteve: How do you go about actually pointing to the part locations with the downfacing camera? I imported my coordinates from a file and there is no option to do this manually it seems.

As long as "manual" is unchecked, you can't do it.  The problem is that if you manually program some of your components with the machine coordinates using this method, you really have to program all of them including the fiducials using the machine's coordinates.   This is why it would be extremely helpful if there were an offset setting that you could change. 

[TankSparks]

You can view where it will place parts with down looking camera even with manual unchecked.
You need to mount file,  step through fiducials then click on part you want to view, then click vision align button.

[elmood]

TankSparks: Thanks for that info! I didn't know about the vision align function because it always seemed greyed out. I guess it wasn't too obvious how to make it enabled, but doing the calibration part of the mounting routine makes sense. We had our first mostly successful run yesterday! I'm definitely finding out what needs to go on our checklist before starting a job each time.

Our placement alignment issues seemed to be due to the way our fiducials (mark points, in Neoden-speak) were being detected. Obviously it's necessary to have at least a few points because the rail system is not 100% accurate at positioning the board... there is always some room for error both side-to-side and while feeding in the PCB. A few things we found after some careful testing:

- The number of mark points and their positions has a big impact on positioning overall on the board. Too few / bad locations chosen will make some parts perfectly aligned while others are off a bit. Our test board is a worst case of an older design we have that doesn't use proper fiducials or rails. It was hard to find good points that would make the alignment bang on all over the PCB.

- The size of the mark point has a big impact! We had some drill holes approx. 3mm dia. (the default size range for detection is 0.8mm to 3.0mm) and they detected every time but there was some error in the detected position, maybe due to shadows in the lighting? Smaller vias seem to work better. It would be good if the Neoden software showed a circle around the detected point, or maybe print out some data after all the points were calculated. They have a nice big debug window on the mount screen but they hardly write any info to it.

- The Neoden software seems to only detect circles... I couldn't find an option in the menus to change the shape to rectangle, so using a copper feature like a resistor pad is not possible.

- The most unfortunate part is that there is no way to know / change the field of view while detecting the points. Occasionally the machine zeros in on the correct point and then all of a sudden jumps to a nearby hole instead. (It takes about 3 pictures of each point) We got in the habit of starting jobs with the Step function until we were sure it got all the points right. On a dense board I think rails might be the only option to ensure there is enough blank area around the point to guarantee reliable detection. How much blank area? This is still a mystery... it would be nice if Neoden would shed some more light on this.

As for placement, it mostly went as expected. Small parts like resistors and caps are bang on and the pick works every time. We had trouble with some larger components not picking up reliably. The nozzle seems to go up and down rapidly if it can't sense vacuum, but it seems far too rapid. Sometimes it would do this even if it had a part. We changed the vacuum detection threshold to a lower value to try helping it but it didn't seem to make much difference. The biggest problem is with the vibration feeder... if it doesn't think it has a part the nozzle just bobs up and down... if it actually did get a part it smashes that part down on top of the new chip which is already moving down the tube. This just makes a little train wreck all over the place. I'm not sure if the pick and place delay settings affect this retry speed since I didn't experiment with this yet.

The final note I'd like to share is that the nozzle sizes that Neoden suggests for various chips doesn't appear to be right. The smallest nozzle works well for all the passive parts but they suggest what seems like too large sizes for the ICs. After getting repeated pick errors I moved down by one size on all the ICs and things seemed to work much better. Perhaps the larger nozzles just don't get a very good seal on the part.

Anyway, we're still fighting with our oven which made toast out of one of our test boards, but I'd say that the Neoden 4 is workable. The software is still not really at 100%, but it does seem to work enough to make a board. Some more polish would go a long way into making the process enjoyable and save a lot of wasted time messing around with stuff though.

[uncle_bob]

Hi

Not in any way specific to this machine (I don't have one ... yet):

A copper circle off in the middle of nowhere with mask pulled back from it is often the most universal fiducial. That's not just on this specific machine. If at all possible we put them completely outside the main board area so there is nothing of similar size or contrast to confuse things. Off the top of my head, no idea of the exact size we use other than 1 to 2 mm seems to be about right.

The gotcha with drill holes is that the "fine people" who did your board may (or may not) have their drill nicely aligned with the top (or bottom) copper. If you need <0.2 mm alignment (0.5 mm pitch parts) a hole that is 0.2 mm off is ..errr ... not going to help.

Bob

[mrpackethead]

This just makes a little train wreck all over the place. ........ Some more polish would go a long way into making the process enjoyable and save a lot of wasted time messing around with stuff though.

Thanks. This really has confirmed what i was feeling.

[Ice-Tea]

Hi

Ok, so simply put, there really isn't any competition at this price point.

Just a general thought: take the price of the machine, add additional engineering hours lost over the course of 10 years, add addtional components tossed in the bin over 10 years and recalculate.

[rx8pilot]

Hi

Ok, so simply put, there really isn't any competition at this price point.

Just a general thought: take the price of the machine, add additional engineering hours lost over the course of 10 years, add addtional components tossed in the bin over 10 years and recalculate.

This is a critical concept to consider. I even wrote a paper for my industry titled "Cheap Can Be Very Expensive". There are up-front costs that are easy to see and compare,  there are operational costs that are very difficult to see. Over the past decade I have owned a small manufacturing business, I caught myself being suckered in by a 'good deal' that turned out to be very expensive. The unexpected expenses were usually lost time and/or bad parts. In the beginning, it seems like a few minutes here and there. A few minutes here and there every day for years add up to a LOT. It also, as the business owner, set me up to be scrambling nights and weekends to catch up - making up for the poor performance of my software or hardware choices. At that point, the money I 'saved' initially was long forgotten and I was bitter and angry working on a Saturday night.

A 'Long tailed' piece of equipment is one that has a low acquisition cost, but a high cost of operation - ink jet printer as an example. A short tailed piece of equipment may have a high initial cost, but simply works after that for a long time with little additional effort or money. In my commercial manufacturing experience, hardware and software that is focused on being low-cost as a primary design goal, tend to be the most expensive in the long run. A $10k machine seems like a deal, unless you have to fiddle with it, miss deadlines, throw away scrap, burn a relationship with a customer, etc. I have no specific knowledge of the N4, but that is what I would be calculating/estimating if I was considering the product. After some poor decisions, I will do anything to avoid a long tailed piece of equipment or software.