Tronstol E1 experience

[PCBprototyping]

Why do you keep covering the reflection and don't fix it? If you cannot find it or remove it, this is not your problem and don't accept the machine as is. It just makes you disapointed and gives other users a wrong information about the machine.
Find attached the images of how it should look like, you can see all black around the nozzles and TQFP and images were taken under normal office lights, no reflection. Recognition is perfect.
For bent head 1, if you swing it to the left, does it stay straight? To check if it's bent, swing it to the left, insert a big, flat nozzle into head 1(CN750), use manual mode to lower the head above the PCB rails and slowly approach the rails. When you're i.e. 0.5 mm above the rails, use the manual rotation in both directions to see, if the nozzle is parallel to the rail(check left and right + front and back parallelism). If it's ok, head is not bent, but the head guide rails are probably loose, maybe screw fell out because of transport or sth. Remove the cover and see, if you can identify a fault.

[lamabrew]

Why do you keep covering the reflection and don't fix it? If you cannot find it or remove it,

Sorry, I'm not following.  The source of the reflections are all of the shiny things in the head assembly:

• hex standoffs
• screw heads
• collar assembly

The fix is to black them out, in my case some electrical tape and sharpie did the trick.  The actual fix would be for Tronstol to anodize the parts black.  Cutting the ROI down helps, though if it had to place large parts then that trick can't be used.  A second fix would be to have the software support setting the ROI of each footprint as well as provide some way to set thresholds/brightness on a per part basis.

These reflections are so far only a problem on certain parts that are smaller and low contrast, specifically SMT electrolytic caps.

The TQFP example is not equivalent to what happens with a 6x6mm SMT cap with a very dark plastic piece; the large TQFP parts covers a lot of reflection sources as well as it's quite a bit brighter and the contrast much higher.

. Remove the cover and see, if you can identify a fault.

The motor shaft does not appear bent, but I also can not push the motor back in position to make it plumb.  I'm guessing something has gone wrong with the guide track/slide, in the video you can see Head 2 is rock solid where as Head 1 is easy to make wobble.

https://photos.app.goo.gl/eDyDMtSW4xYcBz8z8

The motor shaft itself does not seem bent, rotating it the nozzle stays in the same offset position
https://photos.app.goo.gl/2jt6dZb56nnmJUYT8
 
It's not exactly clear how to take the head apart further so I've handed it back to the distri to sort out with Tronstol.  For now I'll just run with Head 2 so I can finish up a project.

[PCBprototyping]

The TQFP example is not equivalent to what happens with a 6x6mm SMT cap with a very dark plastic piece; the large TQFP parts covers a lot of reflection sources as well as it's quite a bit brighter and the contrast much higher.

That's why I added the pictures without TQFP also, just the nozzle which is small and similar to electrolytic capacitor. The recognition doesn't catch any reflection around it and correctly makes a frame around the nozzle.


For the head 1, you can see in the video, that guide rail is moving, so it's not fixed and that's why the head swings.
To check the parallelism, rotating the head by hand is not good enough to see if nozzle is parallel to working area or not. Use the SW manual mode, like I mentioned.

[lamabrew]

That's why I added the pictures without TQFP also, just the nozzle which is small and similar to electrolytic capacitor.

The nozzles have a reasonably bright ring around them that isn't there with the caps.  I think the reflective feet of the cap makes whatever processing the software is doing turn the gain down. For example: https://www.eevblog.com/forum/manufacture/tronstol-e1-experience/?action=dlattach;attach=1740245

The recognition doesn't catch any reflection around it and correctly makes a frame around the nozzle.

I have no explanation as to why your machine acts differently.  I took your pictures and boosted them (see attached example), you have some of the reflections I see but at a much lower level. And some I have don't show up at all on your screen captures.  Maybe your machine was built with different things than mine?  The one the US distri has is different than mine.

There are no knobs on the DCAM that I can find to adjust exposure, etc.  Regardless, black tape and sharpie fixed it for me.  Hopefully someday more people with an E-1 will join in and we'll have another data point(s).

On the head 1 being angled & wobbly, I'm going to let Tronstol tell me what to do. It looks like I can take the (head rotation) motor off with the 4 bottom screws and I assume there will be screws under where the motor sits that attach the bracket to slide/linear bearing, and assume taking that apart will reveal something loose or broken.

Between the blown controller board, the defective left feeder due to factory assembly error, fighting with DCAM and reflections, and now head 1 busted, all in less than 40 hours of the machine being powered on, has been just a tad frustrating.

I tried doing a 'work' run with only head 2 enabled, but the software partially ignores that and tried to have head 1 pick up a nozzle anyway, which fails (mechanically) and wedges the software (i.e. Replacing nozzle dialog just sits there) and requires a power cycle to get things cleared.  By setting all slots of the autochanger to empty I was able to stop head 1 from trying to pick up a nozzle.  Clearly software bugs.

However the software would then only place the parts that matched the nozzle in head 2 and then decide it was done, never offering up changing the nozzle manually. Then it gets confused if I restarted the run with a different nozzle as most of the footprints are set for more than one nozzle size and it wants to place those parts a second time with the new nozzle.  I went back to edit mode and just did the rest that way as it was a small board.  I don't see what I could have set wrong to not have it ask for the next nozzle without losing track of what was already placed.  Buggy software I guess.

[Loky22]

Hallo lamabrew , thanks for your write experience ,
I need take decision for pick and place too and i thought that this machine was better of Neoden4  all do think so . But with real your  experience i see than isn't so.
Neoden4 i see work and is very well problem is old software and dont can upgrade .





 

[PCBprototyping]

That's why I added the pictures without TQFP also, just the nozzle which is small and similar to electrolytic capacitor.

The nozzles have a reasonably bright ring around them that isn't there with the caps.  I think the reflective feet of the cap makes whatever processing the software is doing turn the gain down. For example: https://www.eevblog.com/forum/manufacture/tronstol-e1-experience/?action=dlattach;attach=1740245

The recognition doesn't catch any reflection around it and correctly makes a frame around the nozzle.

I have no explanation as to why your machine acts differently.  I took your pictures and boosted them (see attached example), you have some of the reflections I see but at a much lower level. And some I have don't show up at all on your screen captures.  Maybe your machine was built with different things than mine?  The one the US distri has is different than mine.

There are no knobs on the DCAM that I can find to adjust exposure, etc.  Regardless, black tape and sharpie fixed it for me.  Hopefully someday more people with an E-1 will join in and we'll have another data point(s).

On the head 1 being angled & wobbly, I'm going to let Tronstol tell me what to do. It looks like I can take the (head rotation) motor off with the 4 bottom screws and I assume there will be screws under where the motor sits that attach the bracket to slide/linear bearing, and assume taking that apart will reveal something loose or broken.

Between the blown controller board, the defective left feeder due to factory assembly error, fighting with DCAM and reflections, and now head 1 busted, all in less than 40 hours of the machine being powered on, has been just a tad frustrating.

I tried doing a 'work' run with only head 2 enabled, but the software partially ignores that and tried to have head 1 pick up a nozzle anyway, which fails (mechanically) and wedges the software (i.e. Replacing nozzle dialog just sits there) and requires a power cycle to get things cleared.  By setting all slots of the autochanger to empty I was able to stop head 1 from trying to pick up a nozzle.  Clearly software bugs.

However the software would then only place the parts that matched the nozzle in head 2 and then decide it was done, never offering up changing the nozzle manually. Then it gets confused if I restarted the run with a different nozzle as most of the footprints are set for more than one nozzle size and it wants to place those parts a second time with the new nozzle.  I went back to edit mode and just did the rest that way as it was a small board.  I don't see what I could have set wrong to not have it ask for the next nozzle without losing track of what was already placed.  Buggy software I guess.

That's why I added the pictures without TQFP also, just the nozzle which is small and similar to electrolytic capacitor.

The nozzles have a reasonably bright ring around them that isn't there with the caps.  I think the reflective feet of the cap makes whatever processing the software is doing turn the gain down. For example: https://www.eevblog.com/forum/manufacture/tronstol-e1-experience/?action=dlattach;attach=1740245

The recognition doesn't catch any reflection around it and correctly makes a frame around the nozzle.

I have no explanation as to why your machine acts differently.  I took your pictures and boosted them (see attached example), you have some of the reflections I see but at a much lower level. And some I have don't show up at all on your screen captures.  Maybe your machine was built with different things than mine?  The one the US distri has is different than mine.

There are no knobs on the DCAM that I can find to adjust exposure, etc.  Regardless, black tape and sharpie fixed it for me.  Hopefully someday more people with an E-1 will join in and we'll have another data point(s).

On the head 1 being angled & wobbly, I'm going to let Tronstol tell me what to do. It looks like I can take the (head rotation) motor off with the 4 bottom screws and I assume there will be screws under where the motor sits that attach the bracket to slide/linear bearing, and assume taking that apart will reveal something loose or broken.

Between the blown controller board, the defective left feeder due to factory assembly error, fighting with DCAM and reflections, and now head 1 busted, all in less than 40 hours of the machine being powered on, has been just a tad frustrating.

I tried doing a 'work' run with only head 2 enabled, but the software partially ignores that and tried to have head 1 pick up a nozzle anyway, which fails (mechanically) and wedges the software (i.e. Replacing nozzle dialog just sits there) and requires a power cycle to get things cleared.  By setting all slots of the autochanger to empty I was able to stop head 1 from trying to pick up a nozzle.  Clearly software bugs.

However the software would then only place the parts that matched the nozzle in head 2 and then decide it was done, never offering up changing the nozzle manually. Then it gets confused if I restarted the run with a different nozzle as most of the footprints are set for more than one nozzle size and it wants to place those parts a second time with the new nozzle.  I went back to edit mode and just did the rest that way as it was a small board.  I don't see what I could have set wrong to not have it ask for the next nozzle without losing track of what was already placed.  Buggy software I guess.

Nozzle reflection:
Please, accept finally, that our machine works and yours doesn't and send it back to Neoden USA or request their on site service! It is in warranty and it's not your problem if it came with a dust inside which CLEARLY affects the recognition. Tronstol shows electrolytic capacitor placement in their application video and you can see, they don't have this refelction and placement works just fine:
https://youtu.be/92ktbBHFBig?t=436
Focus on solving the problems, not on searching and discussing why your machine works like it works, because you can clearly see that it has a fault. You are just waisting my time, I told you what to do with my longtime experience in protototyping machines, I can't endlessly reply to your emails and try to convince you to finally send the machine back to Neoden USA and request a repair or replacement. In Europe, you have a few days of time to report faulty machine after the purchase and you recognized a fault immediatelly. I'm not sure, what is the claim period in USA, but I'm sure, buyer is protected more than a seller.
Hope you finally let others solve your problem. You are a user, not a service guy. Thank you.

The boosted reflection is filtered out and doesn't affect the recognition. However, image is not the same as it shows just the area around the needle which can be seen through the cutout, but your image clearly shows an "intruder", which is not part of normal machines and this source of reflection needs to be removed by a service guy.

I can't comment on what happened with head 1, but this is also problem for your seller, so don't waste time thinking about it and let others do the repair.

[MR]

> Please, accept finally, that our machine works and yours doesn't and send it back to Neoden USA or request their on site service!

before issuing such a statement, how about he sends you a strip with those components so you can doublecheck?
My experience is some users use components which might be tricky - which means you just did not run into that issue yet with your projects.

[lamabrew]

You are just wasting my time, I told you what to do with my longtime experience in protototyping machines, I can't endlessly reply to your emails and try to convince you to finally send the machine back to Neoden USA and request a repair or replacement

Apologies, I do not intend to waste anyone's time with these posts.  I am simply relaying my experience with getting an E-1 set up to do small prototype builds.  My reasons were two-fold:

• Get feedback from other users about their experience(s)/be a single common thread on EEVBlog for E-1 users
• Document my journey from going from a bunch of boxes at my door to finished PCBs

I really appreciate the time you have taken to explain/show what your machine does.  Please do not feel you must/need to respond to these posts.

Your machine clearly has some differences from mine (i.e. head assembly reflections) and you didn't experience the mis-assembled feeder, blown controller card, and failed Head 1. The US user that I spoke with before purchasing also did not experience any of those things, so I can write this off as a certain amount of bad luck on my part combined with lack of good QC at Tronstol at the time my E-1 was assembled.  Which was almost a year ago - I'll spare the details on how/why it took so long for me to get it, etc.  I can believe Tronstol has changed/improved things in that time period.

As to returning the machine, that is not practical on many levels.  Mostly the costs involved in trucking a unit, as well as I need it.  I have actually had good results with it assembling so far, I haven't had a chance to post that as well as so far it's been simple things and I want to try a more complex board next week.

There's no need for the US distri to come on site, Tronstol DHL's me the replacement parts and makes a video showing the process to do the repair.  I learn more about the machine this way, though obviously not having to learn how to take out the whole guts of it to replace something would have been the preferred outcome.  NeodenUSA has helped smooth the whole Tronstol help cycle and at this point is very invested in the support.

I do hope Tronstol can keep improving the software to deal with the instabilities as well as make certain things more efficient.  Providing some (per part/footprint) knobs for DCAM to deal with the range of devices encountered would be a huge help.

[PCBprototyping]

> Please, accept finally, that our machine works and yours doesn't and send it back to Neoden USA or request their on site service!

before issuing such a statement, how about he sends you a strip with those components so you can doublecheck?
My experience is some users use components which might be tricky - which means you just did not run into that issue yet with your projects.

Find attached the recognition photos for a bit smaller cap. Recognition is good with both algorithm methods.

[lamabrew]

I see two big difference in the cap in the above vs. mine:
- Base on my cap is not rectangular (see attached picture)
- Leads stick out a bit

Here's a video of a typical failure (after killing the reflections)
 

In that run the Threshold alg always worked, but I have seen it fail sometimes too.  The Forstner edge detect seems to fail almost all the time and on pretty much everything, not just caps.

[PCBprototyping]

Rectangular or square doesn't matter, the biggest difference I see is that image is much better on my picture, maybe your shooting height is too low or too high?
Component doesn't need to be recognized with both algorithms, that's why it's possible to select the best one in package library.
First try to get a better image with shooting height and then try treshold algorithm again, maybe this will correct occassional failed detection. If it still occurs, video would help a lot to see what was going on. It would help Tronstol also to correct the SW if there's a bug.

You can let me know the P/N of the cap you use and I'll buy it here in Europe to show you cap is not a problem.

[PCBprototyping]

Hallo lamabrew , thanks for your write experience ,
I need take decision for pick and place too and i thought that this machine was better of Neoden4  all do think so . But with real your  experience i see than isn't so.
Neoden4 i see work and is very well problem is old software and dont can upgrade .

I used both Neoden 4 and Tronstol E1 and I can say that E1 is much more reliable, precise and with better SW.
Neoden has bad feeder design, which shakes components or lets the plastic tape fall off the support. SW and old Windows platform is also a big drawback.

Problem from Iamabrew is not the machine problem, but some dust or sth creates reflection on his machine, which is clearly identifiable in recognition.
You can send me a board and some components and I'll try them for you, if you want.

[MR]

> Please, accept finally, that our machine works and yours doesn't and send it back to Neoden USA or request their on site service!

before issuing such a statement, how about he sends you a strip with those components so you can doublecheck?
My experience is some users use components which might be tricky - which means you just did not run into that issue yet with your projects.

Find attached the recognition photos for a bit smaller cap. Recognition is good with both algorithm methods.

from my point of view this is also a poor picture.

You're doing something like canny algorithm on the picture, that's also not the best.
The problem is the illumination source, the source is inappropriate for this component, there are quite a few ways that can be tried to fix this issue.

https://www.vision-doctor.com/en/

this website has some valuable information about light sources.

In particular this one:

https://www.vision-doctor.com/en/illumination-techniques/diffuse-incident-light.html

look at the dark spots on the foil, looks familiar since the light is bouncing off and not entering the camera.

[PCBprototyping]

Surface recognition, that vision doctor shows is not important for this cap recognition, as we just need a clear edge to make a frame around it, not surface view. Tronstol E1 uses dark field illumination, which is also described on Vision doctor as an example of machine vision for edge illumination(https://www.vision-doctor.com/en/lighting-for-machine-vision.html). Frame is right on the edge of pins and plastic body, so goal has been achieved. SW calculates the center by calculating mid point of shorter and longer side of frame and, as the frame is precise, center will also be precise.

Keep in mind this is not a 150k pick & place before you dive deep in sofisticated camera, illumination and recognition solutions.

It does edge detection, but doesn't do pin area recognition, etc. So for non-simetrical parts you will need to define offset between center of pad area and center of body + pad area(for flat ribbon cable connectors i.e.) for correct pacement. It can be calculated easily, you just do a test pick-up which will show you the dimensions, recognized by edge detection and calculate the center. Offset is the difference between pin + body center and pick&place file center.

Surface recognition is done by fiducial camera for fiducial recognition, automatic center of the pocket detection, bulk recognition and feeder bank calibration and this one uses top illumination and OpenCV to recognize objects.

[lamabrew]

First try to get a better image with shooting height and then try threshold algorithm again,

What height are you using?  I should probably start with that value.

You can let me know the P/N of the cap you use and I'll buy it here in Europe to show you cap is not a problem.

That's very kind of you to try this out. There's two SMT caps that I've tried:
Panasonic    EEE-1VA101XP      https://octopart.com/search?q=EEE-1VA101XP
Panasonic    EEE-FK1E100AR    https://octopart.com/search?q=EEE-FK1E100AR

[lamabrew]

Some more lessons learned...

Despite just having one working head, I was able to work around the problems described earlier (1] software still tried to pick up a nozzle in the disabled head and 2] would not keep track of what was already placed when a footprint allowed more than one nozzle) and build some simple boards.  The fix for #1 is set the autochanger holder to empty for all slots.  The solution to #2 is only define one nozzle per footprint.

I've learned two more things:
1]
Reverse LEDs, i.e. ones where they will shine down through a hole in the PCB so are in the tape upside down from normal LEDs, can not be scanned with the 3D scanner.  But the DCAM (upward looking camera) is able to figure it out almost all the time.  For the few failures it's easy to fix; if i was doing lots of boards I would figure out why it occasionally missed one but with only 40 LEDs total to place across the run simpler to just correct after placing.

2]
Double check the DCAM camera well before starting a run.  As noted earlier the DCAM manual part check has two bugs: 1] can't tell the system the pick height so it just mashes the head into the part and 2] When you finish (i.e. go to some other function) it shuts off the vacuum and drops the part in to the camera.

I was catching them with a piece of paper but I clearly missed some.

That means a couple of parts ended up on the glass cover over the DCAM, and since this is well below the illumination ring they do not show up in the pictures.  And being way out of the focal plane they were making funny dark blotches that I didn't notice.  They did however make the DCAM totally unreliable, I suspect head movement shaked them around enough to have them be in the way sometimes.

Removing those from the DCAM camera well let me complete the board build.

[PCBprototyping]

First try to get a better image with shooting height and then try threshold algorithm again,

What height are you using?  I should probably start with that value.

You can let me know the P/N of the cap you use and I'll buy it here in Europe to show you cap is not a problem.

That's very kind of you to try this out. There's two SMT caps that I've tried:
Panasonic    EEE-1VA101XP      https://octopart.com/search?q=EEE-1VA101XP
Panasonic    EEE-FK1E100AR    https://octopart.com/search?q=EEE-FK1E100AR

I use almost the same part as EEEFK1E100AR.

It's 5x4.4, height 5.6 mm, I've just set the shooting height in DCAM tool to 5 mm and that's it. Today I measured the height precisely by touching it with the nozzle and then touching the tray height to calculate the difference and I got 5.6 mm.
I get 5.00 x 4.36 with Foerstner and Threshold on DCAM. The difference from last photos is because of 0.6mm height difference as image was bigger than at 0 focus and scaling is therefore wrong. However this doesn't affect the center calculation as everything is proportionally bigger/smaller.
Today I even tried the laser recognition of this cap as it's small enough to fit the laser aperture. It works fine, you just need to change the laser shooting height as default setting is TOP=0.2 mm, which hits the round body and doesn't give you the rotation info. You need to hit it at pin or plastic body level, BOTTOM=0.1 worked for me, dimensions were 4.95 and 4.33.


Just check the height in PDF, enter this value in tools/DCAM/chip thickness and that's it.

I see no reason to buy so similar component to mine, but I'll do it anyway and let you know the result.

Instead of Tools/DCAM, I suggest you to use edit/material/pick test to check the component dimensions and recognition as it's much more cenvenient to work with(use the tray mode so it will put it back after the recgnition). You won't need to manually add the component to the nozzle, it's easy to change package and recognition parameters and it will always put the component back to pick location, so you can try the recognition repeatedly and adjust the parameters.

Can you send the link to reverse LED, you use, so I could check how it looks like?

[lamabrew]

Sorry, I wasn't clear on which height I wanted to know your value for.  I meant the "Shooting Height" in Settings-> System Options

I didn't spend much time on the smaller cap, I don't think I saved any results. Most of my effort was on the 100 uF (6.6x6.6mm base) as that was on the board I was building.

Instead of Tools/DCAM, I suggest you to use edit/material/pick test to check the component dimensions and recognition as it's much more cenvenient to work with(use the tray mode so it will put it back after the recognition). You won't need to manually add the component to the nozzle, it's easy to change package and recognition parameters and it will always put the component back to pick location, so you can try the recognition repeatedly and adjust the parameters.

The Pick test only briefly flashes up the results and my timing to do the screen capture was not good; doing it in the DCAM test it leaves the image up so much easier to capture it.

I see no reason to buy so similar component to mine, but I'll do it anyway and let you know the result.

Agree, Tronstol should be the ones doing this. 

Can you send the link to reverse LED, you use, so I could check how it looks like?

Wurth 156125RS75000 https://octopart.com/search?q=156125RS75000
Wurth 156125VS75000 https://octopart.com/search?q=156125VS75000

They're the same part as the normal ones, they're just in the tape with the dome side facing down. The ones here have a rectangular dome, some versions have rounded domes. Don't know how much difference that would make with DCAM but might be an issue for the 3Dsensor?

[PCBprototyping]

I did not change the shooting height in settings, I did not mean shooting height as a parameter, but actual height where head stops inside the illumination ring. Just play around with component height in tools/dcam and you will see, which one gives you the best image. Shooting height(-26 mm) + chip thickness(5 mm i.e.) gives you head height value(-21). Change chip thickness by 0.1 mm in both directions and you will see the difference. Then use this value as component height in the library. Note, that you need to press enter after each change, otherwise the value won't be updated.

These reverse LEDs are perfect for the laser, I'believe you have a problem, because your shooting them too low with a laser, most probably your setting is TOP=0.2, which will hit the component 0.2 mm below the nozzle and since the LEDs base height is 0.28 mm and has some gaps, you get inconsistent reading(see attachment). Change shooting height to TOP=0.1 and check it.
Shooting the LEDs at lens height might work also, if laser beam doesn't go through the lens, but for sure won't work for round lens as they won't give you the rotation. I recommend shooting it at base height.

[lamabrew]

I did not change the shooting height in settings

Just to make sure I have this right, you left it at the 33 default shown in the manual?

These reverse LEDs are perfect for the laser, [snip] I recommend shooting it at base height.

Makes sense.  Though at this point I'm just about done so I'll save this plan for the reverse LEDs for "next time" as it's good enough with the DCAM for now.

Looks like it will be next week when I get to doing a bigger/more parts board - I still have to pull Head 1 apart to figure out why it's loose and not going to be in the office part of this week.  I'll post what I find.

[PCBprototyping]

It's at -32.5, like it was set from the factory.

[lamabrew]

Update on Head 1 wobble/non parallel operation:

Took things apart over the weekend and turned out the 4 screws that hold the head #1 rail were loose.  Tightened them up, reassembled, realigned the heads, and seems to all be OK now.

How they loosened up so quickly is a mystery.

Getting the motor out was a bit painful and getting it back in was even more painful as the 4 screws that hold it in are very difficult to access with the head on the machine and the bottom plate (3D scanner) in place. I was not inclined to try and take all of that apart to get the motor off. To get at the four rail screws you do need to remove the motor, though I did not need to remove the carriage bracket from the belt to get to the screws.

While everything was opened up I went at everything shiny with the black sharpie and added black heat shrink to the hex standoff to further kill the reflections.

[TJ232]

Hi lamabrew,

Did you managed to fix the above issues?
Did they solved your feeders problem?

It looks like a decent destktop size/weight office machine and despite the higher price for it's class might be a good option if working as advertised.

Any other E1 users around?
Please feel free to share your personal experience.

Thanks,
TJ.

[lamabrew]

Did you managed to fix the above issues?
Did they solved your feeders problem?

It looks like a decent destktop size/weight office machine and despite the higher price for it's class might be a good option if working as advertised.

All of the issues I had have been resolved.  Of course it would have been great not to have had these in first place. I see them as more a QC/QA failure than a fundamental problem with the design of the machine.  Having dealt with similar "off shore" equipment suppliers I went in to this with the expectation that things would come up.  That being said even the big $$$ machines have issues - at least based on conversations with a local CM.

The feed mechanism operates what I would call reasonably well.  There's defintely time needed to futz with settings for certain parts to get reliable pickups but once set things stay stable.  The more I use it the better I'm getting at guessing what/how to tweak.

There are defintely some things that I would call bugs in the software, and there's a number of things they could do to make the system easier to set up and run.  None of them prevent me from assembling boards, though I'm only using this for small batchs (< 20), and mostly for prototypes.  I think it probably would be fine for making 100s of not too complex boards at a time.

My designs are generally conservative with smallest parts being 0603s, though there's 0.5mm pitch QFNs. Placement accuracy seems good and I doubt 0402s would be an issue.  It places 0201 well on the test board, but I have not actually built things with those. My gut feeling is that if I did have a problem I could slow things down to get better repeatability - for me absolute speed isn't a concern.

Loading the reels is a relatively slow process so if doing a high product mix that would be something to consider. The cost savings from not having to purchase the standard feeders is pretty big, and the reduction in CPH from having the pin feed is a total don't care for me.  I am considering picking up another feeder bank so I can swap out a side at a time.

[TJ232]

It's good to know that at least they provide support for such situations.
Did they change the broken parts?


It is working with 2 or 3 fiducials recognition?

How is repeatability? How many total boards did you made in one session

Regarding feeder banks, is it possible to change them on-the-fly or PNP machide need to be fully shutdown before change.