Tronstol E1 experience

[lamabrew]

Now I see that pickup fails as you move down the strip. So far I don't see why.

Two things:

• double sided tape I was using wasn't sticky enough to firmly hold the strip in place.
  
• first and last part wells in the tape were a little twisted due to #1 above.

Redid and picking seems reliable.  I also have ordered the 220 size nozzle (wasn't in the kit I got) as 400 is a bit big.  Calling this one solved.

[lamabrew]

... how you're supposed to pick up a component and have it hold it over the DCAM.

My guess/observations for the procedure for using trays/cut tape sitting in the work area:

• Open Tools->UCam Test
  
• hit Ucam->Open button.  Ucam light should turn on
• Bump the slider to get it to take a new picture of where the head actually is.
• Move the head to center the crosshair over the part you want to use with the DCAM
• Hit "Confirm Position" button.  (nothing will appear to happen)
• Go to Tools->DCam Test
• Adjust "Chip Thickness" to match the part thickness (TBD, this appears semi-functional). Click some other box (not convinced Enter always works) to get it to take the value.
• Hit the Move to DCAM button. Machine will pick up part, but it's going to seriously mash the head down as there's no place to tell it how thick the tray is and it doesn't view this as a component so it can't use the adjusted "pick height" to know how low the head is supposed to go. (and this leads to more problems)
• head should end up over the DCAM
• Hit the down button to have the part moved into the normal scanning position for DCAM (well, no, it doesn't...)
• If you select ROI and hit Algorithm Test it will flash a red circle to show you where it's going to look for the part.  Don't blink or you'll miss it.
• Select the algorithm you wish to test and hit the Algorithm test button
• Wait a second or two
• A bounding box will briefly show in the DCAM display window.  Repeat as needed.
• Without changing anything I found the Threshold Algorithm failed 4 out 5 times, and the Foestner about 1 out of 10
• There are no knobs that I can find to change the imaging results
• There will be some illegible text displayed in the upper left corner of the image window.  Maybe it's supposed to be helpful?
• You can adjust the "chip thickness" (do an up/down after) to change the height that it goes to for the DCAM test, this seems to be the only thing you can control and does make a difference; I could get 100% fails going too low or too high
• Unfortunately the height is a global thing (Changed in Settings -> System Options) so if you find different parts need different heights I think you're screwed
• When you hit any other menu/whatever, the head is going to drop the part into the camera.  I put a piece of paper there to catch it.  Seems like a bug.

Also noted:

• The Rotation Test button moves the head from side to side and does some Algorithm Test. Some number is displayed when it finishes. I believe you should use that number the next time you play the lottery.
• There is a button labeled Save Image.  If it has a purpose, it is unknowable.
• There is a very faint cross hair in the image.  It is not centered on anything but this doesn't seem to affect things.  Purpose of this cross hair is unknown.

The problem is the height you work out here is NOT the height it will hold the part at over the DCAM when you are in File-> Materials.  I think that's because there it's maybe factoring the pick height (which is non-zero to account for the tray/holder thickness) into it's calcs?  Observing where it holds the part it's clearly different than where it did it in the DCAM Test even though the component height was set to match the component height associated with the footprint.

I'm still trying to find a value for Settings-> System Options -> Shooting Height that will work reliably with this particular 'lytic cap.  Right now it seems like it might be possible to get it to work for one part, but then if that setting doesn't work for other parts you're SOL?  Given the E1 feeder system can't handle tall parts IMHO it's kind of critical that "tray mode" be easy to set up and adjust for each part to achieve reliable operation.  Will continue to poke at this in hopes of discovering whatever secret(s) are needed.

[MR]

thank you for the pictures


https://www.eevblog.com/forum/manufacture/tronstol-e1-experience/?action=dlattach;attach=1741112;image

double check that picture it seems like if the motor is almost disconnected or half way connected... push in the motor connector.

[PCBprototyping]

The picture resolution is too low, I can't see the red hatch, can you repost it?

Smudge is from the light coming through the nozzles, try different recognition algorithm, maybe it will ignore it.
Light reflection can also come from some dust on 3D printed camera walls, I already had this problem and it disappeared after I removed this dust with tweezers. Covering the area above and on sides of the laser by hand will help you pinpoint where it's coming from.

For pneumatics on the head, it uses integrated vacuum pump and there's also pressure sensor connected to this line. You can see it's values in manual mode/head.

[lamabrew]

The picture resolution is too low, I can't see the red hatch, can you repost it?

I assume this is the one you had the problem with?  It's 2kx1,5k, and the red hatch shows up well for me in the browser.  Maybe open the link and save it locally?
https://www.eevblog.com/forum/manufacture/tronstol-e1-experience/?action=dlattach;attach=1740233;image

Smudge is from the light coming through the nozzles, try different recognition algorithm, maybe it will ignore it.

By using a pair of tweezers as a pointer I know the ghost in the image is from something about half way between the two heads on the bottom of the plate ( I was way off in guessing it was the needle).  It's probably way out of focus, but whatever it is it didn't come off when I wiped the area.
 
The Forstner algorithm seems to ignore it, but it gets confused by the feet the of the 'lytic cap as they protrude a bit from the plastic foot part of the cap.

I didn't have enough time today to record screen video of the machine trying to pick the cap, I think that is the only way to see it failing vs. trying to describe in words.

I did get to try a small SMT connector and the DCAM seemed to do the right thing - for the rare times the head didn't send the part flying across the room as there's no way to adjust anything in the DCAM test and none of my current nozzles did a good job holding on (I have ordered some different ones).  The body of that connector is tan so the algorithm probably had solid data and didn't get confused by the odd bits hanging off the connector body.

DHL says I'll have the new board tomorrow so getting the 2nd peeler bank working will be the first thing I look at.

[lamabrew]

Did some screen records of DCAM/picking fails and success.

There's 3, the descriptions say what they are.  Summary: caps = fail, connectors = success.

https://youtu.be/5tkmggaozKM

https://youtu.be/clEA1emHqoE

https://youtu.be/eIORyl_S02M

I tried the caps with the 3D sensor; the caps are a little bigger than the 6mm size in the manual but figured was worth a shot.  It seems to not be able to figure out their orientation, always places them skewed.

Passed the info back to NeodenUSA to sort out with Tronstol.  Next up is ripping out the guts to replace the controller board to hopefully solve the right bank feeder/peeler failure.

[lamabrew]

Replacing the controller board solved the busted right side peeler.  Here's a quick video of both sides working:
https://youtu.be/67KOM9j_EfM

Lots of cables in to the main board so it's not a fast thing to take it out and replace. Might have hoped for a simple big header and cable harness but I guess not at this price point.  I added some labels so I wouldn't make an oops.

With that taken care of it's back to trying to find a way to place SMT electrolytics.

[lamabrew]

SMT cap problem solved!

Experimenting with a 2nd cap made it more and more obvious that the ghost reflection was causing all sorts of problems in the system trying to figure out the cap's footprint.  Add a piece of electrical tape under the head between the two nozzle holes to kill the reflection and it seems happy. Though the Forstner algorithm still gets confused so probably stick with just Threshold for now.

Also was able to get some better contrast on the original SMT cap by raising the scan height.  This is not something that has a per part setting, as well as the height in the DCAM test is not the same that it uses in operation (haven't tried to figure that one out yet).  Need to do some more tests of actual pick and place; software crashed during testing...

Here's some screen captures showing the things tried to get to something that seems to work.

https://youtu.be/z08bReq2TwI   4.3x4.3 cap with ROI set to 300, which avoids the ghost reflection on the right side. Locates OK.

https://youtu.be/QpEZArYg1Eo 4.3x4.3 cap with ROI set to 500, ghost reflection on the right side breaks edge search

https://youtu.be/xShbzenNCQk 6x6 cap with ROI set to 300, ghost reflection on the right side ignored and edge search works

https://youtu.be/KL-jun1nOKs 6x6 cap with ROI set to 500 and a piece of black electrical tape added under the head (between the two nozzle holes) removes the ghost reflection; edge search works

https://youtu.be/LjMB7v1t9vk 4.3x4.3 cap with ROI set to 500 and a piece of black electrical tape added under the head (between the two nozzle holes) removes the ghost reflection; edge search in Threshold mode works, Forstner fail

[PCBprototyping]

Scan height can be changed per part by changing the component height. Usually, this is not necessary, but if you need to, you can do it that way, as shooting height(scan height) is calculated with this value and you can trick the machine to go lower/higher.

I think you should try to find the dust that reflects the light and affects the recognition, adding electrical tape is ok, but this is just a coverup, not removing the root cause. Since you cleaned the lenses, I believe it's on the walls surrounding the camera, which are 3D printed and it's rough surface can catch some dust. Happened to me before and I was able to remove it with tweezers.

From what I experienced, red cross being outside of camera center does not affect anything. I'll test some more and let you know.

Instead of DCAM in tools, I mostly use edit/material/pick test to check the dimensions and recognition of components, it will also use the speed settings, so your components don't fly away because of fast movements.
The recognition window always shows the width/length measurements, so you can adjust them accordingly. However, after the recognition, window is cleared, so use the phone camera to see, how the recogniton looked like. After some time, you'll get a filling, what settings are good.

I should mention, that for laser recognition, scanning height is very important for ICs and small electrolitics i.e. For ICs set it to TOP=0.2, so laser doesn't hit the pins and for electrolytics, set it to Bottom=0.2, so laser hits the black plastic part as top is round and laser cannot figure out the rotation, only the center. Max size of components for laser is around 10 mm, it will be less, if component is not picked in center.

[lamabrew]

Scan height can be changed per part by changing the component height. Usually, this is not necessary, but if you need to, you can do it that way, as shooting height(scan height) is calculated with this value and you can trick the machine to go lower/higher.

If I change the component height that's going to throw off the pick and place heights? Or do you mean change the component height to get the shoot height to be better and then change those other two to compensate for the fudged component height?

I think you should try to find the dust that reflects the light and affects the recognition, adding electrical tape is ok, but this is just a coverup, not removing the root cause. Since you cleaned the lenses, I believe it's on the walls surrounding the camera, which are 3D printed and it's rough surface can catch some dust. Happened to me before and I was able to remove it with tweezers.

The reflection is from the head assembly, covering the reflection area with tape making it go away IMHO pretty much leaves no other possibility.  I ran some more experiments, see next post.

I should mention, that for laser recognition, scanning height is very important for ICs and small electrolitics i.e. For ICs set it to TOP=0.2, so laser doesn't hit the pins and for electrolytics, set it to Bottom=0.2, so laser hits the black plastic part as top is round and laser cannot figure out the rotation, only the center. Max size of components for laser is around 10 mm, it will be less, if component is not picked in center.

Manual says 6mm max for the scanner. I did try the 'lytics previously that way and it failed, but perhaps the height was not right.  Regardless I would expect DCAM to work on the caps.

[lamabrew]

I was too optimistic last week, as well as I didn't really think through that I paid for a machine with two heads and taping one off really isn't solving things...I am baffled as to what exactly is causing these reflections.  I could work around it on the caps by setting a small ROI, but that will then break placing larger parts. To me the edge detect algorithm is buggy; even when it does correctly find the edge the feet of the cap protruding a bit cause some small errors in orientation.  Not enough that placement will fail for the cap, but for a fine pitch part it would be off. OTOH those wouldn't have protudy thingees.

If I hold a screwdriver up by the head you can't see it with the DCAM, very little light makes it up there for the way the lighting and camera is set up.  But if I illuminate it with a flashlight you can clearly see the thing that you might think makes the ghost - except if you actually line up the two pictures of the ghosty and the one with the light there's nothing in the right spot. (as an aside, blown up to this size I can read the numbers it computed, but they're not really readable at normal size).

It's clearly something on the head as the pictures with the electrical tape added back shows no problem and the two heads show different reflections. 

The ghost using head 2 is not in the same position as the ghost when using head 1. It does break edge detect. Screen video here: https://youtu.be/zIeSHx1stqM

I can't figure out why Tronstol doesn't see this same problem.  Is there some mechanical difference between my machine and the one they have?  Some software difference?

DCAM photos are in the attached PDF (Forum tools seem too awkward to insert/label screen captures like this.)  I also tried with all lights off and it makes no difference, the DCAM light or something in the head is the source of the ghosty.

[PCBprototyping]

For the head 2 in pdf, reflection is kinda red and it alignes with thin line sticking out of the corner, seems like it's FR4 fibre that's left from the cutting(see photo).
But I can't see anything for the head 1.

Cap recognition is good enough for the placement, there might be some offset as cap pins are typically not very straight and plastic foot can be a bit loose, so recognition frame is not always the same.
Chip recognition is very good from my experience. You can get some misplacement because of some dust on pins if you reuse the IC from the tray(double sided tape makes pins even more sticky) and if pins are bent, so the camera makes bigger frame. Upload a photo of TQFP recognition so I could see if it's ok and in focus.

[MR]

To me the illumination of that part doesn't seem to be right, yes you can get those effects if you put the light the wrong angle.
Try to go there with a torch and add some light in a different angle, maybe add a diffuser (eg. milky acrylic around the leds after the torch experiment).
You need to get a clear picture of the pads via the camera.

I also had optical recognition issues with my Mechatronika machine back then, but those assholes refused to fix those issues. After I wrote my own PP application for that machine I clearly noticed that optical recognition is a hot topic that has to be worked out with the customer if needed, even now I still have to update and enhance my algorithms and detection strategies from time to time. Some components seem to have a very unique production technology which makes nearly every component look unique to the camera.

However your issue seems to be illumination related. You need to get a clear pattern to the camera eye, some weak lines are not enough and you will fall into those 0.1% of customers who might run into such an issue.

Your part should look close to the attached picture (I just modified your picture with Gimp) that's what the optical recognition is looking for.

[SMTech]

Lower end machines seem to have very basic component lighting, if you look at more expensive machines the camera is set back and a "tunnel" of LED's in the camera housing illuminate the image, which tiers of that tunnel illuminate and to what intensity can be adjust to suit the package shape and color you are trying to image. If you have a simple &/diffused light source, you don't have that control and you may simply have to accept that some packages image inconsistently needing very wide acceptable parameters or have high reject rates. As MR show in his doctored image, there's way too much going on in your images, arguably the machine should not even be trying to look at most of those areas or for the shapes that appear there, but then that depends a little on how the machine interprets the camera data and what it measures.  Our previous (low end) machine, did crude edge detection to come up with an XY dimension and have a rough stab at rotation, our Essemtec by contrast can identify feet on legs, BGA balls and package bodies and use all that data to correct alignment and reject for damage (e.g missing balls or bent legs) or rotation outside acceptable limits. It's way better than the cheap method but there are still improvements that could be made, and they could include camera lighting and algorithms, perfect imaging takes a lot of work and sometimes a bit of trial and error to find the parameters that are best.

[lamabrew]

Try to go there with a torch and add some light in a different angle, maybe add a diffuser (eg. milky acrylic around the leds after the torch experiment).
You need to get a clear picture of the pads via the camera.

There's no obvious way to modify the camera/lighting design, it's more or less one large assembly.  My problem seems to be related to some sort of reflection from something in the head - when I tape off the hole(s) things seem to work. Taping off pick head #2 isn't my first choice, I would like to find a better way to solve the reflection problem.  Tronstol has given me two suggestions to try - though I still can't figure out why this seems to be an issue that nobody else seems to have.

It has been suggested by the US distrib that I try and get it just to see the pads vs. the body of the cap.  For me that seems even more daunting than trying to kill the reflection as there's no guidance on how that might be done.  Also I think that until I kill the reflection the edge detection software - in its current state - is going to continue to be confused.

My backup plan is to just cut the ROI down.  If I have a board with small and large parts that need to use the DCAM I'll run the board twice, once with a small ROI and a second pass to do the large part(s) that needs a larger ROI to be seen correctly.  Or if needed place the part by hand as for now this is planned just for use for prototype builds.

[MR]

Here are two pictures of my machine:
Original:


I wrapped some foil from a tray around the lightning source to bundle the light:


from that moment on I never ran into such issues again.

My camera is B/W but you can clearly see the light is bouncing off the fiducial in the corner, it was a HASL-leadfree PCB where the surface is not 100% even. Someone could argue that this is something a manufacturer should take care about since they earn their money with those machines but there are just so many topics the smaller manufacturers have never seen that you shouldn't wonder.

If your pad of the component is so dark the light is not where it is supposed to go.

Who knows where the light of your illumination goes to, certainly not to the camera sensor.

[PCBprototyping]

Try to go there with a torch and add some light in a different angle, maybe add a diffuser (eg. milky acrylic around the leds after the torch experiment).
You need to get a clear picture of the pads via the camera.

There's no obvious way to modify the camera/lighting design, it's more or less one large assembly.  My problem seems to be related to some sort of reflection from something in the head - when I tape off the hole(s) things seem to work. Taping off pick head #2 isn't my first choice, I would like to find a better way to solve the reflection problem.  Tronstol has given me two suggestions to try - though I still can't figure out why this seems to be an issue that nobody else seems to have.

It has been suggested by the US distrib that I try and get it just to see the pads vs. the body of the cap.  For me that seems even more daunting than trying to kill the reflection as there's no guidance on how that might be done.  Also I think that until I kill the reflection the edge detection software - in its current state - is going to continue to be confused.

My backup plan is to just cut the ROI down.  If I have a board with small and large parts that need to use the DCAM I'll run the board twice, once with a small ROI and a second pass to do the large part(s) that needs a larger ROI to be seen correctly.  Or if needed place the part by hand as for now this is planned just for use for prototype builds.

Tronstol DCAM illumination works differently than most of machines on the market, it's side illumination instead of up looking, so you will never get just the pins illuminated if body is almost at the same level. You cannot change the illumination values for DCAM.

As mentioned, head 2 reflection comes from hairline at the corner of head protection board(I believe it's fibre left from the cutout process), you can see that it's at the same spot at both illuminated and non illuminated image and it's kinda red on both images. Try to remove this hairline with the tweezers or something, for sure this is a problem for head 2.
It's not true, that nobody had the reflection problem like you have with head 1, I had it and solved it by thorough investigation where it's coming from and by removing a dust that stuck on the 3D printed groves surrounding the camera(see photo attached). Seems your dust is somewhere else, but it's definitely removable.
Go step by step:
- remove the DCAM lens. Dust, no dust?
- remove the illumination and laser board surrounding the fiducial cameras. Dust, no dust?
- put a tape on side walls of DCAM. Dust, no dust?
- add some none transparent cover over machine cover. Dust, no dust?

There's small chance that the dust is on the camera itself, but I doubt.

[lamabrew]

Here's a picture of the (upward looking) DCAM


(separate question, how do you get inline images without them being attachments?  I can not figure that out despite RTFM)

[lamabrew]

Here's a video of "forcing" the failure with Head 2 holding the part (and offset a bit by using the Rotation button, whatever that is for is unknown) by making the lower standoff on the head assembly more reflective:  https://youtu.be/czEh59rQVpA

Some pictures attached of other mods that seem to allow the cap, when held in Head 2, to always be detected correctly:

Sharpied the hex standoff, then rotated it a bit so it wasn't reflecting directly back down.

Covered the stepper motor screw heads with electrical tape, that got rid of more. (but it's not going to hold, need to find something different).

Blacked out the inside of the white cover just in case, but I don't have any indication that was actually a source of a reflection.

Sharpied all of the cutouts on the head too.  Not perfect but it reduced the problem enough that it *seems* like the edge detection isn't getting confused.  Who knows how it will act tomorrow.

IMHO this is a design issue with the E1.  They need to black out anything around the head assembly.  Making it so the thresholding doesn't get confused would help.  IMHO for parts being scanned with the DCAM camera there should be a ROI setting for the part, that way we could fine tune where it's going to look.  Probably would speed up processing too.

[PCBprototyping]

See attached photo, circled is the hairline that I'm talking about and you need to remove. It clearly made a reflection in the pdf photos. Please, remove it as it makes no sense for me to help you, if you don't want follow the suggestions.
What I noticed also is, that you have the nozzle with velvet protection in head 2. Some dust gets stuck in this velvet and causes the reflection, I had this problem with Neoden K1830, which uses up shooting illumination, so it shines directly on the velvet. I cleaned the velvet with compressed air and it went away.
In the video, reflection seems very straight cut on the right side, which means that it's coming through the board surrounding the heads as the edge of the board makes straight edge in reflection.
I agree that area around the camera should be more protected, but from my investigation it would be enough to make white cover surround the whole area, not just the front and making plexiglass less transparent.

[lamabrew]

See attached photo, circled is the hairline that I'm talking about and you need to remove

Sorry, that was a "before" picture.  It's gone now.  Attached captures of what DCAM sees now with things taped off, sharpied, etc.

However despite the testing in the DCAM test being 100% in terms of part orientation/footprint, pick tests or place tests are only good about half the time.  I suspect slowing things down more might help.  When I started DCAM this quest was almost always failing and after the mods DCAM (in the test dialog) seems to always work.  So I'm hopeful I'll get the placement to be reliable too.

[Styno]

To me it looks like it's more a problem with the lighting (and reflectivity of the capacitor's soldering tabs) than a speed issue. A bit more contrast would help i think.

Edit: Sorry, this was already mentioned earlier. 

[lamabrew]

Just when I thought I could move on...

Ghosty #1 is back...I switched the 220 nozzle to head #1 and put the #65 in head #2.  From trial & error I found that increasing the exposure 1 unit (no idea what that means IRL) the edge detect worked better, but that also increased the reflection brightness (from the multiple sources) and more or less borked using head #2 - hence my switch back to head #1.

#65 has a smaller ring (not sure what you want to call it) so the camera can see more in the hole for head #2.  Figured out that the reflection is from the shank of the nozzle holder (see red circle in pix).  I was able to isolate that by holding a small piece of black tape with tweezers and moving it around, and then comparing the view with and without the extra light.  I can't put tape their as it would interfere with the nozzle pickup (the reflection is from the bottom of the tailpiece on the head). I'm not sure I want to paint around there either as I wouldn't want to get paint up in the barrel of the tailpiece.  IMHO the fix here is Tronstol needs to anodize all the "shiny" things (shafts, tailpiece, screw heads, standoffs, etc) flat black to kill the reflections.


I think without that the E1 DCAM is never really going to work right. (or they get better software that can figure out to ignore these things)

For now I'm going to go back to cutting the ROI down to keep the reflections outside of where the camera is looking.

[lamabrew]

Came up with one more thing to do to show the problem is inherent in the E1. Take a black Sharpie and scribble all over the tailpiece/shank for head 2. (picture is for head #1 holding the cap)

Reflections go way down.  TBD if it's enough to stop breaking the edge detection, I trust the ROI reduction more than this hack.

[lamabrew]

Had to set the E-1 aside again, but now need to put some simple boards together.  During setup I had it swap nozzles on Head 1 (has the autochanger option).

It made a funny noise doing that and now Head 1 is bent sideways and actually loose, i.e. I can pivot it a bit. Guessing maybe a screw popped off somewhere?  Will post when I figure it out.

For now will just use head 2 to build things but annoying that it keeps having issues.  The black tape and black marker you see in the pictures are the fixes to make the camera work, i.e. the shiny metal parts reflect light down into the camera and the software barfs trying to figure out the footprint.