Release candidate for new software version, 20.05.2021

[AnalysIR]

Good news...I just completed a full panel with the latest RC & using the TMC2209 stepper upgrade.

Issues:
- Most were related to me mishandling after placement
- I couldn't get the nozzle to pick-up 10uF caps 0805 using nozzle #2 or #3. As these parts are not fully flat it may be a lack of vacuum power. I remember this before on occasion. I had a similar issue with smaller 4.7uF caps but resolved this using nozzle #2 instead of #3. I will look into this more later, but unlikely to be related to this RC.
- As above editing 'one tape' gave an error, so I just used the tape grid to manually edit. (Occurred on previously defined tapes only)


I also tested recognizing 0805 resistors & caps last evening and it seems to work just trying to use vision to recognize the (very very small) top pad on the component. The advantage of this method is that angular offset is reported which you don't get with hole or pocket.(I didn't use this on this job, but did use pocket recognition wherever possible.)


Question: does the centre x/y offset make use of the angular offset or does it just adjust vertically/horizontally along the XY axes?

[JuKu]

Still not 100% clear. If I jog, is the new position accepted or must it still go thru the vision? Good to know that I can redo the vision, return and not loose context.

You must go to thru the vision. The red pop-up dialog (red to catch your attention) is an interrupt to the measurement routine, so the measurement is still under way behind the scenes. You can jog the machine to a position and/or tune the algorithm so that when you click retry, the interrupted measurement routine tries again and has a better chance to succeed.

Please suggest clearer wording, if you can think some!

[AnalysIR]

Its clear now.

Some of my confusion was that the user manually jogs to the correct position...so why not just accept that position & thus avoid having to (re)run the vision stuff.

....will also be a good launching point for additional/future features as discussed. (skip to next component position or even an alternative tape etc)

[AnalysIR]

The red pop-up dialog (red to catch your attention)

This pops up over the vision functions when changing tab. =>annoying because I have to move it every time.

Might be better to pop it up around top right or bottom left of window/screen. Alternatively, remember where the user moved it to last and and always use that.


I did a lot more testing with identifying 0805 pads instead of holes/pockets. Although it works it can vary a lot with ambient light. So I will stick with pockets for now (LEDs seem OK as the contrast is better and can be got using the full component outline.)

Question: Is there a way of getting 2 pads recognised using vision? like an 0805 resistor/cap.

Z limit check boxes:
I noticed at one point that I got an error saying these were not set. I didn't change them. So I just checked them again and continued on.
I can't see anything that might have caused that, but I did hit the e-stop button a few minutes before. I vaguely remember it happening before with this RC.

[JuKu]

Z limit check boxes:
I noticed at one point that I got an error saying these were not set. I didn't change them. So I just checked them again and continued on.
I can't see anything that might have caused that, but I did hit the e-stop button a few minutes before. I vaguely remember it happening before with this RC.

You hit the e-stop on the right moment during probing. I added some words to the message to clarify.
(This was one of the reasons why I had to redo the probing function. On the old software, this would have needed some work to restore.)

[JuKu]

Question: Is there a way of getting 2 pads recognised using vision? like an 0805 resistor/cap.

Yes, there is. Functionality like that is the reason to update the vision system in the first place. I will do more instructions and examples like this later, but since you asked: Here is a sketch for an instruction page to look at resistors directly. I did this a moment ago; it is not added to the main instructions yet, as it is unfinished and I'd like to actually try it out myself. But the vision example is there. (But note my next post about known bugs.)
https://liteplacer.com/targeting-parts-directly/

[JuKu]

Issues know so far. Here, fixed means fixed in the source repo for those who can compile it by themselves as well as a for myself. Updated RC with the fixes soon.

-Clicking on the image gives wrong size move if show processing is on and zoom is used (fixed)
-Filter features by size does not take zoom into size calculation (fixed)
-Placing 0402 resistors with only 3 or 4 parts results to a pickup at wrong position (fixed, sort of. Now, the first hole needs to be good.)
-On old tape files, an error about bad data in capacity is sometimes reported(fixed)

[AnalysIR]

Tape data file sent via email along with nozzle data for another minor nozzle issue.

-Filter features by size does not take zoom into size calculation (fixed)

Is there an easy workaround for this...e.g. scaling the size to account for zoom?

Yes, there is. Functionality like that is the reason to update the vision system in the first place. I will do more instructions and examples like this later, but since you asked: Here is a sketch for an instruction page to look at resistors directly. I did this a moment ago; it is not added to the main instructions yet, as it is unfinished and I'd like to actually try it out myself. But the vision example is there. (But note my next post about known bugs.)
https://liteplacer.com/targeting-parts-directly/

WOW...this is great.
Comments:
- I tried this out only with 0805 resistors and results are greatly improved....very promising
- Initially, some failed and I figured out that nearby artifacts mess with the shape algorithm. So I made sure these artifacts were covered in black and there was a great improvement.
- Initially, I forgot to turn on all of the lighting, which improved things again
- So finally I manually got 17 resistors in a row recognized and for about 5 of them I just had to make 1 or 2 manual jogs to get them recognized.
- It seems to be very dependent on getting close to dead centre initially. Jogging fixes this, but components will be randomly distributed inside the pocket.
- I also had to use Erosion as the first function to get a good result and my threshold value was much higher than yours (~120ish). I also had to use different sizes for 0805.
- If the initial algorithm failed, it would be useful to allow a fallback to an alternative one...In this case if the component failed then try the pocket or even hole (like an auto recovery). You could just add in an extra Column to the tape table etc.

Possible improvements for this algorithm
- make sure the surrounding area is clean and if possible black for paper tape. No rough edges or corners nearby.
- Consider implementing a new 'Mask' function. This would draw a back/white - round/rectangle area outside of the target area, masking out all irrelevant stuff.(see image) If you could reduce the image size it would surely speed up the processing??? Reason: The shape algorithm which detects circles/rectangles gets confused if there are other artifacts in the vicinity (dirt/dust/loose components/edges). The mask would be applied probably first, but before any shape processing is started.
- Is there a way of improving the algorithm to avoid near misses that are out just one or 2 jogs.


I will try Capacitors later and some more resistors which are in an area with full black background

[JuKu]

> Is there an easy workaround for this...e.g. scaling the size to account for zoom?

For now: If you use (say) 4x zoom, increase the max and min size in “filter features by size” function as well- or wait for next RC.

Mask is already on the todo list, although it will not be exactly mask, but a resize. To speed up the processing, the image needs to have less pixels. I haven’t tried it yet to know how it would be best to show visually. Or if it makes sense as a separate function in the first place, the final result acceptable size and discard distance already gives the necessary info. Maybe it can be done easily enough automatically?

Also on the todo list: a way to save a camera image and use an image from a file as the video source. This would allow you to send examples of difficult situations (like when had to jog etc) and me to get much more feedback of real use cases and material to develop solutions.

[AnalysIR]

tnx...I will try it out

Mask is already on the todo list, although it will not be exactly mask, but a resize. To speed up the processing, the image needs to have less pixels. I haven’t tried it yet to know how it would be best to show visually. Or if it makes sense as a separate function in the first place, the final result acceptable size and discard distance already gives the necessary info. Maybe it can be done easily enough automatically?

Yes reducing size is better to reduce processing. Calling it mask or similar may be easier for users to follow (re-size may be confused with zoom or the existing size boxes)