Hi,
Does the system provide the capability to pick randomly placed components? My components are OD .050" x .010" thick.
Thanks,
Ramdomly placed component pick capability
Re: Ramdomly placed component pick capability
The current software has a loose component function, but that is one by one. It shouldn't be that difficult to enhance so, that it searches parts from a given area. There are some issues that at first thinking, I'm not sure what is the best behavior:
-Parts touching each other might be seen as one part and put the nozzle in between. How to avoid this? Set part size somehow? Set part shape?
-when run out of parts: ask user (especially if the above is solved, to spread out the remaining), error, ignore the rest and move on, ...?
Please share your thoughts!
-Parts touching each other might be seen as one part and put the nozzle in between. How to avoid this? Set part size somehow? Set part shape?
-when run out of parts: ask user (especially if the above is solved, to spread out the remaining), error, ignore the rest and move on, ...?
Please share your thoughts!
Re: Ramdomly placed component pick capability
While testing the vision functions I tried to write a function that can pick a C or R from a set of loose parts.
To start with the result: yes it can be done
So far I was successful picking up Cs (0603) and RGB Leds (WS2812/5050). Rs are still a little bit unreliable.
short sketch of steps:
Step A: no zoom - blurr (to get rid of noise) -threshold - detect contours - filter contour size - delivers collection of contours
Step B: Pick a contour from collection - get contour bounding rect (no rotation) - get rect center = contour center - move down cam to center
Step C: zoom in - mask rectangle a bit larger than max bounding rect can be - rotate nozzle to zero - detect rotated rect - move to rect center and rotate to part rotation - pickup part - rotate to zero - place.
But there ar several problems that can arise:
parts too close: can partially be solved using a size and distance filter but successfully to detect a rotated rect of correct size and angle no other part may be in the masked rect view.
parts are labeled: most R have their value labeled in white on black. The label creates a heap of contours, which can be filtered by min. contour size
parts have different colors: Cs have silver tips and a khaki colored body, Rs have silver tips and a black colored body plus while label.
To get good contours you have to place the parts on a white surface, but the silver tips and the body sometime create separated contours around the tips. So lighting and threshold is very delicate. I also tried colored cardboard as background. Although the cardboard looked uniformly tinted when you zoom in there ar fine fibers of brighter or darker color that can trigger a threshold or contour. I tried more blur but hat to be brought to a level that the part itself was blurred to much.
Maybe other forum members have also experimented in this field and could share their experience?
To start with the result: yes it can be done
So far I was successful picking up Cs (0603) and RGB Leds (WS2812/5050). Rs are still a little bit unreliable.
short sketch of steps:
Step A: no zoom - blurr (to get rid of noise) -threshold - detect contours - filter contour size - delivers collection of contours
Step B: Pick a contour from collection - get contour bounding rect (no rotation) - get rect center = contour center - move down cam to center
Step C: zoom in - mask rectangle a bit larger than max bounding rect can be - rotate nozzle to zero - detect rotated rect - move to rect center and rotate to part rotation - pickup part - rotate to zero - place.
But there ar several problems that can arise:
parts too close: can partially be solved using a size and distance filter but successfully to detect a rotated rect of correct size and angle no other part may be in the masked rect view.
parts are labeled: most R have their value labeled in white on black. The label creates a heap of contours, which can be filtered by min. contour size
parts have different colors: Cs have silver tips and a khaki colored body, Rs have silver tips and a black colored body plus while label.
To get good contours you have to place the parts on a white surface, but the silver tips and the body sometime create separated contours around the tips. So lighting and threshold is very delicate. I also tried colored cardboard as background. Although the cardboard looked uniformly tinted when you zoom in there ar fine fibers of brighter or darker color that can trigger a threshold or contour. I tried more blur but hat to be brought to a level that the part itself was blurred to much.
Maybe other forum members have also experimented in this field and could share their experience? best regards
Manfred
Manfred
Re: Ramdomly placed component pick capability
Hello,
I know I'm replying to a two-years-old post, but I've read it just now, and the first idea that come to my mind is of using a translucent screen - some milky-white plexiglass or sort of, lit from below (or else a LED panel, their price lately has dropped considerably) to be placed on a window of convenient size cut into the LitePlacer base and accommodating on its surface the loose parts, that this way are becoming perfectly black shapes on white, as long as they are not already translucent by themselves...
I know I'm replying to a two-years-old post, but I've read it just now, and the first idea that come to my mind is of using a translucent screen - some milky-white plexiglass or sort of, lit from below (or else a LED panel, their price lately has dropped considerably) to be placed on a window of convenient size cut into the LitePlacer base and accommodating on its surface the loose parts, that this way are becoming perfectly black shapes on white, as long as they are not already translucent by themselves...