LitePlacer.com forums

You can use vision to pick up loose components (always had the ability). The component from pads is more robust way to see parts. It is needed for up camera part correction (in the future); right now you can see ic’s in black tapes, for example. Filtering features for size makes it possible to throw away fake results, noise artifacts etc

- Issue with Job Data seems fixed in latest beta.

-tried out Vision on SOIC8(5x6mm) & QFN32 (5x5mm) (normal and with part upside down)

for QFN32 upside down, got :

p: -2.0, -11.0px; -0.042, -0.231mm; s: 238.7, 236.8px; 5.01, 4.97mm; A: -21.04
Result: X= -0.042, Y= -0.231, A= -21.04
Elapsed time 103ms

for QFN32 Normal got:

p: -39.0, -33.5px; -0.819, -0.704mm; s: 237.5, 237.0px; 4.99, 4.98mm; A: 30.34
Result: X= -0.819, Y= -0.704, A= 30.34
Elapsed time 61ms

When I had both outline&pads checked, I couldn't get a unique result. There were 2 results both identical. So in this situation maybe recognise it is the same rectangle??? (Should they toggle each other??)
Once I unchecked Outline, it worked fine.

For SOIC8 Normal I got :

Filtered for distance, results:
p: -63.0, -120.5px; -1.323, -2.531mm; s: 286.1, 244.8px; 6.01, 5.14mm; A: 31.86
Result: X= -1.323, Y= -2.531, A= 31.86
Elapsed time 115ms

For SOIC8 upside-down I got:

Components from pads:
p: 46.0, -36.0px; 0.966, -0.756mm; s: 298.6, 239.5px; 6.27, 5.03mm; A: -23.89
Filtered for size, results:
No items left.

Finally, I tried out a SOT23....didn't work out initially. Couldn't get a lock on the pads. The nominal package size didn't work. But once I used what the vision was measuring, it did lock on the centre point of the SOT23.

I haven't figured out the difference between Filter by size / Acceptable Size /Size min&max.

I also noticed that when there was another object at the edge of the camera view, the vision created a much larger rectangle (= fail).

At one point I turned on a stronger light, which created a shadow. This effectively made the part larger. However, a threshold adjustment fixed it.

Once you get the scaling for Measure zoom added, it should be much better.

Didn't try out any pickups...just vision tab.

How is the discard distance used. (from center or edge of object. Is it a radius or diameter?)

> When I had both outline&pads checked, I couldn't get a unique result.

You can search for different features (round, rectangles, components). All these go to a list of result candidates. Then the list is filtered by size (min, max) and distance (candidates too far away from the image center are not accepted). The result should be unique. For example, a rectangle likely registers as a component, too. Sometimes you can get the same component by pads and by outline. In real use, you pick what you think gives best results.

> I haven't figured out the difference between Filter by size / Acceptable Size /Size min&max.

Filter by size happens before building the list. It is useful to throw away noise or other unwanted results, especially before combining white pad images before building a component. You then use acceptable size to make sure you got what you wanted. For example, if the vision misses a pad of your SOT23, the resulting component is wrong size, you don't want to use that.

> How is the discard distance used. (from center or edge of object. Is it a radius or diameter?)

When vision is used, the machine goes first to approximate location, and then uses vision to get the exact location. In case the vision misses what you are looking for, you don't want a false match too far away. If the software doesn't get a match from the first frame, it tries again. You might have noticed cases where contrast is not so good, and the circles (or whatever) recognition results drawn flicker. Being able to filter the results makes the system more robust and allows us to use more relaxed algorithms: You can have false results as candidates, but the filtering helps to pick the result we want.

> Is it a radius or diameter?

Simpler. The object center X and center Y has to be less than discard X or Y from the image center.

Thanks Juha...that helps clarify some things.

Here is the latest exe: https://www.liteplacer.com/Downloads/be ... _02021.exe
I built also an install file: https://www.liteplacer.com/Downloads/be ... 1_2021.exe

This might be the release version. From now on, I don't intend to develop it much further right now, unless bugs or easily implement good ideas come up. Rather, I'll write the documentation to make this the official version.

Notable changes since last beta:
- option to draw results on top of unprocessed image works with zoom also
- bug fixes
- tape setup allows jogging with z down
- esc key works on table edit
- retry option on vision errors (with ability to jog, tune vision algorithms etc before retry)
- ability to jog machine before measurement (customer request: allows using a via as a fiducial on a very tight board, for example)

A quick bug and usability fix, the exe: https://www.liteplacer.com/Downloads/be ... 2-2021.exe

No install file this time.

Fixes:
"Reset Placed" button did not reset.
Moved camera location and nozzle setup back to algorithms page. (Although they logically belong to camera setup, you need them when setting up the algorithms.)

Welcome.
The new beta version works great so far.
I did not find any problems, problems,
But....
Is there a way to convert old tape data to a new program format. I can't edit them. (Invalid data at Capacity) when I press OK
It would take a while, too.

Best regards.

Krzysztof

My bad, I'll fix this.

It would be nice if you added when arranging
Loose part of the ability to select a recognition algorithm.

What do you think JuKu?

Pozdrawiam.

Krzysztof

Krzys wrote:It would be nice if you added when arranging
Loose part of the ability to select a recognition algorithm.

What do you think JuKu?

Pozdrawiam.

Krzysztof

Of course, will do.