[chbot] TCS3200 anyone?

[Helmut Walle]

On 09/04/15 19:14, Robin Gilks wrote:
[...]
> We're 100kg into harvesting (and hence washing) walnuts and the next issue
> that needs to be automated is grading. Nuts with dark stains on the shell
> are quite often rotten but in any case they are only accepted by the
> processor as a low grade. Trouble is, a few duff nuts in a bag of good
> ones reduces the grade and hence price of the whole bag.
>
> Does anyone have experience with the TCS3200 programmable colour to
> frequency converter device? Looks pretty simple to interface to :)
[...]

> I think the main issues will be mechanical ones - have the nuts travel
> slow enough to measure their colour and for the sorting gate to operate
> but quick enough to be able to process in a reasonable time.
[...]

Hi Robin,

interesting questions... do you have a collection of photos of good, bad 
and, erm, borderline (almost said "ugly") nuts? Or is it more a 
black-and-white situation with a very wide margin between good and bad, 
and no or very few borderline cases?

Besides the mechanical issues you are mentioning, I believe there may be 
some statistical issues as well. Obviously you want to maximise your 
yield, however not at the expense of degraded quality. So the sorting 
process needs to identify bad nuts with near perfect reliability, 
without throwing out too many good ones.

A camera with image processing software would cost more than the sensor 
you are mentioning, but it might be a good way of proving and tuning an 
algorithm before implementing it in a more cost-efficient way (if that's 
a requirement at all; if you only need one unit why optimise for 
component cost?). - Not sure whether this leads anywhere... 
alternatively, a collection of sample photos could also be used as 
training material for a neural network if the, or one, simple sensor is 
not enough.

Just some fuzzy ideas around this question - (I didn't want to mention 
fuzzy logic too explicitly!).

Kind regards,

Helmut.