Originally Posted by Bernd_
I currently calculate the chain dropping speed every second this should account for variations in dropping speed. Therefor, this is not a major issue. But to account for things like wind noise
people talking and the different characteristics of boat types and hull
materials is a problem. My tests so far had an accuracy of roughly 10%.
I also experimented with different methods like spectral decomposition or graphical analysis of the signal. The Spectral decomposition seemed promising but I am not happy with the reliability
. Therefor, I am considering to use artefical neural networks but in order to do this I need a large set of training data which is the reason I need more recordings form a diversity of boats.
Interesting. The frequency will change with speed, so you need to be clever about spectral analysis. I would think that you could do an FFT and then look for the peak (since wind, voices, etc will be a broad spectrum but the chain rattle is more consistent). The peak frequency would give the speed the chain is going out.
You know roughly what frequencies to look at from a practical standpoint, so you can limit the bandwidth to speed up calculation.
10% accuracy is actually pretty good. Anchoring
isn't so precise! You should think about the user interface. The user should, of course, input the chain size up front, but also the height of the bow roller above the water
. Then, right before heading up to the bow to drop, the user should enter the water depth
and the expected maximum tidal increase (if you were really clever you could have the app know the tide information, but that might be a bit much).
The app could then calculate the chain out and give a real-time reading of scope
. You could colour code the scope
to be red below 3, turn yellow, then green at 5.
Great idea, I think. If you can get it to work in real conditions!