Last modified: 2026-03-27 16:21:19
< 2026-03-25 2026-03-27 >Part of me is thinking I should give up on trying to make a compliant mechanism piano action and just make the bloody keyboard glockenspiel with an ordinary action.
But part of me finds the puzzle irresistible, so I think I am going to persist.
I have this:
In CAD it is showing that the key reaches full 10mm dip at 50g of force, and this moves the hammer about 30mm. Will see how it works in real life.
I don't like that some of the flexures are so thin, seems flimsy. But how else do you stop them from being stiff?
I also would prefer more like 10x lever ratio instead of only 3x.
https://www.youtube.com/watch?v=8Ia-3dpteys
Meh, forget it, it is too springy and vibratey and wobbly. I'm going to forget the whole compliant mechanism boondoggle and just design an ordinary mechanism.
What kind of lever ratio would a typical piano action have?
On this one:
It actually looks like the ratio is only about 2x? The distance between the hammer and the string is only 2x the key travel.
I found an alternative diagram where it is 2.5x. So is it more that it's highly non-linear? Rather than being an overall high ratio. If the first part of the key travel gives very little hammer movement, and then the last part gives a lot, you effectively have a higher lever ratio to accelerate the hammer at the end?
According to ChatGPT a typical ratio is more like 5x or 6x, defined as "hammer travel to let-off / key travel to let-off". What is let-off?
I think my error is that I measured it as though the felt key stop didn't exist. Looking again:
So that's almost 4x now instead of 2x. And this has 6.4x:
I guess aim for 6x overall.
I want to finally make the "Image Measurer" idea from Software Ideas. Now that AI coding is really good this should be pretty easy.
I've done this by writing a SPEC.md file, getting Codex to ask for clarification on things it's not sure about, then getting Codex to update the spec, and finally getting Codex to implement it.
It did a working version in one shot, great success.
And really quite satisfactory with a few refinements:
https://incoherency.co.uk/image-measurer/
< 2026-03-25 2026-03-27 >