This is somewhat of a continuation of x and y off by 1%. There, I realized I was cutting parallelograms–say I programmed two adjacent 100mm squares. When I cut them and hold them on top of each other, they line up well, but when I rotate 1 90°, the lack of orthagonality becomes apparent.
I happened upon these realities when I began attempting to cut a peg board. I made holes in a cross shape so the board could be used both horizontally or vertically, but the widths of the two cuts across the board were generally 1/4" or so off from each other.
Pegboard pattern I attempted to replicate, but my hooks either couldn’t fit in the hole or wouldn’t slide down over the groove…
As I was reading through maslow.cpp with my coffee this morning, I believe I understood that, when tensioning, maslow sets TL and TR belts to their anticipated lengths, and then tightens BR and BL.
The issue I’m seeing is that maslow moves substantially when BR and BL are tensioned. I mounted a measure on my board and ran tension belts. In a casual measurement, maslow moves roughly 7/32" down from the position set by the top belts when the bottom belts are tensioned. Here’s a video. I believe this displacement is due predominantly to the TPU/steel belts stretching. Because firmware (if I’m understanding it cortectly) calculates maslow’s position from the place it expects maslow to be according to it’s top belts, and not where it ends up after being fully tensioned and subsequently displaced, it seems error is being calibrated in to maslows position, regardless of frame size.
I know that frame/bolt rigidity can be a sensitive topic, so I tried to see if I could measure belt stretch more directly so it could be quantified. I was anticipating something in the range of a micron of stretch per mm of belt.
I attached two sets of two plates about 180mm apart on the top right belt, near the anchor point. I measured the space between the two plates. Unmounted/laying on my desk, the plates were 178.87mm apart. When mounted but not tensioned, the spacing measured 178.87mm. When tensioned, I measured 179.08mm.
Because the difference was relatively small, and the blocks I had installed weren’t rigid by the nature of the measurement being taken, I thought it might be worthwhile to try and measure a longer sample. Of course the accuracy of my instruments at larger scales dropped substantially from my calipers. I chose my laser measure which has a stated +/- 1/16" error.
I unhanged, retracted, extended, hanged. When I press tension belts, maslow first moves up to set top belt lengths, then tensions bottom belts. Working quickly, and after several attempts, I was able to measure the top belt lengths after being set both before and after tensioning the bottom belts, arriving at these results:
TL
Untensioned: 5’ 7 & 4/32”
Tensioned: 5’ 7 & 11/32”
TR
Untensioned: 5’ 7 & 9/32”
Tensioned: 5’ 7 & 13/32”
These numbers do show elongation in keeping with the overall movement of maslow that I filmed–roughly 7/32" down, but are also close to the margin of error of the instrument. It would be better to make a larger sample of them, but I think time might be better spent beginning to correlate varying amounts of stretch to tension current measurements.
I suspect factoring in belt stretch may be a boon to dimensional accuracy, but getting it ‘right’ may be tough. Factors include stretch due to gravity, varying stretch at different tension currents, and the reality that the ‘distance’ being measured by maslow’s belts as they move through the encoders would therefore be changing under different tensions as well. The stress-strain curve would be linear in the elastic region, so as long as that’s not being exceeded, it should be calculable…
