Performance Measurement, again
Before I mounted the chuck on the RT backplate I had opened up the chuck back cover to look inside. I was reasonable happy with what I saw. Yes, the basic guts are either very simple or slightly crude, depending on your point of view, but the parts looked solid, the interior seemed clean of any swarf and there was plenty of grease in the right places. So that seemed OK - just heavy.
OK, so I mounted the chuck up on the RT, and try chucking a few bits of ground rod to see how much they wobbled during rotation. The first few did wobble - and rather variable amounts too (this is before the measurements reported earlier). That was a bit odd I thought. So I ran the jaws in all the way and out all the way several times, against as much load as I could manage with my hands, and then tried again. Aha - much better: less TIR and more consistent. From which I infer that the teeth on the jaws or the faces on the scroll (or both, more likely) just needed a little bit of bedding in, or maybe they were not totally free of very fine machining dust? It is cast iron after all. Only then did I do the TIR testing reported in the last chapter.
Now that I had the outputs mechanics done it was time to check the input motor parts. I carefully cleaned the teeth on the two pulleys and the face of the toothed belt and assembled the drive. Then I carefully tighted the belt using the screw mentioned earlier. A loose belt is an invitation to all sorts of problems, but so is a belt under too much tension. I aimed for just a few millimetres of deflection under 'medium finger load'. Yeah, real technical! But the teeth on the bought GT2 belt fitted the teeth on the DIY pulleys very nicely - once the belt was tensioned. If you don't tension the belt it does not seem to fit 'into' the teeth too well.
A few quick spin commands from the Mach keyboard, and it was clear that <i>it runs</i>! Good. (Sigh of relief.) So now to test it a bit more thoroughly.
I reused the optical sensor from before for further testing. First I attached a thin bent shim to the backplate with a small self tapper - tightly. That made the blocking vane. (If you must know, it was the sliding cover off a dead 3.5" floppy disk. Nice hard alloy.) Then I mounted the sensor block so the vane went through it without touching. Yes, the arrangement in the photo does not look 'industrial-strength', but it is not meant to be a permanent fixture. It is there just for these tests. By monitoring the output of the sensor I could measure tiny movements of the RT - and I do mean tiny. So now several trials could be run.
Incidentally, I will mention here that the data logging had to be done by hand. Neither Mach nor a Smooth Stepper have the facility of reading in analog voltages - although I see that the UC-300 interface and UCCNC SW do. Interesting, and noted. I thought of using a USB data logger in parallel with Mach, but running two 'real-time' systems together on one Windows-based machine ... I thought maybe 'not today'. Actually, if you don't run the data logging very fast and you have a reasonable look-ahead buffer in mach3, it should work. Most of the traffic will be by DMA after all.
The first trial was to calibrate the sensor again - for this arrangement. The previous calibration was no longer valid as I had changed the radius to the sensor beam. This let me convert the mV read by the multimeter into degrees of rotation.
Previously I had ran a number of loops back and forth, which gave the hysteresis curve shown before. I had also measured the slope of the response curve for both sides of the hysteresis loop. They were close enough together, given that they were done with the direct drive at 283.333 steps per degree. But those results were fairly coarse. Now I could do better, since the toothed belt drive gave me 1,000 steps/degree. Just to remind you, that means a single step gives 0.06 arc-minutes of rotation, or 3.6 arc-seconds. The aperture of the sensor used is 0.1 mm or 100 microns, but you can't use anywhere near that range of movement.
Here we have a hysteresis loop with readings taken every 0.001 degrees - every step pulse. You can see a slight rounding over at the tips of the loop - that may be the sensor non-linearity. But the gap between the two parts of the curve - that's the HD hysteresis. Overall the sensor sensitivity is 0.000347 degrees/mV (or 0.347 degrees/V). The hysteresis is about 0.008 degrees.
You can also see a few bumps. Some of those are due to rounding-off the DVM readings, while there may also be some slow time-constants involved in the sensor: it was working at a faily high impedance, and I was maybe a bit impatient sometimes.
This is a strange-looking graph, but it does make sense once explained (I hope). It shows what happens when the RT starts at 0 degrees (set to mid-range on the sensor, at about 0.033 degrees), rotates to +45 degrees and then back to 0 degrees, then to -45 degrees and back to zero, a number of times. The zero positions are shown; the 45 degree positions are way off-scale of course. Basically, it shows that the RT always returns to the same zero position, after allowing for hysteresis. Going in one direction you get the lower 'zero point' at 0.033 degrees, while coming back in the other direction gives you the other 'zero point' at 0.045 degrees. So the hysteresis is about 0.012 degrees for a 45 degree swing. But the reproducibility is far, far better.
There are a number of cycles plotted here, and they mostly overlay nicely, but you can see what looks like a slight 'widening' of the line at the lower right. That widening is actually a number of plotted lines just faintly offset from each other in the vertical direction. I believe that is really a slight thermal drift in the sensor, as it is below the resolution of the stepper motor.
I have seen a video of a large and very expensive RT doing just the same, against a dial indicator. When the RT came back to 'zero', the DI went back to zero too. I was impresssed at the time, but it seems an MYOG effort with a Harmonic Drive and a zero-backlash toothed belt can match some of the performance of very expensive commercial units.
Another interesting graph which needs explaining. This time the RT is cycling under significant load (or torque). This is meant to give some guide as to what the RT might do when machining forces are placed on it: will it creep? The torque was created by a lump of steel hanging off an arm, giving a static torque of about 0.46 Newton.metres (N.m). What that means, in more practical engineering terms, is a 3 kg lump of steel hanging off at 150 mm from the axis of rotation. This is a fair lump of steel at a fair distance, and is significantly more torque than a 6 mm cutter would give on a 100 mm diameter bit of aluminium (I think).
The RT was oscillated back and forth a few steps. A small range was used to stay within the scope of the sensor. The left hand bunch of points show the RT being driven from 0.06 degrees to 0.09 degrees and back: the repetition of end points is very good. For the middle third of the data the positions were moved up by 0.02 degrees. The repetition between points stays good. Then for the final one third the positions were brought back down to what they were at the start. There is very little difference in sensor reading between the start and the end, and what there is is probably thermal drift in the sensor anyhow. The RT is holding its position under load: it is not creeping.
Zero Backlash
Can I measure any backlash this way? The smooth continuous nature of the hysteresis curves under no-load conditions suggest there is no classical backlash: there are no flat spots in the response curves. That means the hysteresis which can be seen may be due to slight movements in the harmonic drive spline engagement, or maybe flex in the spline. It's hard to say, and you would need a solid read of the detailed technical explanation of how the HD really works. We are talking about some very fine elastic distortions here. However, I doubt I will see much effect in practice. After all, 0.001 degrees represents 0.87 microns at the periphery of a 100 mm diameter object. Did I go overboard with this? Not a bit: it's nice to have the resolution.
Do I need a brake on the unit? At this stage, with these results, I do not see the need to add a brake. It would have to be quite a powerful brake, and you can't use a brake on a moving rotary table anyhow. You can use a brake on an indexer, but that was not what I want.
Well, there it is. Questions are welcome.
Roger