Maximum Servo Thread Frequency?

but, wait, I'm not sure I believe your calculations. 100 RPM is 1.6667 revs/second. Multiply that by 160,000 (assuming your
40K PPR is the same as an optical encoder with 40K lines) and you get 266667 counts/second. Still, if you want to
run at 1000 or 2000 RPM someday, the USC would not be able to count that.

I've got a 2500 line per rev. codewheel on the spindle giving 10,000 quadrature edges per rev., each quadrature edge equating to ~0.15 µm z axis movement. This is just over 3x the required 0.5 µm target machining accuracy, so ideally I need more than 10,000 pulses per spindle rev to get more than the recommended 10x sampling rate for servo systems?

I've checked the RGH24H specification and it's a 50 nm resolution readhead. Over a short distance the readhead error is at best +/-0.2 µm, which is roughly the same as the target machining accuracy, so it looks like it could do with a little more resolution?

So the encoder frequencies appearing at the USC (450 rpm much lower than recommended for Sandvik threading insert with brass!), the spindle encoder bandwidth is (450/60)*10,000 = 75 kHz, and the RGH24 approximately 11.91 mm/ 50 nm =~ 238 kHz. So both the spindle and z axis encoder frequencies are below the USC maximum 500 kHz

I think in the end, it's probably going to be the runout in the Z-axis bearings which dominate - but on the plus side, the Z-axis only needs to move 20 mm for threading, so

llrjt100 wrote:

but, wait, I'm not sure I believe your calculations. 100 RPM is 1.6667 revs/second. Multiply that by 160,000 (assuming your
40K PPR is the same as an optical encoder with 40K lines) and you get 266667 counts/second. Still, if you want to
run at 1000 or 2000 RPM someday, the USC would not be able to count that.

I've got a 2500 line per rev. codewheel on the spindle giving 10,000 quadrature edges per rev., each quadrature edge equating to ~0.15 µm z axis movement. This is just over 3x the required 0.5 µm target machining accuracy, so ideally I need more than 10,000 pulses per spindle rev to get more than the recommended 10x sampling rate for servo systems?

I've checked the RGH24H specification and it's a 50 nm resolution readhead. Over a short distance the readhead error is at best +/-0.2 µm, which is roughly the same as the target machining accuracy, so it looks like it could do with a little more resolution?

So the encoder frequencies appearing at the USC (450 rpm much lower than recommended for Sandvik threading insert with brass!), the spindle encoder bandwidth is (450/60)*10,000 = 75 kHz, and the RGH24 approximately 11.91 mm/ 50 nm =~ 238 kHz. So both the spindle and z axis encoder frequencies are below the USC maximum 500 kHz

I think in the end, it's probably going to be the runout in the Z-axis bearings which dominate - but on the plus side, the Z-axis only needs to move 20 mm for threading, so

OK, all these frequencies seem OK. But, expecting an overall accuracy of 5 um seems iffy. Unless the machine is massively
rigid and the driven by very smooth servo drives, I have my doubts it will really achieve the required accuracy. Do you even
have instruments to PROVE you are meeting this accuracy?

Jon

I can measure the resulting worm wheel gear error to less than 0.1 arcseconds - the problem is measuring/controlling the error when threading the worm and hobbing the worm wheel, and achieving a sufficiently rigid setup for this.

I've been Googling grinders, and it looks like all the top end kit use hydrostatic bearings, which I suspect are going to be too expensive and bulky for my application. I can use relatively simple dead centres to achieve concentricity of the worm blank/hob - the main issue is the z-axis, and the errors in the linear bearings I guess. The big question is how good will linear cross-roller/ball bearings be over 50 mm length on the z-axis?

Are you building from scratch?

I can build from scratch, but I'm nervous about correctly identifying the things that matter/don't matter in getting a short highly accurate thread, and also spending more time building the machine than using it!

Thanks for the reference to the Bryant Symons Lathe - I guess the quality of the machined worm helix is determined by the straightness of the z axis ways and the transmission errors in the change wheels plus the helix error in the z axis leadscrew.

I've look at air bearings and linear motors, but they are expensive - the nanometer precision light load machining centres use granite beds, with aerostatic bearings and linear motors, and temperature compensation. The tolerances I'm working too aren't as challenging, so perhaps over a short distance of say 50 mm, a z-axis cross-roller linear bearing should be pretty accurate, as you suggest. I've spoken to some people who are achieving pretty good results on a standard Hardinge CNC collet lathe, with very light spring passes. Because I'm only producing a small worm, I'm thinking I can put my money into higher precision parts to create a much smaller, more accurate machine.

I'm thinking of using a granite surface plate and attaching the hardware to it - one of the problems is how to line up the z-axis linear bearings with a fixed headstock and moving tailstock - perhaps the z-axis bearings could run the whole length (probably < 500 mm), with the head stock clamped permanently to the linear rails, and the tailstock sliding? Both the headstock and tailstock are simple dead centre supports - I just need a mirror image pair which can be accurately aligned, with the z/x carriage moving in between (50 mm Z, 150 mm X). The worm blank will be driven by a lathe dog - I'm thinking of mounting a timing belt driven pulley on a bearing on one of the dead centre holders...

That sounds like very traditional lathe practice. lots of lathes have a headstock that clamps to the same straight bed as the carriage slides on.

I wonder if this can be achieved with linear profile rails? - in other words, do clamping elements exist, in addition to the ball/cross-roller carriages?

I'll have a poke around cnczone to see what I can find.

Normally LinuxCNC servo systems are run with velocity mode drives.
This means the drive itself runs the current (torque) and velocity PI loops
and LinuxCNC runs the position loop (with a velocity command output)

On most systems the nested control loops run at slower and slower frequencies
as you get closer to the controller.

For example on Fanuc's latest drives the current loop runs at 32 KHz, the velocity loop
at 16 KHz and the position loop (the part LinuxCNC typically does) at 4 KHz.

This means the drive itself runs the high speed loops and effectively does
interpolation between position way-points. This interpolation also means servo accuracy is not
limited to velocity/servoperiod but something much less, that is determined (in part) by how
closely the drive follows the velocity command between position way-points.

For this reason and the limited bandwidth of the mechanical drive system
(maybe 100 Hz at the most), there is little to be gained in increasing LinuxCNCs
position loop sample rate above a few KHz.

Would you be able to suggest a LinuxCNC compatible drive which will give me the high speed current and velocity loops as per the Fanuc control?

These people have some pretty high performance MOSFET drives with 28 KHz PWM/current loop
(most larger IGBT based drives are much slower)

www.teknic.com/products/eclipse-servo-mo...pecs-eclipse-drives/