There are quantization problems with other than simple CAD models, however.
Curves and low-tangent cuts are particularly subject to quantization error
effects -- think of "jaggies" in graphics rendering.
I agree with you, though, that hexapods have unexpected quantization
errors. While a cartesian machine can cut an axis parallel surface with no
quantization induced variations (although the exact position of the cut
line or plane may be off due to quantization), a hexapod will intrinsically
have a surface roughness introduced by the quantization.
In my new hexapod, I have an actuator quantization error (theoretical) of
0.000025 inches. Each actuator is servo driven with 8192 encoder events
per revolution, and a 0.2" per revolution displacement. Depending on where
the platform is operating, one paper analyzing hexapod accuracy finds that
the expected platform error is about 1/2 of the actuator error. If so,
then the peak error should be on the order of 13 microinches. Of course,
that's with everything perfect, which strangely enough it never is.
What are good ways to measure surface smoothness? It would be interesting
sometime to attempt to cut a plane, and see what the error actually is, and
somehow estimate what portion is quantization related, vibration related,
calibration related, and simple actuator error related.
-- Carl
toggle quoted message
Show quoted text
At 12:13 PM 8/31/2006, Mike Pogue wrote:
Dennis Schmitz wrote:
Not really. Since CAD models are built in a cartesian coordinate system,
somewhere along the way you need to translate into a motor space. Whenever
you translate co-ordinate systems, you get quantization error at the
encoder
(or step) resolution.
In the cartesian axis case, I don't think quantization error is a given,
because the step resolution can match the coordinate system resolution.
In my case, for example, steppers are 200 steps/rev, and leadscrews
are 10TPI. That gives a quantization of 0.0005".
So, as long as I specify coordinates in 0.001" increments (which is what
I normally do), there's an exact representation in motor coordinates
(and no quantization error).
Mike