Deviation problem with configured axis

kostas wrote:
I don't think that G61 will override any velocity and acceleration values

I stand corrected if that is the case, I assumed that if the sole purpose of G61 is to reach the end of a path, it must at some stage control trajectory if that would result in an over-run.

There is nothing obvious I can see in your .ini file.
Accel and velocity look normal, base thread and servo thread are the bog standard defaults from stepconf.

Your scale figure intrigues me, SCALE = 500.187570339 is very precise.
Are you using imperial ball screws on a metric machine or some unusual gearing?

ArcEye wrote:

Your scale figure intrigues me, SCALE = 500.187570339 is very precise.
Are you using imperial ball screws on a metric machine or some unusual gearing?

Hmm, I guess the number is just an internal calculated value. I configured the machine with stepconf wizard. The values I entered are
- 2000 steps per revolution
- 10 microsteps
- 39,985mm leadscrew pitch

Leadscrew pitch should be 4mm (metric trapezoid spindle with high precision nuts made from ZX-100K) but it seems to be a bit less than 4mm. 39.985mm gave the best results so far, scaling different ways from 50 to 150mm.

r0sewhite wrote:

Leadscrew pitch should be 4mm (metric trapezoid spindle with high precision nuts made from ZX-100K) but it seems to be a bit less than 4mm. 39.985mm gave the best results so far, scaling different ways from 50 to 150mm.

Are you sure this isn't the nut's backlash or any calipers error? We're talking about 0.0015mm error, which is pretty small. I would just trust the screw's pitch and go with the 4mm value.

kostas wrote:

Are you sure this isn't the nut's backlash or any calipers error? We're talking about 0.0015mm error, which is pretty small. I would just trust the screw's pitch and go with the 4mm value.

I always eliminated backslash by running the machine a few millimeters against the caliper before starting a scale run. I don't think that a machine can be calibrated with backslash. If the backslash is, let's say 0.2mm and you run 100mm, you'll get 100-0.2=99.8mm. If you scale 50mm, you would end up with 49.8mm. There is no consistent ratio between these two distances, so it's impossible to find a pitch value that matches more than one fix distance.

Of course my caliper could be wrong, but to be honest: I trust my caliper more than the screw's pitch.

Finally I've got my interupt problem solved. I found out that rtai_smi was not loaded. I started "/etc/init.d/realtime start" and got a message that the path for the SMI module was wrong.

I checked rtapi.conf character for character multiple times but couldn't find any error. Then I had the idea to simply copy the line below (which is loading rtai_hal) and change hal to smi. And now it works...

Don't ask me what was wrong. However, no SMI interupts and no warnings anymore. After half an hour max jitter ist still below 30.000.

So, did sorting out your latency spike from SMI resolve the undershoot on your Y axis?

It would make sense if it was latency related, because the error increased with feed rate.

Sorry for not replying the past days. When I got the smi module running, it was almost 2:30 in the morning and I was too tired tu run any tests. The next days I had too much work to care for the machine.

I just checked it and the problem remains: A test run under stepconf wizard gives perfect results but if I run the same distance in EMC2, the faster I run, the shorter is the driven distance. With F200 I get only 99.48mm instead of 100 and with F400 I don't even get 99mm. Only with G61 the distance is correct.

I contacted the designer of my machine a few days ago and he told me that my caliper must be wrong: His trapezoid spindles do no only meet the requirements of DIN 103 (max. deviation on 300mm: +/- 0.15mm), actually they don't even have a deviation higher than 0.05mm on 300mm. This guy is well known for high precision cnc machines so his statement is reliable.

I think we can exclude missing steps due to latency problems because max jitter is not more than 30.000 now (entered 50.000 in stepconf wizard to be on the save side). Also the machine always meets 0.00mm again when running back to 0 position and runs perfect in axis test of stepconf wizard.

The screw is very good: In stepconf wizard I can scale any distance between 5 and 200mm (my caliper doesn't make more) anywhere on the axis and the results never differ more than 0.01mm from what I entered. The thread is even more accurate on the overall length than I expected.

There is no backslash at all in the nuts. Meanwhile I tried to scale a potential backslash but couldn't find any. This is how I checked it.
- 1st test run 10mm with running a few millimeters towards the caliper before the test run
- 2nd test run 10mm with running a few millimeter away from the caliper before the test run

So what other reason could be for EMC2 to lose distance when running faster?

I think there is no problem at all. When in G64 (the default), toolpath is never exactly what the g-code program dictates.