Chris Radek wrote: > Yes I think so. That's why Stuart S has his pid maxoutput set to 8, > which is the same as the machine's maxvel. That lets pid units be > real world units inches/sec. > > Yes, this way of setting it up is starting to make sense (after only 10 years of making everything conform to 10 V as the nominal engineering unit). > It turns out he could have used scales of 1 all the way through and > set his amp gains so 1 volt = 1 inch/sec and his numbers would have > been really simple. But in general that's not going to work well > unless you happen to have an 8 or 9 inch/sec machine. > > Mine is about that speed too, and I *did* adjust the amps so the dacs > are 1v = 1ips. I have an OUTPUT_SCALE of 10 (because of the mesa full > scale being +-1 instead of +-10 at the hal pin). I have pid maxoutput > of 10. My tunings are P=300 D=2 FF1=1 and the fine adjustment of tach > gain and dac offset were just done with the amp knobs. I have > excellent response and following with acceleration at 35. > Nice! I think my servo amps (and motors) are way too wimpy, but I'm envious of that acceleration. > >> The limit of the PID output is set by parameter pid.n.maxoutput which is >> set (in my >> sample files) by ini parameter PID_MAXVEL, not a really accurate >> description. >> I have it set to 12.0, which also looks wrong! >> > > Yeah I wondered in his case whether it should be set to some number > slightly higher than the machine maxvel for some headroom. Or maybe > to whatever number allows up to 10 at the dac. Maybe a separate > setting is better for that reason. > > Yes, if the PID output ever maxes out, it does really horrible things to loop stability. You get integral windup as well as just going open-loop for a moment. The recovery is very abrupt and discontinuous. > It's tempting to leave it unset (unlimited) to give people one less > number to worry about, but you can get integrator windup forever then. > Exactly. > > It would be nice to have good documentation for how a velocity mode > setup should work. I have gone through it with several people - I > don't think it's written anywhere. People think they are overly > complicated, but dang they give unbeatable results once they're set > up. If you start with a tuned working velocity loop like Stuart S > has on his running machine, you can't go wrong. > > Yes, I've been trying to tell Igor this for some time. I guess he's never SEEN a properly-tuned velocity servo. Sometimes in my shop I'd demo this by shutting off the EMC positioning loop and trying to deflect the servo. It was amazing how well the velocity loop would resist outside forces. >> You must know this as you are using "the source IS the documentation". >> > > Yep, you're not kidding. > Well, I really do intend to add some documentation to the PPMC driver, but I'm up to my eyeballs now in nuclear gear manufacturing, and that pays a lot better than the EMC stuff.
I could do a new Wiki tuning page that specifically goes over velocity servos, and use my Bridgeport as the test case. It is only a bit atypical. I had no IDEA what bandwidth was appropriate, and was starting with some class-D audio amp designs, so ended up putting WAY too much bandwidth in them, they start to roll off at 1 KHz. Testing with a signal generator, they gave good response out to 10 KHz, using the motors as speakers. Also, the whole axis drive system on my Bridgeport conversion is too weak, it is adapted from an Excellon PC board drilling machine, which moved a small aluminum table on air bearing sliders. So, I keep the acceleration at 3 inch/sec^2 to keep from snapping timing belts. Still, I think it could produce representative scope traces of the likely situations and what to do to fix each symptom. Most of what I have on my tuning page now is reasonably apropos, although not an analytical approach. Voltage amps roughly simulate a sloppy velocity servo. Jon ------------------------------------------------------------------------------ Start uncovering the many advantages of virtual appliances and start using them to simplify application deployment and accelerate your shift to cloud computing. http://p.sf.net/sfu/novell-sfdev2dev _______________________________________________ Emc-users mailing list Emc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-users
