2 years back I picked up a 1992 Fadal 4020; 28-inch Z axis, 10k spindle, DC servos. I had a cart full of Dynomotion boards but started making parts with a drip feed to the old CNC88. Present day: everything was going fine until my X-axis resolver started on the fritz. The cabling looked rough as well. So, I ordered about $1k in parts from Cali to make it right. As I was waiting for parts to show up...the wait got the best of me. I didn't want my Fadal "Fawne" to be trapped in a wait-for-old-yet-expensive-parts-ecosystem. To my surprise; that shopping cart still had the Kflop, Kanalog, and Konnect boards waiting for me! Click.
Not the first or the last to do a Fadal retrofit. CNCzone users Drassk and Mmurray70 have done most (maybe all?) of the legwork getting a Dynomotion setup running smoothly. That can be read through here:
https://www.cnczone.com/forums/dynomoti ... rofit.html
I just wanted to document my retrofit here for the benefit of the community.
Fawne Fadal, Unchained. (4020 Fadal Retrofit)
Moderators: TomKerekes, dynomotion
Re: Fawne Fadal, Unchained. (4020 Fadal Retrofit)
The plan is to keep the DC servos and use Omron E6B2-CWZ1X 2000P/R encoders. For the drives, I was planning on using some Copley Controls AEM-180-18(Good deal on Fleabay). They would need a custom backplane made but offer some fun features like autotune. I do have concerns about amperage with these drives though. In the hunt for drives, I was able to snag 3 AMC drives that are drop-in replacements for the Glentech GA4568EA drives. The plan has changed to modifying my original drives to cut out the velocity loop as described in the CNC zone post. I also picked up a spare DC servo which I think will very quickly become a 4th axis. There is a part I machine right now that is just begging for 3+1 capabilities.
Currently parsing through all the C-code and Dynomotion wiki until parts start to arrive. Will follow up with some pictures.
Currently parsing through all the C-code and Dynomotion wiki until parts start to arrive. Will follow up with some pictures.
Re: Fawne Fadal, Unchained. (4020 Fadal Retrofit)
Well, plans for the servo drives have changed drastically. Through some more researching, I have finally settled on some drives. New ones. Junus JSP-180-30 from Copley controls. 100% digital drive, 15 amps continuous, up to 180V DC supply voltage. They seem to offer some unique oppurtunities but I won't get into that until I get a chance to test them. What really excites me is that they are modern and off-the-self. Don't have to rely on eBay or others for parts.
With the conversion, progress has been made. The old controller and OP have been removed and sold. Kflop, Kanalogue and Konnect boards are mounted inside the cabinet along with the ribbon cable breakout. Tach/resolver cables pulled and encoder cables have been run. All 3 servos converted over to the encoders mentioned above.
With the conversion, progress has been made. The old controller and OP have been removed and sold. Kflop, Kanalogue and Konnect boards are mounted inside the cabinet along with the ribbon cable breakout. Tach/resolver cables pulled and encoder cables have been run. All 3 servos converted over to the encoders mentioned above.
Re: Fawne Fadal, Unchained. (4020 Fadal Retrofit)
The X-axis is setup! Well, mechanically. With the motor wired just with power in a vice on the floor I attempted to setup the motor with the drive. Took a long time to get the software to connect. (Recieved an email back the next day from Copley support saying support was 'forgotten' for my series drive. By this time I had found a legacy version that was working well with a legacy version of the usb-rs232-rj11-rxtx setup I was rocking. This non-support issue has been fixed in their next version I am told.)
Once connected, the setup went real smooth. Put in motor spec and hit autotune. No dials or pots to tweak! In 5 minutes the values for the torque loop in the drive were spot on. It also calculated the proper signaling bandwidth to use. A little over my head but I thought it was fancy. I attempted a step response move.
Moves as one would expect with only a very small P value. Started fartin' around with values till about an hour later realized I didn't know what I didn't know about servo tuning. Spent my free time this weekend rereading about 8 pages of this thread:
https://www.cnczone.com/forums/dynomoti ... ted-6.html
Looking forward to getting some time this week to tune in the X axis and get the other 2 wired up and moving as well. I may rerun the torque loop autotune now that the servo has been bolted back up to the machine.
Once connected, the setup went real smooth. Put in motor spec and hit autotune. No dials or pots to tweak! In 5 minutes the values for the torque loop in the drive were spot on. It also calculated the proper signaling bandwidth to use. A little over my head but I thought it was fancy. I attempted a step response move.
https://www.cnczone.com/forums/dynomoti ... ted-6.html
Looking forward to getting some time this week to tune in the X axis and get the other 2 wired up and moving as well. I may rerun the torque loop autotune now that the servo has been bolted back up to the machine.
Re: Fawne Fadal, Unchained. (4020 Fadal Retrofit)
-
TomKerekes
- Posts: 2546
- Joined: Mon Dec 04, 2017 1:49 am
Re: Fawne Fadal, Unchained. (4020 Fadal Retrofit)
Hi Batty,
Very good. But several things:
#1 the Acceleration is set extremely high relative to that level of Jerk. Jerk is normally set around 10X the Acceleration value in order to to apply the full acceleration in around 0.1 seconds. With those levels of A and J it would take 200 seconds to apply full acceleration. And by this time the max V would be far exceeded. So the acceleration is entirely jerk limited (triangular rather than trapezoidal). There is nothing necessarily wrong with entirely Jerk limited motion but mainly the A value can be misleading.
We have an included tool called TripsWPF.exe that will plot acceleration with any parameters. Here is an example with your settings:
Notice the A ramps to a little over 1e6 before needing to ramp back down. You might try reducing A down to something like 1e6 and then increase J to see if performance improves.
#2 You didn't post the IIR Filters Screen but I'm guessing the filters are all cleared. Notice the Output (green) has something like 150 count spikes in it. This is because with D gain of 150 each encoder count change causes a 150 count impulse. Often the Amplifier can't react to this and effectively smooths out the spikes but adding a 2nd order Low Pass filter of around 500Hz Q=1.4 should reduce the spikes. (This is normally placed in Filter #2 so it also applies to feed forward).
#3 The largest error occurs as a lag during the acceleration and as an overshoot during deceleration. Acceleration Feed Forward might be able to reduce this. Start with a very small number and increase until the initial lag is reduced as much as possible without causing a positive error.
HTH
Very good. But several things:
#1 the Acceleration is set extremely high relative to that level of Jerk. Jerk is normally set around 10X the Acceleration value in order to to apply the full acceleration in around 0.1 seconds. With those levels of A and J it would take 200 seconds to apply full acceleration. And by this time the max V would be far exceeded. So the acceleration is entirely jerk limited (triangular rather than trapezoidal). There is nothing necessarily wrong with entirely Jerk limited motion but mainly the A value can be misleading.
We have an included tool called TripsWPF.exe that will plot acceleration with any parameters. Here is an example with your settings:
#2 You didn't post the IIR Filters Screen but I'm guessing the filters are all cleared. Notice the Output (green) has something like 150 count spikes in it. This is because with D gain of 150 each encoder count change causes a 150 count impulse. Often the Amplifier can't react to this and effectively smooths out the spikes but adding a 2nd order Low Pass filter of around 500Hz Q=1.4 should reduce the spikes. (This is normally placed in Filter #2 so it also applies to feed forward).
#3 The largest error occurs as a lag during the acceleration and as an overshoot during deceleration. Acceleration Feed Forward might be able to reduce this. Start with a very small number and increase until the initial lag is reduced as much as possible without causing a positive error.
HTH
Regards,
Tom Kerekes
Dynomotion, Inc.
Tom Kerekes
Dynomotion, Inc.
Re: Fawne Fadal, Unchained. (4020 Fadal Retrofit)
Tom, thanks for the insight! That Trip States program seems quite nifty.
1) I will spend some more time with the J and A values.
2) Correct, my IIR screen has zeros across the board currently. Will get the 500Hz Q=1.4 added to FIlter #2.
3) I was actually thinking about adding some feed-forward! I have just not gotten to it yet. In day-to-day use I doubt the machine will ever have more than 200lbs on the table so could use 0-200lbs of table weights as a 'proving range' for feed-forward values.
The tuning of the PID took me a while to get to those numbers because they were way bigger than I thought they would be. 75 P and 150 D just seemed really large to all the other examples I have read through.
Even with these large numbers, I was getting an error plot with a lot less error than the OE drives so was quite pleased.
Thursday afternoon I think I may have started to find my reason/problem.
Wednesday I spent some time removing a big rat's nest of unused wires. Felt great to not be tripping over them anymore. Also added in and wired up drives for Y and Z. On machine power-up, a few things happened:
A) X-axis would run away regardless of gain polarity speed.
B) No signal from the encoder on Y.
C) Amp power stage would not enable.
For the X-axis issue, I eventually tried a P of just 1 and gained control (no runaway). So my tune went bad somehow. For the encoder, the OEM bracket that used to hold the resolver was .050 shorter than the others! This ruined the encoder by putting a lot of force on the input shaft from the motor output shaft. Would have been nice to catch this before the motor was reinstalled under the machine... Lastly, this took me a few hours and a scope, I had a DC voltage float(ground issue) on my OE power supply. The Dynomotion boards and the AMP chassis are all grounded to ground, but we were not getting a low enough signal to the SSR that enables the emergency circuit. Landed a wire to make the old DC power supply relative to everything else and we were off to the races. I believe this may have been affecting the signal voltage to the amps and tune. After this, the X and Z started responding quite nicely to initial tuning steps with P under 5. Y axis has something going on mechanically, I think the PO had the oldham coupler shimmed instead of replacing it.
The current plan is to get the Y sorted mechanically, get XYZ tuned and better my A/J/FF values. I will report back when I have more.
1) I will spend some more time with the J and A values.
2) Correct, my IIR screen has zeros across the board currently. Will get the 500Hz Q=1.4 added to FIlter #2.
3) I was actually thinking about adding some feed-forward! I have just not gotten to it yet. In day-to-day use I doubt the machine will ever have more than 200lbs on the table so could use 0-200lbs of table weights as a 'proving range' for feed-forward values.
The tuning of the PID took me a while to get to those numbers because they were way bigger than I thought they would be. 75 P and 150 D just seemed really large to all the other examples I have read through.
Even with these large numbers, I was getting an error plot with a lot less error than the OE drives so was quite pleased.
Thursday afternoon I think I may have started to find my reason/problem.
A) X-axis would run away regardless of gain polarity speed.
B) No signal from the encoder on Y.
C) Amp power stage would not enable.
For the X-axis issue, I eventually tried a P of just 1 and gained control (no runaway). So my tune went bad somehow. For the encoder, the OEM bracket that used to hold the resolver was .050 shorter than the others! This ruined the encoder by putting a lot of force on the input shaft from the motor output shaft. Would have been nice to catch this before the motor was reinstalled under the machine... Lastly, this took me a few hours and a scope, I had a DC voltage float(ground issue) on my OE power supply. The Dynomotion boards and the AMP chassis are all grounded to ground, but we were not getting a low enough signal to the SSR that enables the emergency circuit. Landed a wire to make the old DC power supply relative to everything else and we were off to the races. I believe this may have been affecting the signal voltage to the amps and tune. After this, the X and Z started responding quite nicely to initial tuning steps with P under 5. Y axis has something going on mechanically, I think the PO had the oldham coupler shimmed instead of replacing it.
The current plan is to get the Y sorted mechanically, get XYZ tuned and better my A/J/FF values. I will report back when I have more.