Implementing a software drag knife
- chowderhead
- Offline
- Senior Member
-
Less
More
- Posts: 57
- Thank you received: 11
29 Nov 2023 22:26 #286817
by chowderhead
Implementing a software drag knife was created by chowderhead
I've read the
QTPlasmaC with six axis
post and follow most of the discussion regarding the shortcomings in calculating heading using dx and dy. I'm not sure what the state tags are as discussed by
rodw
, maybe it's a better approach? Will someone please point me in the right direction to learn more?
At any rate, I have a heading rt component working and would at least like to test, but I'm not clear about how to wire the heading output (in degrees) to the C-axis/joint. I'm using a 7i96 card, if that matters.
At any rate, I have a heading rt component working and would at least like to test, but I'm not clear about how to wire the heading output (in degrees) to the C-axis/joint. I'm using a 7i96 card, if that matters.
Please Log in or Create an account to join the conversation.
30 Nov 2023 12:22 - 30 Nov 2023 12:26 #286864
by timo
Replied by timo on topic Implementing a software drag knife
I have no idea how to do it 
The a axis of Andy Pughs gear hobber runs synchronized with the spindle. I guess your problem is very similar to "electronic gearing",
I cannot find the thread right now, there was a complete hal file with a lot of useful tricks that might be worth reading through.
forum.linuxcnc.org/38-general-linuxcnc-q...er-slave-axes#256984 here might be something (I did not read it).
The a axis of Andy Pughs gear hobber runs synchronized with the spindle. I guess your problem is very similar to "electronic gearing",I cannot find the thread right now, there was a complete hal file with a lot of useful tricks that might be worth reading through.
forum.linuxcnc.org/38-general-linuxcnc-q...er-slave-axes#256984 here might be something (I did not read it).
Last edit: 30 Nov 2023 12:26 by timo.
The following user(s) said Thank You: chowderhead
Please Log in or Create an account to join the conversation.
30 Nov 2023 17:17 #286890
by Aciera
Replied by Aciera on topic Implementing a software drag knife
This is example hal config for an open loop stepper axis, your angular position would come in place of 'joint.3.motor-pos-command'.
#*******************
# --- AXIS A ---
#*******************
# --- PID SETUP ---
setp pid.a.Pgain [JOINT_3]P
setp pid.a.Igain [JOINT_3]I
setp pid.a.Dgain [JOINT_3]D
setp pid.a.bias [JOINT_3]BIAS
setp pid.a.FF0 [JOINT_3]FF0
setp pid.a.FF1 [JOINT_3]FF1
setp pid.a.FF2 [JOINT_3]FF2
setp pid.a.deadband [JOINT_3]DEADBAND
setp pid.a.maxoutput [JOINT_3]MAX_OUTPUT
setp pid.a.error-previous-target TRUE
setp pid.a.maxerror .01
# --- STEPGEN SETUP ---
setp hm2_7i76e.0.stepgen.03.dirsetup [JOINT_3]DIRSETUP
setp hm2_7i76e.0.stepgen.03.dirhold [JOINT_3]DIRHOLD
setp hm2_7i76e.0.stepgen.03.steplen [JOINT_3]STEPLEN
setp hm2_7i76e.0.stepgen.03.stepspace [JOINT_3]STEPSPACE
setp hm2_7i76e.0.stepgen.03.position-scale [JOINT_3]STEP_SCALE
setp hm2_7i76e.0.stepgen.03.step_type 0
setp hm2_7i76e.0.stepgen.03.control-type 1
setp hm2_7i76e.0.stepgen.03.maxaccel [JOINT_3]STEPGEN_MAXACCEL
setp hm2_7i76e.0.stepgen.03.maxvel [JOINT_3]STEPGEN_MAXVEL
#--- OPEN LOOP STEPPER SIGNALS ---
net a-index-enable <=> pid.a.index-enable
net a-enable <= joint.3.amp-enable-out => pid.a.enable => hm2_7i76e.0.stepgen.03.enable
net a-pos-cmd <= joint.3.motor-pos-cmd => pid.a.command
net a-vel-cmd <= joint.3.vel-cmd => pid.a.command-deriv
net a-output <= pid.a.output => hm2_7i76e.0.stepgen.03.velocity-cmd
net a-pos-fb <= hm2_7i76e.0.stepgen.03.position-fb => pid.a.feedback => joint.3.motor-pos-fb
# --- SETUP HOME / LIMIT SWITCH SIGNALS ---
net a-home-sw-auto <= hm2_7i76e.0.7i76.0.0.input-10-not
net a-home-sw => joint.3.home-sw-in
The following user(s) said Thank You: chowderhead
Please Log in or Create an account to join the conversation.
30 Nov 2023 17:20 #286891
by Aciera
Replied by Aciera on topic Implementing a software drag knife
The above is for a mesa 7i76e interface card.
State tags refers to an information channel in linuxcnc that would, ideally, publish the current heading of the trajectory.
Alas it does not do that because nobody has implemented yet.
State tags refers to an information channel in linuxcnc that would, ideally, publish the current heading of the trajectory.
Alas it does not do that because nobody has implemented yet.
Please Log in or Create an account to join the conversation.
01 Dec 2023 04:44 #286929
by rodw
github.com/LinuxCNC/linuxcnc/pull/900
Replied by rodw on topic Implementing a software drag knife
I started that. It would be worth resurrecting by somebody with better skills than I hadThe above is for a mesa 7i76e interface card.
State tags refers to an information channel in linuxcnc that would, ideally, publish the current heading of the trajectory.
Alas it does not do that because nobody has implemented yet.
github.com/LinuxCNC/linuxcnc/pull/900
The following user(s) said Thank You: Aciera
Please Log in or Create an account to join the conversation.
- chowderhead
- Offline
- Senior Member
-
Less
More
- Posts: 57
- Thank you received: 11
01 Dec 2023 17:45 #286979
by chowderhead
Replied by chowderhead on topic Implementing a software drag knife
I'm definitely not that guy - my C skills are nonexistent.
The post by Aciera highlighted the non-triviality of this endeavor given the heading component output, which is really joint.n.pos-cmd, needs to be translated to joint.n.motor-pos-cmd, which includes offsets (or vice versa) quickly but in small steps to keep from tossing following errors. My hacky approach is to limit drag knife functionality to only when executing G1, G2 or G3 while running a g-code file. Here's my logic yielding that conclusion, please pick it apart:
My C-axis has tiny inertia and is therefore the fastest axis on the machine; it should be able to keep up with changes in heading that the trajectory planner (I think that's the right component) comes up with for the much slower x and y axes given coordinated moves. This can't work when jogging or running single lines in MDI as the machine goes from no heading (stationary) to some heading (moving) instantaneously, but it should work when executing conterminous G1, G2 or G3 while running a g-code file. I need to figure out how to deal with interstitial G0's...
This goes out the window if a correction is made causing the heading to make a wild turn, which was pointed out in the QTPlasmaC post. I should be able to trap for that and just not change the heading when it occurs, or calculate a moving average heading (noise reduction). It may also be a problem if the updates from the 7i96 are laggy?
My cracked head says this yields drag knife-like functionality when running a g-code file, with the knife offset being fperiod, yes?
Would it be possible to implement a slaved joint, akin the dual joints used by gantry machines to drive a single axis, except the slaved joint has a commanded position calculated from two masters? The trick is driving the slaved joint to its target heading as quickly as possible instead of synchronizing the move.
Thanks for the help!
The post by Aciera highlighted the non-triviality of this endeavor given the heading component output, which is really joint.n.pos-cmd, needs to be translated to joint.n.motor-pos-cmd, which includes offsets (or vice versa) quickly but in small steps to keep from tossing following errors. My hacky approach is to limit drag knife functionality to only when executing G1, G2 or G3 while running a g-code file. Here's my logic yielding that conclusion, please pick it apart:
My C-axis has tiny inertia and is therefore the fastest axis on the machine; it should be able to keep up with changes in heading that the trajectory planner (I think that's the right component) comes up with for the much slower x and y axes given coordinated moves. This can't work when jogging or running single lines in MDI as the machine goes from no heading (stationary) to some heading (moving) instantaneously, but it should work when executing conterminous G1, G2 or G3 while running a g-code file. I need to figure out how to deal with interstitial G0's...
This goes out the window if a correction is made causing the heading to make a wild turn, which was pointed out in the QTPlasmaC post. I should be able to trap for that and just not change the heading when it occurs, or calculate a moving average heading (noise reduction). It may also be a problem if the updates from the 7i96 are laggy?
My cracked head says this yields drag knife-like functionality when running a g-code file, with the knife offset being fperiod, yes?
Would it be possible to implement a slaved joint, akin the dual joints used by gantry machines to drive a single axis, except the slaved joint has a commanded position calculated from two masters? The trick is driving the slaved joint to its target heading as quickly as possible instead of synchronizing the move.
Thanks for the help!
Please Log in or Create an account to join the conversation.
01 Dec 2023 19:41 #286987
by Aciera
Replied by Aciera on topic Implementing a software drag knife
You could also try to use a custom kinematic.
I attach my sim config of a disc saw. The custom kinematic 'xyzc_tangetial.comp' is included you can try out the config by installing the custom component using halcompile.
I attach my sim config of a disc saw. The custom kinematic 'xyzc_tangetial.comp' is included you can try out the config by installing the custom component using halcompile.
Attachments:
The following user(s) said Thank You: chowderhead
Please Log in or Create an account to join the conversation.
- chowderhead
- Offline
- Senior Member
-
Less
More
- Posts: 57
- Thank you received: 11
05 Dec 2023 16:48 #287289
by chowderhead
Replied by chowderhead on topic Implementing a software drag knife
Gotta say, this is the best forum I've ever been on. Thanks!
I actually got the open loop stepper component working pretty well, though it seems very hacky:
1) Set FERROR and MIN_FERROR ridiculously high, effectively eliminating any chance a following error will be thrown.
2) When the component is active, found that open loop eliminates MAX_VELOCITY and MAX_ACCELERATION from AXIS and JOINT, leaving only STEPGEN_MAX_VEL and STEPGEN_MAX_ACC active. Which makes sense.
3) Set STEPGEN settings to quite high values to get a quick reaction time; ends up being pretty damn quick.
4) Set threshold values for dx and dy, below which no heading update occurs, eliminating wild hairs on motion correction.
The only thing I don't like is, given the PID controller is P only, it has a noticeable overshoot. Any recommendations for minimizing it?
I actually got the open loop stepper component working pretty well, though it seems very hacky:
1) Set FERROR and MIN_FERROR ridiculously high, effectively eliminating any chance a following error will be thrown.
2) When the component is active, found that open loop eliminates MAX_VELOCITY and MAX_ACCELERATION from AXIS and JOINT, leaving only STEPGEN_MAX_VEL and STEPGEN_MAX_ACC active. Which makes sense.
3) Set STEPGEN settings to quite high values to get a quick reaction time; ends up being pretty damn quick.
4) Set threshold values for dx and dy, below which no heading update occurs, eliminating wild hairs on motion correction.
The only thing I don't like is, given the PID controller is P only, it has a noticeable overshoot. Any recommendations for minimizing it?
Please Log in or Create an account to join the conversation.
05 Dec 2023 17:40 #287294
by Aciera
Replied by Aciera on topic Implementing a software drag knife
Why is the PID p only? The PID hal component offers all you want.
# --- PID SETUP ---
setp pid.a.Pgain [JOINT_3]P
setp pid.a.Igain [JOINT_3]I
setp pid.a.Dgain [JOINT_3]D
setp pid.a.bias [JOINT_3]BIAS
setp pid.a.FF0 [JOINT_3]FF0
setp pid.a.FF1 [JOINT_3]FF1
setp pid.a.FF2 [JOINT_3]FF2
setp pid.a.deadband [JOINT_3]DEADBAND
setp pid.a.maxoutput [JOINT_3]MAX_OUTPUT
setp pid.a.error-previous-target TRUE
setp pid.a.maxerror .01Please Log in or Create an account to join the conversation.
- chowderhead
- Offline
- Senior Member
-
Less
More
- Posts: 57
- Thank you received: 11
05 Dec 2023 21:48 #287314
by chowderhead
Replied by chowderhead on topic Implementing a software drag knife
Truth is, I tried the classic PID tuning routine and only made it worse. Stock from Pncconf seems to be best.
I find that I can pretty much eliminate overshoot by tuning STEPGEN_MAX_ACC and STEPGEN_MAX_VEL.
I find that I can pretty much eliminate overshoot by tuning STEPGEN_MAX_ACC and STEPGEN_MAX_VEL.
Please Log in or Create an account to join the conversation.
Time to create page: 0.220 seconds