blog/data/2022-07-16-linuxcnc-pid-autotuning.txt

   1 Title: Automatic LinuxCNC servo PID tuning?
   2 Tags: english, debian, linuxcnc, 3d-printer, robot
   3 Date: 2022-07-16 22:30
   4
   5 <p>While working on a CNC with servo motors controlled by the
   6 <a href="https://en.wikipedia.org/wiki/LinuxCNC">LinuxCNC</a>
   7 <a href="https://en.wikipedia.org/wiki/PID_controller">PID
   8 controller</a>, I recently had to learn how to tune the collection of values
   9 that control such mathematical machinery that a PID controller is.  It
  10 proved to be a lot harder than I hoped, and I still have not succeeded
  11 in getting the Z PID controller to successfully defy gravity, nor X
  12 and Y to move accurately and reliably.  But while climbing up this
  13 rather steep learning curve, I discovered that some motor control
  14 systems are able to tune their PID controllers.  I got the impression
  15 from the documentation that LinuxCNC were not.  This proved to be not
  16 true.</p>
  17
  18 <p>The LinuxCNC
  19 <a href="http://linuxcnc.org/docs/html/man/man9/pid.9.html">pid
  20 component</a> is the recommended PID controller to use.  It uses eight
  21 constants <tt>Pgain</tt>, <tt>Igain</tt>, <tt>Dgain</tt>,
  22 <tt>bias</tt>, <tt>FF0</tt>, <tt>FF1</tt>, <tt>FF2</tt> and
  23 <tt>FF3</tt> to calculate the output value based on current and wanted
  24 state, and all of these need to have a sensible value for the
  25 controller to behave properly.  Note, there are even more values
  26 involved, theser are just the most important ones.  In my case I need
  27 the X, Y and Z axes to follow the requested path with little error.
  28 This has proved quite a challenge for someone who have never tuned a
  29 PID controller before, but there is at least some help to be found.
  30
  31 <p>I discovered that included in LinuxCNC was this old PID component
  32 at_pid claiming to have auto tuning capabilities.  Sadly it had been
  33 neglected since 2011, and could not be used as a plug in replacement
  34 for the default pid component.  One would have to rewriting the
  35 LinuxCNC HAL setup to test at_pid.  This was rather sad, when I wanted
  36 to quickly test auto tuning to see if it did a better job than me at
  37 figuring out good P, I and D values to use.</p>
  38
  39 <p>I decided to have a look if the situation could be improved.  This
  40 involved trying to understand the code and history of the pid and
  41 at_pid components.  Apparently they had a common ancestor, as code
  42 structure, comments and variable names were quite close to each other.
  43 Sadly this was not reflected in the git history, making it hard to
  44 figure out what really happened.  My guess is that the author of
  45 <a href="https://github.com/LinuxCNC/linuxcnc/blob/master/src/hal/components/at_pid.c">at_pid.c</a>
  46 took a version of
  47 <a href="https://github.com/LinuxCNC/linuxcnc/blob/master/src/hal/components/pid.c">pid.c</a>,
  48 rewrote it to follow the structure he wished pid.c to have, then added
  49 support for auto tuning and finally got it included into the LinuxCNC
  50 repository.  The restructuring and lack of early history made it
  51 harder to figure out which part of the code were relevant to the auto
  52 tuning, and which part of the code needed to be updated to work the
  53 same way as the current pid.c implementation.  I started by trying to
  54 isolate relevant changes in pid.c, and applying them to at_pid.c.  My
  55 aim was to make sure the at_pid component could replace the pid
  56 component with a simple change in the HAL setup loadrt line, without
  57 having to "rewire" the rest of the HAL configuration.  After a few
  58 hours following this approach, I had learned quite a lot about the
  59 code structure of both components, while concluding I was heading down
  60 the wrong rabbit hole, and should get back to the surface and find a
  61 different path.</p>
  62
  63 <p>For the second attempt, I decided to throw away all the PID control
  64 related part of the original at_pid.c, and instead isolate and lift
  65 the auto tuning part of the code and inject it into a copy of pid.c.
  66 This ensured compatibility with the current pid component, while
  67 adding auto tuning as a run time option.  To make it easier to identify
  68 the relevant parts in the future, I wrapped all the auto tuning code
  69 with '#ifdef AUTO_TUNER'.  The end result behave just like the current
  70 pid component by default, as that part of the code is identical.  The
  71 <a href="https://github.com/LinuxCNC/linuxcnc/pull/1820">end result
  72 entered the LinuxCNC master branch</a> a few days ago.</p>
  73
  74 <p>To enable auto tuning, one need to set a few HAL pins in the PID
  75 component.  The most important ones are <tt>tune-effort</tt>,
  76 <tt>tune-mode</tt> and <tt>tune-start</tt>.  But lets take a step
  77 back, and see what the auto tuning code will do.  I do not know the
  78 mathematical foundation of the at_pid algorithm, but from observation
  79 I can tell that the algorithm will, when enabled, produce a square
  80 wave pattern centered around the <tt>bias</tt> value on the output pin
  81 of the PID controller.  This can be seen using the HAL Scope provided
  82 by LinuxCNC.  In my case, this is translated into voltage (+-10V) sent
  83 to the motor controller, which in turn is translated into motor speed.
  84 So at_pid will ask the motor to move the axis back and forth.  The
  85 number of cycles in the pattern is controlled by the
  86 <tt>tune-cycles</tt> pin, and the extremes of the wave pattern is
  87 controlled by the <tt>tune-effort</tt> pin.  Of course, trying to
  88 change the direction of a physical object instantly (as in going
  89 directly from a positive voltage to the equivalent negative voltage)
  90 do not change velocity instantly, and it take some time for the object
  91 to slow down and move in the opposite direction.  This result in a
  92 more smooth movement wave form, as the axis in question were vibrating
  93 back and forth.  When the axis reached the target speed in the
  94 opposing direction, the auto tuner change direction again.  After
  95 several of these changes, the average time delay between the 'peaks'
  96 and 'valleys' of this movement graph is then used to calculate
  97 proposed values for Pgain, Igain and Dgain, and insert them into the
  98 HAL model to use by the pid controller.  The auto tuned settings are
  99 not great, but htye work a lot better than the values I had been able
 100 to cook up on my own, at least for the horizontal X and Y axis.  But I
 101 had to use very small <tt>tune-effort<tt> values, as my motor
 102 controllers error out if the voltage change too quickly.  I've been
 103 less lucky with the Z axis, which is moving a heavy object up and
 104 down, and seem to confuse the algorithm.  The Z axis movement became a
 105 lot better when I introduced a <tt>bias</tt> value to counter the
 106 gravitational drag, but I will have to work a lot more on the Z axis
 107 PID values.</p>
 108
 109 <p>Armed with this knowledge, it is time to look at how to do the
 110 tuning.  Lets say the HAL configuration in question load the PID
 111 component for X, Y and Z like this:</p>
 112
 113 <blockquote><pre>
 114 loadrt pid names=pid.x,pid.y,pid.z
 115 </pre></blockquote>
 116
 117 <p>Armed with the new and improved at_pid component, the new line will
 118 look like this:</p>
 119
 120 <blockquote><pre>
 121 loadrt at_pid names=pid.x,pid.y,pid.z
 122 </pre></blockquote>
 123
 124 <p>The rest of the HAL setup can stay the same.  This work because the
 125 components are referenced by name.  If the component had used count=3
 126 instead, all use of pid.# had to be changed to at_pid.#.</p>
 127
 128 <p>To start tuning the X axis, move the axis to the middle of its
 129 range, to make sure it do not hit anything when it start moving back
 130 and forth.  Next, set the <tt>tune-effort</tt> to a low number in the
 131 output range.  I used 0.1 as my initial value.  Next, assign 1 to the
 132 <tt>tune-mode</tt> value. Note, this will disable the pid controlling
 133 part and feed 0 to the output pin, which in my case initially caused a
 134 lot of drift.  In my case it proved to be a good idea with X and Y to
 135 tune the motor driver to make sure 0 voltage stopped the motor
 136 rotation.  On the other hand, for the Z axis this proved to be a bad
 137 idea, so it will depend on your setup.  It might help to set the
 138 <tt>bias</tt> value to a output value that reduce or eliminate the
 139 axis drift.  Finally, after setting <tt>tune-mode</tt>, set
 140 <tt>tune-start</tt> to 1 to activate the auto tuning.  If all go well,
 141 your axis will vibrate for a few seconds and when it is done, new
 142 values for Pgain, Igain and Dgain will be active.  To test them,
 143 change <tt>tune-mode</tt> back to 0.  Note that this might cause the
 144 machine to suddenly jerk as it bring the axis back to its commanded
 145 position, which it might have drifted away from during tuning.  To
 146 summarize with some halcmd lines:</p>
 147
 148 <blockquote><pre>
 149 setp pid.x.tune-effort 0.1
 150 setp pid.x.tune-mode 1
 151 setp pid.x.tune-start 1
 152 # wait for the tuning to complete
 153 setp pid.x.tune-mode 0
 154 </pre></blockquote>
 155
 156 <p>After doing this task quite a few times while trying to figure out
 157 how to properly tune the PID controllers on the machine in, I decided
 158 to figure out if this process could be automated, and wrote a script
 159 to do the entire tuning process from power on.  The end result will
 160 ensure the machine is powered on and ready to run, home all axis if it
 161 is not already done, check that the extra tuning pins are available,
 162 move the axis to its mid point, run the auto tuning and re-enable the
 163 pid controller when it is done.  It can be run several times.  Check
 164 out the
 165 <a href="https://github.com/SebKuzminsky/MazakVQC1540/blob/bon-dev/scripts/run-auto-pid-tuner">run-auto-pid-tuner</a>
 166 script on github if you want to learn how it is done.</p>
 167
 168 <p>My hope is that this little adventure can inspire someone who know
 169 more about motor PID controller tuning can implement even better
 170 algorithms for automatic PID tuning in LinuxCNC, making life easier
 171 for both me and all the others that want to use LinuxCNC but lack the
 172 in depth knowledge needed to tune PID controllers well.</p>
 173
 174 <p>As usual, if you use Bitcoin and want to show your support of my
 175 activities, please send Bitcoin donations to my address
 176 <b><a href="bitcoin:15oWEoG9dUPovwmUL9KWAnYRtNJEkP1u1b">15oWEoG9dUPovwmUL9KWAnYRtNJEkP1u1b</a></b>.</p>