blog/archive/2022/07/07.rss

   1 <?xml version="1.0" encoding="ISO-8859-1"?>
   2 <rss version='2.0' xmlns:lj='http://www.livejournal.org/rss/lj/1.0/'>
   3         <channel>
   4                 <title>Petter Reinholdtsen - Entries from July 2022</title>
   5                 <description>Entries from July 2022</description>
   6                 <link>http://www.hungry.com/~pere/blog/</link>
   7
   8
   9         <item>
  10                 <title>Automatic LinuxCNC servo PID tuning?</title>
  11                 <link>http://www.hungry.com/~pere/blog/Automatic_LinuxCNC_servo_PID_tuning_.html</link>
  12                 <guid isPermaLink="true">http://www.hungry.com/~pere/blog/Automatic_LinuxCNC_servo_PID_tuning_.html</guid>
  13                 <pubDate>Sat, 16 Jul 2022 22:30:00 +0200</pubDate>
  14                 <description>&lt;p&gt;While working on a CNC with servo motors controlled by the
  15 &lt;a href=&quot;https://en.wikipedia.org/wiki/LinuxCNC&quot;&gt;LinuxCNC&lt;/a&gt;
  16 &lt;a href=&quot;https://en.wikipedia.org/wiki/PID_controller&quot;&gt;PID
  17 controller&lt;/a&gt;, I recently had to learn how to tune the collection of values
  18 that control such mathematical machinery that a PID controller is.  It
  19 proved to be a lot harder than I hoped, and I still have not succeeded
  20 in getting the Z PID controller to successfully defy gravity, nor X
  21 and Y to move accurately and reliably.  But while climbing up this
  22 rather steep learning curve, I discovered that some motor control
  23 systems are able to tune their PID controllers.  I got the impression
  24 from the documentation that LinuxCNC were not.  This proved to be not
  25 true.&lt;/p&gt;
  26
  27 &lt;p&gt;The LinuxCNC
  28 &lt;a href=&quot;http://linuxcnc.org/docs/html/man/man9/pid.9.html&quot;&gt;pid
  29 component&lt;/a&gt; is the recommended PID controller to use.  It uses eight
  30 constants &lt;tt&gt;Pgain&lt;/tt&gt;, &lt;tt&gt;Igain&lt;/tt&gt;, &lt;tt&gt;Dgain&lt;/tt&gt;,
  31 &lt;tt&gt;bias&lt;/tt&gt;, &lt;tt&gt;FF0&lt;/tt&gt;, &lt;tt&gt;FF1&lt;/tt&gt;, &lt;tt&gt;FF2&lt;/tt&gt; and
  32 &lt;tt&gt;FF3&lt;/tt&gt; to calculate the output value based on current and wanted
  33 state, and all of these need to have a sensible value for the
  34 controller to behave properly.  Note, there are even more values
  35 involved, theser are just the most important ones.  In my case I need
  36 the X, Y and Z axes to follow the requested path with little error.
  37 This has proved quite a challenge for someone who have never tuned a
  38 PID controller before, but there is at least some help to be found.
  39
  40 &lt;p&gt;I discovered that included in LinuxCNC was this old PID component
  41 at_pid claiming to have auto tuning capabilities.  Sadly it had been
  42 neglected since 2011, and could not be used as a plug in replacement
  43 for the default pid component.  One would have to rewriting the
  44 LinuxCNC HAL setup to test at_pid.  This was rather sad, when I wanted
  45 to quickly test auto tuning to see if it did a better job than me at
  46 figuring out good P, I and D values to use.&lt;/p&gt;
  47
  48 &lt;p&gt;I decided to have a look if the situation could be improved.  This
  49 involved trying to understand the code and history of the pid and
  50 at_pid components.  Apparently they had a common ancestor, as code
  51 structure, comments and variable names were quite close to each other.
  52 Sadly this was not reflected in the git history, making it hard to
  53 figure out what really happened.  My guess is that the author of
  54 &lt;a href=&quot;https://github.com/LinuxCNC/linuxcnc/blob/master/src/hal/components/at_pid.c&quot;&gt;at_pid.c&lt;/a&gt;
  55 took a version of
  56 &lt;a href=&quot;https://github.com/LinuxCNC/linuxcnc/blob/master/src/hal/components/pid.c&quot;&gt;pid.c&lt;/a&gt;,
  57 rewrote it to follow the structure he wished pid.c to have, then added
  58 support for auto tuning and finally got it included into the LinuxCNC
  59 repository.  The restructuring and lack of early history made it
  60 harder to figure out which part of the code were relevant to the auto
  61 tuning, and which part of the code needed to be updated to work the
  62 same way as the current pid.c implementation.  I started by trying to
  63 isolate relevant changes in pid.c, and applying them to at_pid.c.  My
  64 aim was to make sure the at_pid component could replace the pid
  65 component with a simple change in the HAL setup loadrt line, without
  66 having to &quot;rewire&quot; the rest of the HAL configuration.  After a few
  67 hours following this approach, I had learned quite a lot about the
  68 code structure of both components, while concluding I was heading down
  69 the wrong rabbit hole, and should get back to the surface and find a
  70 different path.&lt;/p&gt;
  71
  72 &lt;p&gt;For the second attempt, I decided to throw away all the PID control
  73 related part of the original at_pid.c, and instead isolate and lift
  74 the auto tuning part of the code and inject it into a copy of pid.c.
  75 This ensured compatibility with the current pid component, while
  76 adding auto tuning as a run time option.  To make it easier to identify
  77 the relevant parts in the future, I wrapped all the auto tuning code
  78 with &#39;#ifdef AUTO_TUNER&#39;.  The end result behave just like the current
  79 pid component by default, as that part of the code is identical.  The
  80 &lt;a href=&quot;https://github.com/LinuxCNC/linuxcnc/pull/1820&quot;&gt;end result
  81 entered the LinuxCNC master branch&lt;/a&gt; a few days ago.&lt;/p&gt;
  82
  83 &lt;p&gt;To enable auto tuning, one need to set a few HAL pins in the PID
  84 component.  The most important ones are &lt;tt&gt;tune-effort&lt;/tt&gt;,
  85 &lt;tt&gt;tune-mode&lt;/tt&gt; and &lt;tt&gt;tune-start&lt;/tt&gt;.  But lets take a step
  86 back, and see what the auto tuning code will do.  I do not know the
  87 mathematical foundation of the at_pid algorithm, but from observation
  88 I can tell that the algorithm will, when enabled, produce a square
  89 wave pattern centered around the &lt;tt&gt;bias&lt;/tt&gt; value on the output pin
  90 of the PID controller.  This can be seen using the HAL Scope provided
  91 by LinuxCNC.  In my case, this is translated into voltage (+-10V) sent
  92 to the motor controller, which in turn is translated into motor speed.
  93 So at_pid will ask the motor to move the axis back and forth.  The
  94 number of cycles in the pattern is controlled by the
  95 &lt;tt&gt;tune-cycles&lt;/tt&gt; pin, and the extremes of the wave pattern is
  96 controlled by the &lt;tt&gt;tune-effort&lt;/tt&gt; pin.  Of course, trying to
  97 change the direction of a physical object instantly (as in going
  98 directly from a positive voltage to the equivalent negative voltage)
  99 do not change velocity instantly, and it take some time for the object
 100 to slow down and move in the opposite direction.  This result in a
 101 more smooth movement wave form, as the axis in question were vibrating
 102 back and forth.  When the axis reached the target speed in the
 103 opposing direction, the auto tuner change direction again.  After
 104 several of these changes, the average time delay between the &#39;peaks&#39;
 105 and &#39;valleys&#39; of this movement graph is then used to calculate
 106 proposed values for Pgain, Igain and Dgain, and insert them into the
 107 HAL model to use by the pid controller.  The auto tuned settings are
 108 not great, but htye work a lot better than the values I had been able
 109 to cook up on my own, at least for the horizontal X and Y axis.  But I
 110 had to use very small &lt;tt&gt;tune-effort&lt;tt&gt; values, as my motor
 111 controllers error out if the voltage change too quickly.  I&#39;ve been
 112 less lucky with the Z axis, which is moving a heavy object up and
 113 down, and seem to confuse the algorithm.  The Z axis movement became a
 114 lot better when I introduced a &lt;tt&gt;bias&lt;/tt&gt; value to counter the
 115 gravitational drag, but I will have to work a lot more on the Z axis
 116 PID values.&lt;/p&gt;
 117
 118 &lt;p&gt;Armed with this knowledge, it is time to look at how to do the
 119 tuning.  Lets say the HAL configuration in question load the PID
 120 component for X, Y and Z like this:&lt;/p&gt;
 121
 122 &lt;blockquote&gt;&lt;pre&gt;
 123 loadrt pid names=pid.x,pid.y,pid.z
 124 &lt;/pre&gt;&lt;/blockquote&gt;
 125
 126 &lt;p&gt;Armed with the new and improved at_pid component, the new line will
 127 look like this:&lt;/p&gt;
 128
 129 &lt;blockquote&gt;&lt;pre&gt;
 130 loadrt at_pid names=pid.x,pid.y,pid.z
 131 &lt;/pre&gt;&lt;/blockquote&gt;
 132
 133 &lt;p&gt;The rest of the HAL setup can stay the same.  This work because the
 134 components are referenced by name.  If the component had used count=3
 135 instead, all use of pid.# had to be changed to at_pid.#.&lt;/p&gt;
 136
 137 &lt;p&gt;To start tuning the X axis, move the axis to the middle of its
 138 range, to make sure it do not hit anything when it start moving back
 139 and forth.  Next, set the &lt;tt&gt;tune-effort&lt;/tt&gt; to a low number in the
 140 output range.  I used 0.1 as my initial value.  Next, assign 1 to the
 141 &lt;tt&gt;tune-mode&lt;/tt&gt; value. Note, this will disable the pid controlling
 142 part and feed 0 to the output pin, which in my case initially caused a
 143 lot of drift.  In my case it proved to be a good idea with X and Y to
 144 tune the motor driver to make sure 0 voltage stopped the motor
 145 rotation.  On the other hand, for the Z axis this proved to be a bad
 146 idea, so it will depend on your setup.  It might help to set the
 147 &lt;tt&gt;bias&lt;/tt&gt; value to a output value that reduce or eliminate the
 148 axis drift.  Finally, after setting &lt;tt&gt;tune-mode&lt;/tt&gt;, set
 149 &lt;tt&gt;tune-start&lt;/tt&gt; to 1 to activate the auto tuning.  If all go well,
 150 your axis will vibrate for a few seconds and when it is done, new
 151 values for Pgain, Igain and Dgain will be active.  To test them,
 152 change &lt;tt&gt;tune-mode&lt;/tt&gt; back to 0.  Note that this might cause the
 153 machine to suddenly jerk as it bring the axis back to its commanded
 154 position, which it might have drifted away from during tuning.  To
 155 summarize with some halcmd lines:&lt;/p&gt;
 156
 157 &lt;blockquote&gt;&lt;pre&gt;
 158 setp pid.x.tune-effort 0.1
 159 setp pid.x.tune-mode 1
 160 setp pid.x.tune-start 1
 161 # wait for the tuning to complete
 162 setp pid.x.tune-mode 0
 163 &lt;/pre&gt;&lt;/blockquote&gt;
 164
 165 &lt;p&gt;After doing this task quite a few times while trying to figure out
 166 how to properly tune the PID controllers on the machine in, I decided
 167 to figure out if this process could be automated, and wrote a script
 168 to do the entire tuning process from power on.  The end result will
 169 ensure the machine is powered on and ready to run, home all axis if it
 170 is not already done, check that the extra tuning pins are available,
 171 move the axis to its mid point, run the auto tuning and re-enable the
 172 pid controller when it is done.  It can be run several times.  Check
 173 out the
 174 &lt;a href=&quot;https://github.com/SebKuzminsky/MazakVQC1540/blob/bon-dev/scripts/run-auto-pid-tuner&quot;&gt;run-auto-pid-tuner&lt;/a&gt;
 175 script on github if you want to learn how it is done.&lt;/p&gt;
 176
 177 &lt;p&gt;My hope is that this little adventure can inspire someone who know
 178 more about motor PID controller tuning can implement even better
 179 algorithms for automatic PID tuning in LinuxCNC, making life easier
 180 for both me and all the others that want to use LinuxCNC but lack the
 181 in depth knowledge needed to tune PID controllers well.&lt;/p&gt;
 182
 183 &lt;p&gt;As usual, if you use Bitcoin and want to show your support of my
 184 activities, please send Bitcoin donations to my address
 185 &lt;b&gt;&lt;a href=&quot;bitcoin:15oWEoG9dUPovwmUL9KWAnYRtNJEkP1u1b&quot;&gt;15oWEoG9dUPovwmUL9KWAnYRtNJEkP1u1b&lt;/a&gt;&lt;/b&gt;.&lt;/p&gt;
 186 </description>
 187         </item>
 188
 189         </channel>
 190 </rss>