elastoplastic extension #560

mathl79 · 2022-11-03T22:14:02Z

implementation of the plasticity extension idea to the elasicity plugin.
see: discussion #549

new xml config parameter yield stress ("yield" [Pa]). Yield separates the elastic from the plastic domain
generation of userdata output: von mieses equivalent stress, curvature, strain(change of curvature)
skin/composit stress coloring
skin per vertex color added to the datamodel
CMAKE config of elasticity.dll changed to destination output folder \plugin\
new test model xml file

googletest is not implemented yet --> need help to find sufficient test criterias

skin/composit stress coloring implementation

google-cla · 2022-11-03T22:14:07Z

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

mathl79 · 2022-11-03T22:25:18Z

Hi!
I never started a pull request before.
So in order to fulfill the contribution requirements and adopt best practices, i am hoping for some beginner friendly remarks and some guidance.

quagla · 2022-11-04T17:54:57Z

Hi @mathl79 I think you have changed something in the default behavior of the cable? It looks like the test for the exact solution of the cantilever beam fails.

Great that you are visualizing the stresses, this is something I always wanted to do!

mathl79 · 2022-11-06T05:44:52Z

thank you for taking a look.
The string loaded elasticity test model does not have the required "yield" Parameter as it is a elastic test only.
I need to set a infinite yield value in case the "yield" Parameter in the XML is missing at the cable constructor.
In that way, the test model should be loaded and run as expected.

quagla · 2022-11-07T11:57:42Z

Do you mean that you updated CantileverIntoCircle in elasticity_test? I don't see that file in your commit (which is probably why the checks are failing)

mathl79 · 2022-11-08T06:57:00Z

The default elastic behavior should no be changed, if the new parameter "yield" is infinite. I missed to implement the default setting and remove the parameter check in this pull request, so the elasticity_test fails.

In the meantime i reworked my version to have a material model with strain hardening.

quagla · 2022-11-08T08:13:50Z

Great! I think it's fine if you keep pushing new commits to this branch, the PR should be updated automatically.

inner deformation friction corrected body coloring in case of multiple cable bodies

quagla · 2022-11-08T16:58:00Z

Thanks for the revised version! My main comments are:

Why did you add so much damping to the original cable examples?
Did you try the original examples and particularly to run elasticity_test? The default (purely elastic) behaviour has changed and it seems incorrect.
More generally, can we break the PR into a visualization and a plasticity PR or would that be impossible?

mathl79 · 2022-11-08T19:21:00Z

I tried to implement a plasticity deformation proportional damping by reducing the torque depending on the change of omega an the Volume.
Now i see, that this additional damping is overlapping with the elastic joint damping. This needs to be separated.

I did not run the tests - but better should have...

So I will change back to the standard plugin models, add extra plastic test models and separate the viscous damping by only applying it, if elements are in plastic domain.

I try to separate the coloring feature from the plasticity as you suggested.
For the coloring, i assume it should be configurable by XML options, right ?
-definition of range (min, max, auto)
-constant RGB color below lower limit
-constant RGB color above upper limit
-color palette
-enable/disable

What else needs to be user controllable ?

Should any scalar properties other than equivalent stress be selectable in future ?

Besides the separation from the plasticity, zhe stress needs to be calculated for the elasticity model anyway, in order to be able to show it.

Btw - what is a good place in the source for the color conversion and palette generation tools, that I have put inside the plugin code from now?

Is this something, that could be used generally or should it stay plugin specific?

quagla · 2022-11-08T21:20:14Z

Yes always run tests before committing :)

For the stress visualisation, those are really good questions. I think plugins should decide what to write in your skinuserdata array (as you have done), but if nothing is written, then you have to make sure the current behaviour is maintained. In the future, maybe also the core engine will write other quantities in that array. So, those colouring options could be added either to the visual or more likely to the skin section of the XML, while which quantity to write (e.g. stress) should be an option of the specific plugin. So disabled should be the default behaviour and will be automatically enabled if the plugin writes something.

@yuvaltassa WDYT? Any better idea than using skins for stress visualization?

mathl79 · 2022-11-09T09:41:38Z

Hi,
i did run the tests on my windows machine and all configured tests do pass.
But the PluginTest.CantileverIntoCircle is not even executed. Also the github build check showes, that the windows-2022 build does not run the elasticity tests, but fails at installing the file elasticity.dll into the plugin folder.

I don't know how to correct the config tests.

When building /debugging elasticity.dll, the file is placed into the /bin folder and needs to be copied into the /bin/plugin/ subfolder manually to be recognized as plugin.

I think, that is also linked to the problem above.

quagla · 2022-11-09T09:53:36Z

OK that's interesting, there is something broken with Windows then because on the main branch the dll gets copied. @saran-t what are your thoughts?

As mentioned in my previous comment, could you please open a new PR for the visualisation part? You can keep this one open as well and we will get back to it once we complete the visualisation. How does that sound to you? Maybe in the meantime the Windows issues will be fixed.

mathl79 · 2022-11-09T09:54:03Z

https://github.com/deepmind/mujoco/blob/7b7fc09d1fc3d9a970b2123bc41c4d1fbcff35d6/test/CMakeLists.txt#L69-L71

This code excludes the plugin-test from the test config for Windows builds. what is the reason for this?

quagla · 2022-11-09T10:07:41Z

Good point, we definitely need to fix Windows. Now I remember we disabled the test because we had some issues loading the DLL. I wonder if your change to CMakeList fixes that? Perhaps you want to open a third PR with only that change and also enabling again the plugin tests for Windows by deleting the lines you quoted in your post?

quagla · 2022-11-11T09:53:45Z

My suggestion is that we can start from the visualization PR since that doesn't requires the windows tests.

quagla · 2022-11-17T13:08:26Z

Hi @mathl79 I'm pleased to let you know that now you can run plugin tests on Windows as well, see a9781d9

mathl79 · 2022-11-17T13:22:01Z

Perfect, thanks!
I was also at the point to implement the runtime dll loading, as I was not able to link the plugin into the elasticity-test executable at compile time.

mathl79 · 2022-11-17T13:24:50Z

btw. - you can test the branch for the stress skin visuals without the plasticity in my fork.

quagla · 2022-11-17T14:56:40Z

Looks like we have to rollback the commit unfortunately, hopefully we'll be able to fix it and activate the windows tests in the next days.

quagla · 2022-11-17T17:28:38Z

Good news, Windows is fixed if you rebase from 043ed27 you'll be able to run Windows tests.

For the visualisation, are you going to create a separate PR from your branch or would you prefer if we have a look and apply the changes ourselves?

mathl79 · 2022-11-17T18:42:40Z

Perfect ! Thanks for fixing again!

For me it absolutely ok, if you take my approach as basis to implement the coloring by our own.
I do have the feeling, that my implementation is not generic enough to fit into a public release.

quagla · 2022-11-28T11:04:35Z

Hi @mathl79 we are having a look the colouring internally (see #599).

I wanted to ask you what is the rationale for the Von Mises stress computed directly from the curvature stresses without the membrane ones? Does this plastic model fit your needs?

mathl79 · 2022-11-29T06:24:18Z

Hi @quagla,
my intention for the coloring was just to have an visual indicator for the effects of plasticity.
The average bending stress is for my case enough information and the projected surface stress calculation of the skin elements was/is to difficult for me.
Also the coloring on bone level allows non-skin composites to be colorized.

Regarding the plasticity model:
In the meantime i changed my plasticity model in the way, that omega0 remains unchanged, but the internal bending stress / remaining strain is stored. In that way, the hysteresis behavior is working correctly now, which i did not achive with my first attempt.

I hope, that i will have some spare time the next days to finalize my testing and commit that change into my branch.

br,

Mathias

quagla · 2022-11-29T10:32:32Z

Great to hear! Do you think you can write a unit test in elasticity_test comparing the reference vs computed hysteresis so any future change will make sure to not break your intended behaviour?

For the skin colouring, after an internal discussion it was decided to do it in a different way so we will have to implement it ourselves. I agree however that colouring geometries (bones) is probably already giving plenty of information so when you are committing your final changes you could perhaps leave the geom colouring but remove the skin part if you can?

By the way, I added a Jet color palette in my branch, maybe you like it https://github.com/deepmind/mujoco/pull/599/files#diff-f8ed3e4597989d57bd3588f824c10fe89b140ca2665a05dd3b8878f2f6f6c23b

into mujoco_elastoplastic

quagla · 2022-11-30T10:27:40Z

Thanks for the update!

Tip: I think you have to set the parameters of CantileverIntoCircle such that perfect elasticity is obtained without triggering a warning because warnings cause test failures.

[ RUN ] PluginTest.CantileverIntoCircle
D:\a\mujoco\mujoco\test\fixture.cc(43): error: Failed
mju_user_warning: YieldStress is notset: perfect elasticity is assumed !

quagla

I left some comments. First of all, great work! However some things must be address before continuing:

The original tests and the original examples must be working.
A new test for plasticity should be added.
It would be great to remove the visualization part from the branch, unless absolutely necessary (we should use the test to validate the results instead).

quagla · 2022-12-01T14:02:06Z

plugin/elasticity/cable.cc

+  else {
+      dYield_UtY = UtStress - YieldStress; //strain hardening stress delta
+      k_deltaStrain_G = 5 / (UtStrain - YieldStrain_G); //exponential constant for torsional strain hardening
+      k_deltaStrain_E = 5 / (UtStrain - YieldStrain_E); //exponential constant for bending strain hardening


Are these 3 variables unused?

quagla · 2022-12-01T14:02:50Z

plugin/elasticity/cable.cc

@@ -26,6 +26,29 @@
 namespace mujoco::plugin::elasticity {
 namespace {

+
+//create color scale palette 
+void scalar2rgba(float rgba[4], mjtNum Scalar, mjtNum min, mjtNum max) {


As mentioned in the main discussion, it would be great to remove the coloring part from this branch.

quagla · 2022-12-01T14:03:17Z

plugin/elasticity/cable.cc

-      - stiffness[2]*(omega[2] - omega0[2]) / stiffness[3],
-  };
-
+void LocalForce(mjtNum qfrc[3], mujoco::plugin::elasticity::Cable::stiffness_ Sel, mujoco::plugin::elasticity::Cable::stiffness_consts_ Sconst,


Is the full namespace needed?

quagla · 2022-12-01T14:03:57Z

plugin/elasticity/cable.cc

+        omega[2] - omega0[2]
+    };
+
+    //actual strain


could you always add a space between // and the comment?

quagla · 2022-12-01T14:05:09Z

plugin/elasticity/cable.cc

+         d_omega[2] / (Sel.L_Dz + omega0[2]),
+    };
+
+    mjtNum Yield_stress_top[] = {


Could you document this plastic model with e.g. a paper or a link and/or a more detailed explanation in the code?

quagla · 2022-12-01T14:06:01Z

plugin/elasticity/cable.cc

+  }
+  }
+  else {
+      mju_warning("YieldStress is notset: perfect elasticity is assumed !");


Please update elasticity_test such that this warning is not triggered.

quagla · 2022-12-01T14:06:51Z

model/plugin/belt.xml

-      <joint kind="main" damping="0.1"/>
-      <geom type="box" size="0.03535533906 0.02 0.0015" rgba=".8 .2 .1 1" group="3"/>
-      <skin subgrid="3"/>
+      <joint kind="main" damping="10"/>


If I load this model, it immediately explodes.

quagla · 2022-12-01T14:07:31Z

model/plugin/coil.xml

      <geom type="capsule" size=".005" rgba=".8 .2 .1 1"/>
    </composite>
    <composite prefix="actuated" type="cable" curve="cos(s) sin(s) s" count="41 1 1"
      size="0.251327412 .1 4" offset="0.25 0 .05" initial="fixed">
      <plugin plugin="mujoco.elasticity.cable">
        <!--Units are in Pa (SI)-->
-        <config key="twist" value="5e8"/>
-        <config key="bend" value="15e8"/>
+        <config key="twist" value="1e10"/>


When I pull the second spring, at some point it looks like the simulation blows up.

quagla · 2022-12-02T19:36:28Z

We have added coloring of the geometries using a callback, please see 0d52fea

quagla · 2023-03-01T11:42:22Z

Hi @mathl79 are you still interested in submitting this?

elastoplastic extension

e3c6331

skin/composit stress coloring implementation

quagla self-requested a review November 4, 2022 11:10

mathl79 and others added 2 commits November 8, 2022 12:45

exponential strain hardening model implemented

ca21db1

inner deformation friction corrected body coloring in case of multiple cable bodies

Merge branch 'deepmind:main' into mujoco_elastoplastic

0786aa0

mathl79 marked this pull request as ready for review November 8, 2022 14:43

Merge branch 'deepmind:main' into mujoco_elastoplastic

ed50400

mathl79 and others added 3 commits November 30, 2022 08:27

hysteresis plasticity

836a12f

Merge branch 'mujoco_elastoplastic' of https://github.com/mathl79/mujoco

5735248

into mujoco_elastoplastic

Merge branch 'main' into mujoco_elastoplastic

51aa954

quagla requested changes Dec 1, 2022

View reviewed changes

elastoplastic extension #560

Are you sure you want to change the base?

elastoplastic extension #560

Conversation

mathl79 commented Nov 3, 2022 • edited

google-cla bot commented Nov 3, 2022

mathl79 commented Nov 3, 2022

quagla commented Nov 4, 2022

mathl79 commented Nov 6, 2022

quagla commented Nov 7, 2022 • edited

mathl79 commented Nov 8, 2022 • edited

quagla commented Nov 8, 2022

quagla commented Nov 8, 2022

mathl79 commented Nov 8, 2022

quagla commented Nov 8, 2022 • edited

mathl79 commented Nov 9, 2022

quagla commented Nov 9, 2022 • edited

mathl79 commented Nov 9, 2022

quagla commented Nov 9, 2022

quagla commented Nov 11, 2022

quagla commented Nov 17, 2022

mathl79 commented Nov 17, 2022

mathl79 commented Nov 17, 2022

quagla commented Nov 17, 2022

quagla commented Nov 17, 2022

mathl79 commented Nov 17, 2022

quagla commented Nov 28, 2022 • edited

mathl79 commented Nov 29, 2022 • edited

quagla commented Nov 29, 2022

quagla commented Nov 30, 2022

quagla left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

quagla commented Dec 2, 2022

quagla commented Mar 1, 2023

mathl79 commented Nov 3, 2022 •

edited

quagla commented Nov 7, 2022 •

edited

mathl79 commented Nov 8, 2022 •

edited

quagla commented Nov 8, 2022 •

edited

quagla commented Nov 9, 2022 •

edited

quagla commented Nov 28, 2022 •

edited

mathl79 commented Nov 29, 2022 •

edited