Add elastic inversion capability using batch materials #353

dewenyushu · 2023-06-09T23:42:15Z

Initial demonstration of stress gradient calculation using batch material in the elastic material inversion problem.

dewenyushu · 2023-06-09T23:52:48Z

ping @zachmprince and @lynnmunday

moosebuild · 2023-06-12T16:05:59Z

Job Test on 42897d3 wanted to post the following:

View the site here

This comment will be updated on new commits.

Add documentation and tests Refs idaholab#352

hugary1995

Cool, thanks!

doc/content/source/userobjects/BatchStressGrad.md

hugary1995 · 2023-06-18T21:55:35Z

doc/content/source/vectorpostprocessors/AdjointStrainStressGradInnerProduct.md

+
+## Description
+
+This `VectorPostprocessor` computes the gradient of objective function with respect to interested parameter for elastic problem. Specifically,


First state the objective function please.

hugary1995 · 2023-06-18T21:57:34Z

doc/content/source/vectorpostprocessors/AdjointStrainStressGradInnerProduct.md

+\begin{equation}
+  \frac{\text{d}\boldsymbol{F}}{\text{d}p} = - \int \underbrace{(\boldsymbol{L}\lambda)^T}_{\text{adjoint strain}}  \underbrace{\frac{\partial \boldsymbol{C}}{\partial p} (\boldsymbol{L} u)}_{\text{stress gradient}} \text{d}\Omega,
+\end{equation}
+where $\boldsymbol{F}$ is the objective function, $\boldsymbol{L}$ is the differential operator for the elasticity problem, $\lambda$ is the adjoint displacement, $\boldsymbol{C}$ is the elastic tensor, $p$ is the interested parameter, and $u$ is displacement from the forward problem.


Suggested change

where $\boldsymbol{F}$ is the objective function, $\boldsymbol{L}$ is the differential operator for the elasticity problem, $\lambda$ is the adjoint displacement, $\boldsymbol{C}$ is the elastic tensor, $p$ is the interested parameter, and $u$ is displacement from the forward problem.

where $\boldsymbol{F}$ is the objective function, $\boldsymbol{L}$ is the symmetric strain operator for the elasticity problem, $\lambda$ is the adjoint displacement, $\boldsymbol{C}$ is the elasticity tensor, $p$ is the interested parameter, and $u$ is displacement from the forward problem.

It's not clear to me from this description whether p should be a scalar-valued property or can be vector-valued (or a set of scalar-valued properties). I guess I can find out in the actual source codes...

hugary1995 · 2023-06-18T21:59:12Z

doc/content/source/vectorpostprocessors/AdjointStrainStressGradInnerProduct.md

+\end{equation}
+where $\boldsymbol{F}$ is the objective function, $\boldsymbol{L}$ is the differential operator for the elasticity problem, $\lambda$ is the adjoint displacement, $\boldsymbol{C}$ is the elastic tensor, $p$ is the interested parameter, and $u$ is displacement from the forward problem.
+
+Specifically, this object takes the `adjoint strain` and the `stress gradient` as inputs, computes the double contraction of the two inputs, and integrate over the entire domain. Specifically, the `adjoint strain` is the strain from the adjoint problem, the `stress gradient` should be a batch material (e.g., [BatchStressGrad](/BatchStressGrad.md)).


I get a heart attack every time I see "stress gradient", but then later realize that you really mean the derivative of stress w.r.t. the the parameters which is an entirely different story.

Haha, sorry about the confusion. Yeah I meant the gradient of stress wrt to the interested parameter here but wanted it to be concise. I'll use gradient of stress w.r.t. parameter to hopefully avoid further confusion.

hugary1995 · 2023-06-18T22:00:47Z

include/userobjects/BatchStressGrad.h

+    // tuple representation
+    BatchMaterialUtils::TupleStd,
+    // output data type
+    RankTwoTensor,


So... this implies that p is a scalar-valued property. This is fine for now.

When is p not scalar valued?

We will need multiple such objects for each p component for that case

ya, as Dewen said we need multiple instances of the object to calculate the derivatives w.r.t. each p, for example we need two objects for the two Lame parameters (more when the stress-strain relation has more parameters). I feel like there will be an overhead in doing this. But I'm not sure how much that overhead will cost us.

hugary1995 · 2023-06-18T22:03:10Z

src/userobjects/BatchStressGrad.C

+{
+  auto params = BatchStressGradParent::validParams();
+  params.addRequiredParam<MaterialPropertyName>(
+      "elastic_tensor_derivative", "Name of the elastic tensor derivative material property.");


Suggested change

"elastic_tensor_derivative", "Name of the elastic tensor derivative material property.");

"elasticity_tensor_derivative", "Name of the elasticity tensor derivative material property.");

here and a few other places: there's no such a thing called elastic tensor.

hugary1995 · 2023-06-18T22:06:34Z

src/userobjects/BatchStressGrad.C

+
+void
+BatchStressGrad::batchCompute()
+{


Perhaps a comment here or in the header stating that this is a placeholder until we can swap this out for "actually" batched computation.

Good point, will do!

hugary1995 · 2023-06-18T22:07:06Z

src/vectorpostprocessors/AdjointStrainStressGradInnerProduct.C

+  params.addRequiredParam<MaterialPropertyName>(
+      "adjoint_strain_name", "Name of the strain property in the adjoint problem");
+
+  params.addClassDescription("Compute the gradient for elastic material inversion");


Suggested change

params.addClassDescription("Compute the gradient for elastic material inversion");

params.addClassDescription("Compute the parameter gradient for linear elastic material inversion");

Co-authored-by: Gary (Tianchen) Hu <thu@anl.gov>

moosebuild · 2023-06-19T17:31:48Z

Job Precheck on 87062dd wanted to post the following:

Your code requires style changes.

A patch was auto generated and copied here
You can directly apply the patch by running, in the top level of your repository:

curl -s https://mooseframework.inl.gov/blackbear/docs/PRs/353/clang_format/style.patch | git apply -v

Alternatively, with your repository up to date and in the top level of your repository:

git clang-format 256094a972d35961f4c4df64c7139b59fc2f51a7

dewenyushu · 2023-06-19T17:56:30Z

@hugary1995 Thanks for your review! I guess this indicates a viable path towards using NEML stress derivatives in inverse optimization problems? ;)

If so I'll focus on resolving the civet tests failures and ping you again once I think it is ready

hugary1995 · 2023-06-19T18:06:34Z

@hugary1995 Thanks for your review! I guess this indicates a viable path towards using NEML stress derivatives in inverse optimization problems? ;)

Yes, I see that you are getting correct answers in the regression test :)
I'll be working on NEML2 next month.

If so I'll focus on resolving the civet tests failures and ping you again once I think it is ready

Yes, please.

zachmprince · 2023-06-20T22:55:27Z

doc/config.yml

@@ -19,6 +19,8 @@ Content:
        root_dir: ${MOOSE_DIR}/modules/stochastic_tools/doc/content
    ray_tracing:
        root_dir: ${MOOSE_DIR}/modules/ray_tracing/doc/content
+    optimization:
+        root_dir: ${MOOSE_DIR}/modules/optimization/doc/content


Add this to the modules:/content: list below:

- help/development/analyze_jacobian.md

zachmprince · 2023-06-20T22:56:04Z

doc/config.yml

@@ -83,6 +85,7 @@ Extensions:
            misc: !include ${MOOSE_DIR}/modules/misc/doc/sqa_misc.yml
            xfem: !include ${MOOSE_DIR}/modules/xfem/doc/sqa_xfem.yml
            ray_tracing: !include ${MOOSE_DIR}/modules/ray_tracing/doc/sqa_ray_tracing.yml
+            optimization: !include ${MOOSE_DIR}/modules/optimization/doc/sqa_optimization.yml


Add this to enable the algorithm extension:

MooseDocs.extensions.algorithm: active: True

Cool! Thanks Zach!

moosebuild · 2023-06-21T15:47:56Z

Job Coverage on 42897d3 wanted to post the following:

Coverage

	af945f	#353 42897d
	Total	Total	+/-	New
Rate	94.25%	94.29%	+0.04%	96.77%
Hits	1736	1766	+30	30
Misses	106	107	+1	1

Diff coverage report

Full coverage report

This comment will be updated on new commits.

dewenyushu · 2023-06-21T21:35:59Z

This should be ready @hugary1995

hugary1995

Looks good to me. I only have one question.

hugary1995 · 2023-06-25T22:14:42Z

doc/content/source/vectorpostprocessors/AdjointStrainStressGradInnerProduct.md

+The gradient of objective function with respect to one scalar parameter ($p_i$) is
+\begin{equation}
+  \frac{\text{d}{F}}{\text{d}p_i} = - \int \underbrace{(\boldsymbol{L}\lambda)^T}_{\text{adjoint strain}}  \underbrace{\frac{\partial \boldsymbol{C}}{\partial p_i} (\boldsymbol{L} u)}_{\text{gradient of stress}} \text{d}\Omega,
+\end{equation}
+where ${F}$ is the objective function, $\boldsymbol{L}$ is the symmetric strain operator for the elasticity problem, $\lambda$ is the adjoint displacement, $\boldsymbol{C}$ is the elasticity tensor, and $p_i$ is the interested scalar parameter.


What happens to the regularization term? Do we assume it to be zero, or perhaps dR/dp is handled elsewhere?

Regularization, I believe, will happen on the main app side. Either in the solver or in the evaluation of the objective function. But we haven't implemented anything concretely yet. As such, I don't think it will be the model app's responsibility for computing the gradient.

I see. So I would envision some basic regularizations implemented in moose/modules/optimization, and each app could inherit and implement their own regularizations.

Either way, I think the regularization term will be handled by a different object, and so this documentation doesn't need any modification in that regard.

lynnmunday · 2023-06-23T20:14:54Z

src/userobjects/BatchStressGrad.C

+InputParameters
+BatchStressGrad::validParams()
+{
+  auto params = BatchStressGradParent::validParams();


Will you add a addClassDescription? Or is this going to be a test/src object?

Good point. Just did!

lynnmunday · 2023-06-26T23:07:42Z

include/userobjects/BatchStressGrad.h

+    // tuple representation
+    BatchMaterialUtils::TupleStd,
+    // output data type
+    RankTwoTensor,


When is p not scalar valued?

lynnmunday · 2023-06-26T23:33:05Z

@bwspenc or @dschwen Would one of you look at this PR. This is how we would like to interface with NEML2 for material parameter derivatives for inverse optimization.

bwspenc · 2023-06-27T19:18:21Z

include/userobjects/BatchStressGrad.h

@@ -0,0 +1,33 @@
+//* This file is part of the MOOSE framework


These files will all need the BlackBear header

lynnmunday · 2023-06-27T19:56:49Z

@dewenyushu we need to try moving this to isopod and include blackbear with isopod.

dewenyushu · 2023-06-27T23:14:36Z

Move contributions to idaholab/isopod#78

dewenyushu mentioned this pull request Jun 9, 2023

Batch material gradient for elastic property inversion idaholab/isopod#76

Closed

dewenyushu force-pushed the elastic_inversion branch from e494fc0 to 28cdd39 Compare June 9, 2023 23:43

dewenyushu force-pushed the elastic_inversion branch 4 times, most recently from bcb47f0 to 6028552 Compare June 15, 2023 20:03

Add elastic inversion capability using batch materials

c682a8d

Add documentation and tests Refs idaholab#352

dewenyushu force-pushed the elastic_inversion branch from 6028552 to c682a8d Compare June 15, 2023 20:38

hugary1995 suggested changes Jun 18, 2023

View reviewed changes

dewenyushu and others added 2 commits June 19, 2023 08:52

Update doc/content/source/userobjects/BatchStressGrad.md

225ffbb

Co-authored-by: Gary (Tianchen) Hu <thu@anl.gov>

Update doc/content/source/userobjects/BatchStressGrad.md

938e055

Co-authored-by: Gary (Tianchen) Hu <thu@anl.gov>

dewenyushu force-pushed the elastic_inversion branch from 87062dd to 29a5307 Compare June 19, 2023 17:36

zachmprince reviewed Jun 20, 2023

View reviewed changes

dewenyushu force-pushed the elastic_inversion branch from 29a5307 to ba3d672 Compare June 21, 2023 15:24

hugary1995 approved these changes Jun 25, 2023

View reviewed changes

lynnmunday reviewed Jun 26, 2023

View reviewed changes

Edit documentation and address comments

42897d3

dewenyushu force-pushed the elastic_inversion branch from ba3d672 to 42897d3 Compare June 27, 2023 00:42

bwspenc requested changes Jun 27, 2023

View reviewed changes

dewenyushu closed this Jun 27, 2023

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add elastic inversion capability using batch materials #353

Add elastic inversion capability using batch materials #353

dewenyushu commented Jun 9, 2023

dewenyushu commented Jun 9, 2023 •

edited

Loading

moosebuild commented Jun 12, 2023 •

edited

Loading

hugary1995 left a comment

hugary1995 Jun 18, 2023

hugary1995 Jun 18, 2023

hugary1995 Jun 18, 2023

dewenyushu Jun 19, 2023

hugary1995 Jun 18, 2023

lynnmunday Jun 26, 2023

dewenyushu Jun 27, 2023

hugary1995 Jun 27, 2023

hugary1995 Jun 18, 2023

hugary1995 Jun 18, 2023

dewenyushu Jun 19, 2023

hugary1995 Jun 18, 2023

moosebuild commented Jun 19, 2023

dewenyushu commented Jun 19, 2023

hugary1995 commented Jun 19, 2023

zachmprince Jun 20, 2023

zachmprince Jun 20, 2023

dewenyushu Jun 21, 2023

moosebuild commented Jun 21, 2023 •

edited

Loading

dewenyushu commented Jun 21, 2023

hugary1995 left a comment

hugary1995 Jun 25, 2023

zachmprince Jun 26, 2023

hugary1995 Jun 26, 2023

lynnmunday Jun 23, 2023

dewenyushu Jun 27, 2023

lynnmunday Jun 26, 2023

lynnmunday commented Jun 26, 2023

bwspenc Jun 27, 2023

lynnmunday commented Jun 27, 2023

dewenyushu commented Jun 27, 2023


		## Description

		This `VectorPostprocessor` computes the gradient of objective function with respect to interested parameter for elastic problem. Specifically,

	where $\boldsymbol{F}$ is the objective function, $\boldsymbol{L}$ is the differential operator for the elasticity problem, $\lambda$ is the adjoint displacement, $\boldsymbol{C}$ is the elastic tensor, $p$ is the interested parameter, and $u$ is displacement from the forward problem.
	where $\boldsymbol{F}$ is the objective function, $\boldsymbol{L}$ is the symmetric strain operator for the elasticity problem, $\lambda$ is the adjoint displacement, $\boldsymbol{C}$ is the elasticity tensor, $p$ is the interested parameter, and $u$ is displacement from the forward problem.

	"elastic_tensor_derivative", "Name of the elastic tensor derivative material property.");
	"elasticity_tensor_derivative", "Name of the elasticity tensor derivative material property.");

	params.addClassDescription("Compute the gradient for elastic material inversion");
	params.addClassDescription("Compute the parameter gradient for linear elastic material inversion");

		@@ -0,0 +1,33 @@
		//* This file is part of the MOOSE framework

Add elastic inversion capability using batch materials #353

Add elastic inversion capability using batch materials #353

Conversation

dewenyushu commented Jun 9, 2023

dewenyushu commented Jun 9, 2023 • edited Loading

moosebuild commented Jun 12, 2023 • edited Loading

hugary1995 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

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

moosebuild commented Jun 19, 2023

dewenyushu commented Jun 19, 2023

hugary1995 commented Jun 19, 2023

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

moosebuild commented Jun 21, 2023 • edited Loading

Coverage

dewenyushu commented Jun 21, 2023

hugary1995 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

lynnmunday commented Jun 26, 2023

Choose a reason for hiding this comment

lynnmunday commented Jun 27, 2023

dewenyushu commented Jun 27, 2023

dewenyushu commented Jun 9, 2023 •

edited

Loading

moosebuild commented Jun 12, 2023 •

edited

Loading

moosebuild commented Jun 21, 2023 •

edited

Loading