Process load update and new uniform mesher (beta) #735

albeanth · 2022-06-24T22:08:25Z

Description

This addresses #705 and provides the beginnings of a new uniform mesher; though is not an ARMI converter like the original/current but is rather an ARMI changer.

The primary outcomes of this PR are:

Blueprint assembly definitions are allowed to be axially disjoint in a core and have detailedAxialExpansion: False.

The introduction of a new uniform mesher that aims to minimize (and in some cases remove) inter-block material smearing that is a result of the current uniform mesh converter

armi/armi/reactor/converters/uniformMesh.py

Lines 35 to 40 in 4894a56

    
           .. warning: This procedure can cause numerical diffusion in some cases. For example,  
        
               if a control rod tip block has a large coolant block below it, things like peak 
        
               absorption rate can get lost into it. We recalculate some but not all  
        
               reaction rates in the re-mapping process based on a flux remapping. To avoid this, 
        
               finer meshes will help. Always perform mesh sensitivity studies to ensure appropriate 
        
               convergence for your needs.

the new framework settings, primaryAssemblyToConserve and secondaryAssemblyToConserve provide options for users to provide strings that correspond to assembly Flags that group assemblies together and aim to best preserve their material interfaces. E.g., setting primaryAssemblyToConserve: "control" would give control assemblies highest priority in preserving their block bounds and would aim to minimize (or eliminate) inter-block material smearing (e.g., control into plenum).
Note, only assemblies of a single flag type can be grouped together.
The "primary" and "secondary" settings were included to to enable priorities for multiple assembly types.

Two other points:

the new unit tests that come with this PR are particularly heavy... Future commits will be included to alleviate their cost.

I included the new uniform mesher into the axial expansion changer here.

armi/armi/reactor/converters/axialExpansionChanger.py

Lines 288 to 295 in 4894a56

    
           # loop through again now that the reference is adjusted and adjust the non-fuel assemblies. 
        
           refAssem = r.core.refAssem 
        
           axMesh = refAssem.getAxialMesh() 
        
           for a in r.core.getAssemblies(includeBolAssems=True): 
        
               # See ARMI Ticket #112 for explanation of the commented out code 
        
               a.setBlockMesh( 
        
                   axMesh 
        
               )  # , conserveMassFlag=True, adjustList=adjustList)

To improve ease of use, I changed the constructor for class AxialExpansionChanger and had to make a slew of associated adjustments.

Checklist

The code is understandable and maintainable to people beyond the author.
Tests have been added/updated to verify that the new or changed code works.
There is no commented out code in this PR.
The commit message follows good practices.
All docstrings are still up-to-date with these changes.

Summary: - removing makeAxialSnapList and replacing with new uniform mesh changer - now enables disjoint assemblies to pass processLoading with detailedAxialExpansion: False

Summary: - to enable users to change which assembly groupings get mesh conservation priority, the variables "primaryAssemblyFlag" and "secondaryAssemblyFlag" were moved to be global settings. - also updating runLog statements to make ARMI output cleaner. Users can increase the verbosity to debug to include more information and even set "printLikeBlockSmears = True" in getCoreWideUniformMesh if they _really_ want a lot of info. This could be made a runtime setting, but it seems less useful in a general sense, so leaving it as a source code change seems reasonable.

Summary: - checkForSmearing had no self use, so it was made a function rather than a class method

Summary: - things are waaaaay clearer and more organized if moved over into armi.reactor.converters. And sets things up a little better to eventually replace the old uniformMesher - adding unit tests to ensure mass conservation at the block, assembly, and core levels. also has tests for exceptions to keep coverage up.

A few adjustments for other unit tests to pass: 1. applyCoreWideUniformMesh --> loop over getChildren() instead of getAssemblies to preserve assembly order in core. This is needed in test_reactors.py::HexReactorTests::test_isPickleable. 2. set self.core.p.axialMesh to self.core.finalAllAxialMeshPoints() --> this is needed to account for b.p.axMesh and the test in is needed in test_reactors.py::HexReactorTests::test_addMoreNodes. 3. insert 0.0 into uniMesher.unformMesh. Since core.p.axialMesh is set with findAllAxialMeshPoints(), the bottom mesh value of 0.0 is included now, so it needs to be added so this test passes. TODO: 1. What to do if self.core.findAllAxialMeshPoints() returns a mesh with sub 1 cm meshes? 2. During process load, we don't _alwayss_ need to call the uniform mesher if detailedAxial is False and if self if a core. Really, it only needs to be run if the blueprints assemblies have disjoint axial meshes. This check should be added.

- This aligns with practice in the original uniform mesh converter. It also makes sense, actually. A users intent in setting b.p.axMesh != 1 is to allow the neutronics mesh to be finer than the geometric/armi mesh in the blueprints. However, the original intended neutronics mesh is lost once it passes through the uniform mesher. Once the uniform mesher changes the axial mesh, the intent of having a b.p.axMesh!= 1 is lost and may (well, likely) not make sense.

- for process load, if the assemblies are not disjoint, there is no reason to run the uniform mesher (it takes a little time and if it's not useful, then don't run it).

- removing this is opening up a rabbit hole

- Only affects _manageCoreMesh(r) which is called from axiallyExpandCoreThermal/Percent. - for ease of use, adding primary/secondaryAssemblyToConserve flags to constructor

- with the removal of makeAxialSnapList from processLoad, it needs to be added here for the two tests in Class AssemblyInReactor_TestCase to pass. - these tests very well may be the focus of another armi/github issue moving forward (TBD).

albeanth · 2022-06-24T22:15:07Z

@onufer, @jakehader, @keckler, @mgjarrett, @john-science

john-science · 2022-06-24T22:41:04Z

@albeanth The tests are all borked. Just FYI.

albeanth · 2022-06-24T22:42:02Z

Yeah... I'm seeing that and don't know why. They pass on my machine....

After running tox locally:

  py38: OK (326.52 seconds)
  lint: OK (490.03=setup[317.39]+cmd[172.64] seconds)
  cov: OK (297.69 seconds)
  congratulations :) (1114.34 seconds)

armi/tests/detailedAxialExpansion/refSmallReactorBase.yaml

john-science · 2022-06-25T21:53:36Z

Yeah... I'm seeing that and don't know why. They pass on my machine....

Two classic problems that can cause this:

You didn't push all your code to GitHub. (check git status to see if this is true)
Try running your tests exactly as GitHub does (python -m tox -e test)

changing so that strings are converted to flags through Flags.fromString

- this should have gone in with the prvious commit.

- I mean, how else are you supposed to fix failing tests? :D

albeanth · 2022-06-27T23:34:46Z

Adding some context for commit 54832a6.

The changes in this PR to reactor.py::Core::processLoading remove the call to makeAxialSnapList, which in turn, avoids a call to b.p.topIndex. So those params are never set and therefore the number of differences in the bookkeeping unit test is reduced.

jakehader · 2022-06-28T00:01:15Z

armi/settings/fwSettings/globalSettings.py

@@ -154,6 +156,28 @@ def defineSettings() -> List[setting.Setting]:
            "block. Used for axial mesh refinement.",
            schema=vol.All(vol.Coerce(int), vol.Range(min=0, min_included=False)),
        ),
+        setting.Setting(
+            CONF_PRIMARY_ASSEMBLY_CONSERVE,
+            default="fuel",


I recommend adding a setter to this setting that transforms the inputs into valid flags so that we arent passing string inputs around.

As I understand it this is exactly why the Flags system was created - to avoid passing around strings - so this totally makes sense.

Where should the setter go? Possibly on reactors.py::Core::setOptionsFromCs? Or is there somewhere else you have in mind?

@mgjarrett

- addresses comment by @mgjarrett terrapower#735 (comment)

albeanth · 2022-06-28T14:53:03Z

armi/reactor/converters/uniformMesh_V2_beta.py

+    for param in sourceBlk.getParamNames():
+        b.p[param] = sourceBlk.p[param]
+    for c in sourceBlk.getChildren():
+        b.add(c.__copy__())


@john-science Just wanted to bring this section to your attention since you have the open issue, #706, on deepcopies.

I made a concerted effort to avoid using deepcopy when creating a copy of an assembly -> block -> component, but gave in with components. I decided to use the Component::__copy__ method with the hope that when #706 is resolved, it too will be resolved. Anyway, just figured I'd get your eyes on this section in particular.

Thanks for keeping it in mind.

My concern is that "fixing" that will break a dozen major plugins, so it won't be possible for months.

Fingers crossed I'm just paranoid. We'll find out!

john-science · 2022-06-28T19:02:24Z

This PR temporarily in a holding cycle while we fix the affects on downstream projects.

Soooooon!

albeanth added 17 commits June 9, 2022 13:09

add axial shield & 1 duct above & below -> control

55c1e84

Merge branch 'main' into detAxExp_TestRx_CRUpdate

3a6e293

initial commit

d4d7ac3

Summary: - removing makeAxialSnapList and replacing with new uniform mesh changer - now enables disjoint assemblies to pass processLoading with detailedAxialExpansion: False

Merge branch 'main' into processLoadUpdate

acb3243

docstring updates + making checkForSmearing a func

c27ed1b

Summary: - checkForSmearing had no self use, so it was made a function rather than a class method

checking if assemblies are axially disjoint

1af6068

- for process load, if the assemblies are not disjoint, there is no reason to run the uniform mesher (it takes a little time and if it's not useful, then don't run it).

put back makeAxialSnapList

1fa5d3e

- removing this is opening up a rabbit hole

adding AssemblyToConserve to fw settings

db6427a

introduce uniformMesh_V2 to axialExpansionChanger

8eed0a8

- Only affects _manageCoreMesh(r) which is called from axiallyExpandCoreThermal/Percent. - for ease of use, adding primary/secondaryAssemblyToConserve flags to constructor

adjust assembly test for new processLoad

9a6626a

- with the removal of makeAxialSnapList from processLoad, it needs to be added here for the two tests in Class AssemblyInReactor_TestCase to pass. - these tests very well may be the focus of another armi/github issue moving forward (TBD).

Merge branch 'main' into processLoadUpdate

c075367

removing unused commented out code

147afbc

black formatting

df5f4b8

albeanth marked this pull request as ready for review June 24, 2022 22:15

mgjarrett reviewed Jun 24, 2022

View reviewed changes

armi/tests/detailedAxialExpansion/refSmallReactorBase.yaml Outdated Show resolved Hide resolved

albeanth added 2 commits June 27, 2022 07:51

Merge branch 'main' into processLoadUpdate

9aca58a

settings need to be strings

f5d5824

changing so that strings are converted to flags through Flags.fromString

onufer requested review from john-science and jakehader June 27, 2022 17:32

albeanth added 3 commits June 27, 2022 14:28

it helps to update dependencies!

35be996

- this should have gone in with the prvious commit.

bug for when difference > 1 on first block

c03ad51

black formatting

9419519

albeanth added 2 commits June 27, 2022 15:49

cleaning up/unifying constructor for uniMesh_V2

16168e4

fix test by changing test @john-science

54832a6

- I mean, how else are you supposed to fix failing tests? :D

jakehader reviewed Jun 28, 2022

View reviewed changes

fix extra space

b8ba5f2

- addresses comment by @mgjarrett terrapower#735 (comment)

albeanth commented Jun 28, 2022

View reviewed changes

making non-public facing methods protected

1d5a422

albeanth added 4 commits June 29, 2022 11:19

Merge branch 'main' into processLoadUpdate

bf95166

Merge branch 'main' into processLoadUpdate

685c157

Merge branch 'main' into processLoadUpdate

8b23e35

Merge branch 'main' into processLoadUpdate

1d9e53c

albeanth marked this pull request as draft July 13, 2022 22:20

albeanth closed this Aug 3, 2022

albeanth deleted the processLoadUpdate branch August 3, 2022 22:50

john-science mentioned this pull request Aug 10, 2022

Allow users to always provide input for non-uniform assemblies in the blueprints #705

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Process load update and new uniform mesher (beta) #735

Process load update and new uniform mesher (beta) #735

albeanth commented Jun 24, 2022 •

edited

Loading

albeanth commented Jun 24, 2022

john-science commented Jun 24, 2022

albeanth commented Jun 24, 2022 •

edited

Loading

john-science commented Jun 25, 2022

albeanth commented Jun 27, 2022

jakehader Jun 28, 2022

albeanth Jun 28, 2022

albeanth Jun 28, 2022

john-science Jun 29, 2022

john-science commented Jun 28, 2022

	.. warning: This procedure can cause numerical diffusion in some cases. For example,
	if a control rod tip block has a large coolant block below it, things like peak
	absorption rate can get lost into it. We recalculate some but not all
	reaction rates in the re-mapping process based on a flux remapping. To avoid this,
	finer meshes will help. Always perform mesh sensitivity studies to ensure appropriate
	convergence for your needs.

	# loop through again now that the reference is adjusted and adjust the non-fuel assemblies.
	refAssem = r.core.refAssem
	axMesh = refAssem.getAxialMesh()
	for a in r.core.getAssemblies(includeBolAssems=True):
	# See ARMI Ticket #112 for explanation of the commented out code
	a.setBlockMesh(
	axMesh
	) # , conserveMassFlag=True, adjustList=adjustList)

Process load update and new uniform mesher (beta) #735

Process load update and new uniform mesher (beta) #735

Conversation

albeanth commented Jun 24, 2022 • edited Loading

Description

Checklist

albeanth commented Jun 24, 2022

john-science commented Jun 24, 2022

albeanth commented Jun 24, 2022 • edited Loading

john-science commented Jun 25, 2022

albeanth commented Jun 27, 2022

jakehader Jun 28, 2022

Choose a reason for hiding this comment

albeanth Jun 28, 2022

Choose a reason for hiding this comment

albeanth Jun 28, 2022

Choose a reason for hiding this comment

john-science Jun 29, 2022

Choose a reason for hiding this comment

john-science commented Jun 28, 2022

albeanth commented Jun 24, 2022 •

edited

Loading

albeanth commented Jun 24, 2022 •

edited

Loading