[Core] Load stepping feature by sunethwarna · Pull Request #13877 · KratosMultiphysics/Kratos

sunethwarna · 2025-10-10T12:51:08Z

📝 Description

** NO BEHAVIOR CHANGE IS THERE **

This PR adds load stepping controlling mechanism to try converging non-linear non-converged solutions by load stepping, and changing the time step. This is designed to work in two ways.

Load stepping control, so the load should be dependent on t variable to this to work.
In dynamic simulations, by reducing delta_t, and increasing mass effect to make the non-converging time steps to converge.

This requires all the entities used in the simulations does not hold any private member variables storing the state such as DISPLACEMENTS or VELOCITY since this requires resetting them. Currently this does not work with following element types:

CrBeamElement3D2N
CrBeamElement2D2N

I will wait until #13432 merged, so that there won't be conflicts from this branch to that.

🆕 Changelog

Added StepController class
Added DefaultStepController to have the existing behavior with no changes at all
Added GeometricStepController following the abaqus implementation of load stepping.

…into core/adaptive-time

AlejandroCornejo · 2025-10-10T15:26:35Z

kratos/python_scripts/analysis_stage.py

+                KratosMultiphysics.Logger.PrintInfo(self.__class__.__name__, f"Did not converge for time = {self.time}.")
+
+            current_step_controller_time = time_begin
+            while not step_controller.IsCompleted(current_step_controller_time, is_converged):


@KratosMultiphysics/technical-committee do we want to over complicate the base AnalysisStage like this? I believe that the feature is useful but...

I agree, this is lots of specialized code, IMO the base AnalysisStage is not the place for this

AlejandroCornejo · 2025-10-10T15:27:42Z

applications/StructuralMechanicsApplication/python_scripts/step_controller.py

+            "convergence_factor"                         : 1.5,
+            "delta_t_init"                               : 1.0,
+            "delta_t_min"                                : 1.0,
+            "delta_t_max"                                : 1.0,


delta_time for example would be clearer. What is divergence factor?

RiccardoRossi

Blocking since the design needs to be reviewed and discussed by @KratosMultiphysics/technical-committee .

@sunethwarna can you schedule this for one of the upcoming meetings?

sunethwarna · 2025-11-30T19:14:27Z

moved all the implementatiosn from core to the StructApp under a new derrived analysis.

loumalouomega · 2025-12-01T08:44:17Z

 Running StructuralMechanicsApplication tests
Traceback (most recent call last):
  File "/__w/Kratos/Kratos/bin/Custom/applications/StructuralMechanicsApplication/tests/test_StructuralMechanicsApplication.py", line 247, in <module>
    from test_step_controller import TestStepControllers as TTestStepControllers
  File "/__w/Kratos/Kratos/bin/Custom/applications/StructuralMechanicsApplication/tests/test_step_controller.py", line 4, in <module>
    from KratosMultiphysics.step_controller import DefaultStepController, GeometricStepController
ModuleNotFoundError: No module named 'KratosMultiphysics.step_controller'

Looks like is missing something

…abaqus_extension

...StructuralMechanicsApplication/python_scripts/structural_mechanics_load_stepping_analysis.py

RiccardoRossi

nice PR. I have some comments (the most important is that the Predict needs to be treated as in other analysis stages), but overall i like the idea

RiccardoRossi · 2026-02-20T13:41:01Z

...StructuralMechanicsApplication/python_scripts/structural_mechanics_load_stepping_analysis.py

+
+            # find the appropriate step controller for the given time frame
+            step_controller = None
+            for interval_utility, step_controller in reversed(list_of_step_controllers):


why don't you combine the step controllers in one? my guess is that it would be cleaner not to have to handle a list of step controllers

The reasoning is, the users wants to use different step controllers in different time ranges. This makes it possible. If you combine it to one, that granularity is gone as far as i understand.

What @RiccardoRossi probably means is to create smth like a CompoundStepController that applies different controllers on different time ranges.

The effect and flexibility is essentially the same, but the complexity is hidden in CompoundStepController instead of AnalysisStage. I think that's a valid point.

By the way, why is the list reversed?

matekelemen · 2026-02-26T12:09:47Z

applications/StructuralMechanicsApplication/python_scripts/step_controller.py

+
+    IsCompleted(current_time: float, is_converged: bool) -> bool
+        Check if the stepping process has been completed.
+    """


Please convert all docstrings to doxygen style.

matekelemen · 2026-02-26T12:11:51Z

applications/StructuralMechanicsApplication/python_scripts/step_controller.py

+        default_parameters = Kratos.Parameters("""{
+            "type": "default_step_controller"
+        }""")


For future compatibility with JSON schemas, it'd be great if you could pull default params to a separate function (smth like GetDefaultParameters).

matekelemen · 2026-02-26T12:12:45Z

applications/StructuralMechanicsApplication/python_scripts/step_controller.py

+    Attributes:
+        __divergence_factor: Factor by which the time step is reduced after a failed attempt.
+        __convergence_factor: Factor by which the time step is increased after successful attempts.
+        __delta_t_init: Initial time step size.
+        __delta_t_min: Minimum allowed time step size.
+        __delta_t_max: Maximum allowed time step size.
+        __max_number_of_sub_steps: Maximum number of sub-steps allowed.
+        __number_of_successful_attempts_for_increment: Number of successful steps required to increase the time step.
+        __number_of_failed_attempts_for_termination: Number of failed steps allowed before termination.
+        __sub_stepping_iteration: Counter for the current sub-step iteration.
+        __success_full_attempts_count: Counter for successful attempts.
+        __failed_attempts_count: Counter for failed attempts.
+        __delta_time: Current time step size.
+        __time_begin: Start time of the stepping process.
+        __time_end: End time of the stepping process.


These are internal variables and must not be in the docs.

matekelemen · 2026-02-26T12:15:52Z

applications/StructuralMechanicsApplication/python_scripts/step_controller.py

+    @abstractmethod
+    def Initialize(self, time_begin: float, time_end: float) -> None:
+        """
+        Initializes the step controller with the specified start and end times for current time step from the solver.
+
+        Args:
+            time_begin (float): The starting time of the current time step from the solver.
+            time_end (float): The ending time of the current time step from the solver.
+
+        Returns:
+            None
+        """
+        pass


Why have a separate Initialize function? I'd rather construct a fresh step controller for every time step that gets initialized on construction and destroyed at the end of the time step.

This would help reduce confusion with the initialize/finalize chaos we have with solvers/processes/elements/schemes/strategies. I don't think we need yet another layer of init/fin ops to keep track of.

matekelemen · 2026-02-26T12:20:04Z

applications/StructuralMechanicsApplication/python_scripts/step_controller.py

+            if self.__success_full_attempts_count > self.__number_of_successful_attempts_for_increment:
+                # now we increment the step
+                self.__delta_time = min(self.__convergence_factor * self.__delta_time, self.__delta_t_max)
+                Kratos.Logger.PrintInfo(self.__class__.__name__, f"Incrementing the delta time to {self.__delta_time}")


Please hide printing status info behind a verbosity check. We don't always need to pollute our output with this.

matekelemen · 2026-02-26T12:25:23Z

...StructuralMechanicsApplication/python_scripts/structural_mechanics_load_stepping_analysis.py

+
+            # find the appropriate step controller for the given time frame
+            step_controller = None
+            for interval_utility, step_controller in reversed(list_of_step_controllers):


What @RiccardoRossi probably means is to create smth like a CompoundStepController that applies different controllers on different time ranges.

The effect and flexibility is essentially the same, but the complexity is hidden in CompoundStepController instead of AnalysisStage. I think that's a valid point.

matekelemen · 2026-02-26T12:26:18Z

...StructuralMechanicsApplication/python_scripts/structural_mechanics_load_stepping_analysis.py

+
+            # find the appropriate step controller for the given time frame
+            step_controller = None
+            for interval_utility, step_controller in reversed(list_of_step_controllers):


By the way, why is the list reversed?

matekelemen · 2026-02-26T12:28:04Z

...StructuralMechanicsApplication/python_scripts/structural_mechanics_load_stepping_analysis.py

+        def __get_serializer():
+            return Kratos.FileSerializer("serialization", Kratos.SerializerTraceType.SERIALIZER_NO_TRACE, True)


ugh

I'd like to see an in-memory solution to this. Please open a new PR or issue while we still have this in mind.

matekelemen · 2026-02-26T12:29:38Z

...StructuralMechanicsApplication/python_scripts/structural_mechanics_load_stepping_analysis.py

+        parameters = Kratos.Parameters(parameter_file.read())
+
+    model = Kratos.Model()
+    simulation = StructuralMechanicsAnalysis(model, parameters)


I thought you defined the new analysis type for a reason.

…abaqus_extension

matekelemen and others added 15 commits September 2, 2025 15:26

exclude linear solver specialization from untiy builds

d788159

Merge branch 'master' of https://github.com/kratosmultiphysics/kratos …

f80cb78

…into core/adaptive-time

wip

f7af5dd

debug

4cc4a5f

minor

7d5d959

adding step controller

5088c97

analysis stage

4d51a8e

minor

642cdba

minor

c3d4d4a

add docs

8c0d36f

name change

f84d74e

move step controller list to analysis stage

a0c5ba1

add unit tests

ae40c52

update tests

ab19da0

add tests to suite

2c2121b

sunethwarna self-assigned this Oct 10, 2025

sunethwarna requested review from a team as code owners October 10, 2025 12:51

sunethwarna requested review from RiccardoRossi, matekelemen, pooyan-dadvand, roigcarlo and rubenzorrilla October 10, 2025 12:51

sunethwarna added Feature Kratos Core Python labels Oct 10, 2025

revert changes

58976c6

AlejandroCornejo reviewed Oct 10, 2025

View reviewed changes

RiccardoRossi requested changes Oct 10, 2025

View reviewed changes

sunethwarna requested review from AlejandroCornejo, RiccardoRossi and philbucher November 30, 2025 19:13

sunethwarna added 12 commits December 3, 2025 04:27

minor

727e6aa

Merge remote-tracking branch 'origin/master' into core/adaptive_time_…

a5df261

…abaqus_extension

Merge remote-tracking branch 'origin/master' into core/adaptive_time_…

a3b365a

…abaqus_extension

add DataOnly capabilities to serializer

56038d8

add data only to the file serializer

5072790

remove unnecessary allocation

234931f

file serializer modification

8c51270

expose the data only parameter to python in serializer

f7ae46f

add possibility to read submp data in model part with serializer

877f0a2

add a unit test for serializer data only

7edf135

modify the analysis to use serializer for pseudo time step

df193d1

Merge remote-tracking branch 'origin/master' into core/adaptive_time_…

a9b31b0

…abaqus_extension

sunethwarna enabled auto-merge February 10, 2026 23:00

sunethwarna disabled auto-merge February 11, 2026 09:33

sunethwarna commented Feb 18, 2026

View reviewed changes

...StructuralMechanicsApplication/python_scripts/structural_mechanics_load_stepping_analysis.py Outdated Show resolved Hide resolved

RiccardoRossi reviewed Feb 20, 2026

View reviewed changes

...StructuralMechanicsApplication/python_scripts/structural_mechanics_load_stepping_analysis.py Outdated Show resolved Hide resolved

RiccardoRossi reviewed Feb 20, 2026

View reviewed changes

RiccardoRossi moved this from 👀 Next meeting TODO to 🏗 In progress in Technical Commiittee Feb 23, 2026

move predict

7ebec70

sunethwarna enabled auto-merge February 25, 2026 18:04

matekelemen disabled auto-merge February 26, 2026 12:08

matekelemen reviewed Feb 26, 2026

View reviewed changes

sunethwarna added 3 commits March 3, 2026 07:25

Merge remote-tracking branch 'origin/master' into core/adaptive_time_…

f5f0e2c

…abaqus_extension

convert comments

f125553

add pseudo step test

be0f55b

		def __get_serializer():
		return Kratos.FileSerializer("serialization", Kratos.SerializerTraceType.SERIALIZER_NO_TRACE, True)

Conversation

sunethwarna commented Oct 10, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

RiccardoRossi left a comment

Choose a reason for hiding this comment

Uh oh!

sunethwarna commented Nov 30, 2025

Uh oh!

loumalouomega commented Dec 1, 2025

Uh oh!

Uh oh!

Uh oh!

RiccardoRossi left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants