Added NaNReporter and HighMaReporter to LettuceCFD

[MaxBille]

Description

Closes #174
Closes #272
Adding two reporters, that check for signs of simulation crashes (NaN in f or Ma > 0.3 in u) and abort simulation with respective abort messages, optional final vtk-frame and (in case of HighMaReporter) optional output file with locations of the 100 highest velocity magnitudes for crash-troubleshooting and analysis.

NaNReporter:

• checks f in interval steps for any NaN value. If NaN is found, simulation is aborted.
• if outdir is provided, lists indices q,x,y,[z] for all found NaN values (attention: produces large files for large intervals, because NaN replicates through the domain fast.
• "old" argument (True, False) is for backwards compatibility with lettuce that has no aborting functionality in the simulation.py class. No vtk output in this case...
• vtk argument (True, False) allows for vtk output of the current single flow state frame to outdir (if vtk_dir is not specified) or vtk_dir (if specified)

HighMaReporter:

• calculates mach number from norm(u(f)) every interval steps and checks if Ma > 0.3 occurs. If Ma > 0.3 is found, simulation is aborted. The location (in LU coordinates) of up to 100 highest Ma numbers in the domain are reported to a file in outdir, if outdir argument is provided.
• "old" argument (True, False) is for backwards compatibility with lettuce that has no aborting functionality in the simulation.py class. No vtk output in this case...
• vtk argument (True, False) allows for vtk output of the current single flow state frame to outdir (if vtk_dir is not specified) or vtk_dir (if specified)
• relies on unravel_index() method for torch.tensors, defined in the same file, for sorting and selection of high-Ma-locations.

DISCLAIMER: originally written for a modified version of lettuceCFD, based on Version 0.2.3; Adaptation for new Lettuce Master pending...

Checklist

• This pull request is associated to an issue
• This PR contains a description
• Add someone else as reviewer and wait for approval before merging.

[pep8speaks]

Hello @MaxBille! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

• In the file tests/reporter/test_nan_reporter.py:

Line 7:29: E128 continuation line under-indented for visual indent
Line 14:80: E501 line too long (111 > 79 characters)

Comment last updated at 2024-09-05 09:53:25 UTC

[PhiSpel]

1. Could you check if we need to set non_native in BreakableSimulation? If so, can you rename to BreakableSimulationNonNative?
2. Maybe rename example to FailingTGVandObstacle.py?

[MaxBille]

@PhiSpel would you mind, reviewing the changes I've made, regarding your change requests and recommendation? :) (if you have the time...)

[PhiSpel]

Some comments and changes still.
A nice setup!

Please remove lettuce/ext/_reporter/nan_repoter.py (the comments still apply to lettuce/ext/_reporter/failure_reporter.py!)

Generally, please revisit the typing of inputs and outputs of the classes and functions :)

[McBs]

Do the reporters only work with this specific Simulation class (i.e. BreakableSimulation)?
If not, I prefer to initialize this class within the example script (i.e. header).

[McBs]

Hi @MaxBille
Is there an update on this?

[PhiSpel]

@McBs yes, it should be BreakableSimulation since we manipulate simulation.flow.i. If simulation this is not BreakableSimulation, it will continue running because call() contains a blind for-loop (

lettuce/lettuce/_simulation.py

Line 243 in d04e156

for _ in range(num_steps):

) that won't be broken.

[PhiSpel]

Typing is still incomplete

[PhiSpel]

lgtm