The folder radcluster_code/digital_twin drives the calibration and the verification of the model against measured microstructures. A campaign is a designed set of runs that is distributed over several machines.

Parts

File	Content
design.py	generates the one canonical design of a campaign
run_ensemble.py	the worker that runs the rows assigned to one machine
merge_and_sobol.py	merges the results and estimates the sensitivity indices
check_machine.py	compares a fixed probe run on this machine with the reference
verify.py and the folder verification	the verification pass and the generators of reports, figures and tables

Running on several machines

Git carries the design, the workbook and the code, but not the solver: each machine compiles its own. Before a machine joins a campaign, the script check_machine.py runs a fixed probe and compares its results with the reference values.

Note that a machine that fails this check must not be started, because its rows cannot be pooled with the others.

Each machine writes its own results file, and the rows are appended as they complete. A restart skips the rows that are already present, so that a campaign can be stopped, submitted again and tallied at any time. The campaign runs on the UCLA Hoffman2 cluster as an array job. See Running on the Hoffman2 cluster.

Experimental database

The folder radcluster_code/Eurofer_micro_database contains a spreadsheet of loop and cavity number densities and mean sizes for neutron-irradiated and ion-irradiated ferritic-martensitic steels, together with a notebook that fits and plots them.

See also: Verification and diagnostics