OpenMC PWR pin-cell depletion datasets for machine-learning surrogates

Two OpenMC fuel-depletion datasets for training and evaluating machine-learning surrogates of PWR fuel depletion.

casl_3305_runs_inter.h5 (518 MiB) holds 3,305 independent pin-cell depletions, each of them 100 steps of 10 days. Every run draws a fresh operating state at every step, sampled uniformly: specific power 0 to 41.4 W/g, fuel temperature 600 to 1200 K, moderator temperature 550 to 600 K, cladding temperature 570 to 670 K, moderator density 0.74 to 1.00 g/cm3, and boron 0 to 1000 ppm. What you get is a spread of operating histories rather than a single trajectory.

beavrs_cycle1_daily.h5 (8.1 MiB) is one quarter pin-cell depletion following the measured BEAVRS Cycle 1 power history, 575 days at 1-day steps. Nothing is sampled here, and power is the only quantity that varies. The history is adapted from the MIT CRPG BEAVRS benchmark.

In both models the boundaries are reflective so they both represent an infinite lattice of PWR fuel pins.

Everything was generated with OpenMC v0.15.2 and ENDF/B-VII.1 cross sections. The cross sections are not included, since they are a separate download of roughly 7 GB from openmc.org. The data itself is plain HDF5 and reads with h5py alone.

Also included: generation_inputs.zip, which holds the generator configs, both depletion chains and the power history, and PROVENANCE.json, which repeats the same parameters in machine-readable form with SHA-256 checksums.

Please read README.md in this record before using the data. It covers the column reference, the NaN-padded final row of every run, how run boundaries are found, and the known limitations. The main one worth flagging up front is that the generator seed was not recorded, so a regeneration will be statistically equivalent but not identical.

Data: CC-BY-4.0. Code: MIT,