EntangledSingularity/in718-lpbf-meltpool-F1

IN718 laser powder-bed fusion melt pools (F1 design)

Public listing, gated files (request access on this page). One row per simulated single laser track in IN718 (EOS M290 class): settings in, melt-pool size, energy terms, cooling and health out. Simulated with laserbeamFoamGPU (a CUDA port of LaserbeamFoam: VOF metal + gas, ray-traced Fresnel absorption, Marangoni flow, recoil, evaporation, resolved powder) on one RTX 5090.

Config	Folder	What
default	F1c2/	campaign 2 of the F1 design: 1 of 100 planned rows so far, updated every 10 finished rows
pilot_F1_r00	pilot_F1_r00/	the frozen pilot (52 rows, stopped 2026-10-22): f_M sampled 1.5-3.0, mixed QC and early-stop rules. Kept for reference; never mix it with campaign 2

Use: pd.read_parquet("F1c2/data/closure_meltpool_F1.parquet").query("status == 'success'"). QC failures (qc_fail) keep their outputs with qc_reasons; failures (fail) keep fail_reason.

Read before use

• Not calibrated (column calibration_status). The fixed model constants are the untuned ones; against NIST AM-Bench 2022 tracks the model gives widths 15-38 % too wide and depths 3-18 % too shallow, and the width error grows with scan speed. A calibration found no constants inside their published ranges that bring the tracks within 10 %.
• f_M changed from the pilot. f_M multiplies the Marangoni coefficient dsigma/dT (physical value 1). Campaign 2 samples 0.5-2.0 on a log scale; the pilot sampled 1.5-3.0.
• Mesh bias of the 4 um cells (against 3 um): slow, deep keyholes about 8-11 % too wide, hot-plate pools about 7 % too narrow; fast and high-power keyholes within 2 %.
• Preheat at the beam's arrival (neighbour-track rows): T_pre_K / gradT_lat_K_m are the imposed initial field; T_pre_arrival_K / gradT_lat_arrival_K_m are what the beam meets after the field has cooled while the beam travels (use these).
• absorptivity is absorbed / (lambda_A x P) (<= 1); absorptivity_nominal is absorbed / P.
• Use phi_achieved, not phi_target, for the powder packing.

Each folder's README.md has the full dataset card (every column, unit and meaning) and the file layout; schema.json lists the columns of the default config.