Process Parameters

Each row in process_parameters.csv describes one simulation. The index column is the same number as the HDF5 filename inside the zips, so the CSV is the single point of entry for filtering before any HDF5 is opened.

Columns

Column	Description
index	Unique simulation identifier; matches the HDF5 filename (e.g. index=16039 -> 16039.h5).
geometry	Base part geometry. One of 'rectangular', 'concave', 'convex'.
curvature_radius	Geometry-specific curvature / corner radius, in mm.
bottom_radius	Bottom radius of the formed cup, in mm.
wall_angle	Wall angle of the formed cup, in degrees.
material_scaling_factor	Scaling factor on the base DP600 flow curve (1.0 = nominal). Dimensionless.
sheet_metal_thickness	Initial blank sheet thickness (SHTK), in mm.
friction_coefficient	Coulomb friction coefficient between blank and tools (FC). Dimensionless.
blankholder_force	Force applied by the blank holder (BF), in Newtons.
split	Recommended ML data split: 'train' (~80%), 'test' (~10%), 'val' (~10%); deterministic, seed=42.
rddac	True if the simulation belongs to the RDDAC sub-study (otherwise part of the corner blocks).

Geometry

The geometry column takes three values, one per family of corner shapes. The curvature_radius column samples two extremes per family.

geometry	Description	curvature_radius
rectangular	Rectangular corners	30 / 40 mm
concave	Inward corners	50 / 150 mm
convex	Outward corners	100 / 150 mm

bottom_radius is sampled at 5 / 7.5 / 10 mm and wall_angle at 10 / 20 / 30 deg across all three families.

The three dimensional parameters are defined on the punch as shown below: bottom_radius rounds the floor-to-wall transition, wall_angle is measured between the wall and the vertical axis, and curvature_radius controls the side / corner arcs that distinguish the geometry families.

Each family is rendered below as a formed blank at the end of each operation; the three panels per row share a colour scale of vertical height (Z, in mm) so the depth of the cup is directly comparable across families.

OP10 — after springback

OP20 — after cutting

Split

The split column carries the recommended train / val / test partition. The partition was chosen so that each split sees every geometry family and a representative range of material and process parameters.

RDDAC membership

The boolean rddac column flags whether a simulation belongs to the RDDAC sub study. 396 of the 32,466 simulations are flagged. DDACSDataset(where=lambda row: row["rddac"]) streams only the sub study.

Sample

index,geometry,curvature_radius,bottom_radius,wall_angle,material_scaling_factor,sheet_metal_thickness,friction_coefficient,blankholder_force,split,rddac
16039,rectangular,30,5,10,0.9,0.95,0.05,100000.0,train,False
16040,rectangular,30,5,10,0.9,0.95,0.06,100000.0,train,False

Filtering recipe

import pandas as pd

df = pd.read_csv("./data/process_parameters.csv")

rectangular  = df[df["geometry"] == "rectangular"]
thin_sheet   = df[df["sheet_metal_thickness"] < 0.97]
rddac_only   = df[df["rddac"]]

The same predicate is accepted by DDACSDataset(where=...), which keeps IO scaled to the surviving rows rather than the full 32,466.