|
Persistent Identifier
|
doi:10.18419/DARUS-5589 |
|
Publication Date
|
2026-07-13 |
|
Title
| Real Deep Drawing and Cutting (RDDAC) Dataset |
|
Subtitle
| Real measurements of a real deep drawing and cutting experiment with multiple modalities |
|
Alternative Title
| RDDAC |
|
Author
| Baum, Sebastianhttps://ror.org/04vnq7t77ORCIDhttps://orcid.org/0009-0001-0604-8764
Heinzelmann, Pascalhttps://ror.org/04vnq7t77ORCIDhttps://orcid.org/0009-0000-0638-7320 |
|
Point of Contact
|
Use email button above to contact.
Baum, Sebastian (University of Stuttgart)
Heinzelmann, Pascal (University of Stuttgart)
IAS (University of Stuttgart)
IFU (University of Stuttgart) |
|
Description
| Code, quick-start examples and full documentation: https://github.com/BaumSebastian/RDDAC.
The Real Deep Drawing and Cutting (RDDAC) Dataset is the experimental counterpart to the DDACS simulation dataset. It contains physical measurements from a deep drawing and cutting process for DP600 sheet metal, captured on modified quadratic cups, and is intended for studying the deviation between finite element simulation and physical reality.
The dataset comprises approximately 9000 experiments spanning two base geometries (concave, convex) crossed with three blankholder forces (100, 300, 500 kN) and three lubrication patterns (coarse, medium, fine), giving 18 categories with up to 500 repetitions each. Each experiment is stored as a HDF5 file and captures, where available: press force and process signals (load cells and temperature); a sheet thickness traverse; an oil film traverse; and 3D laser scans (height and luminescence) of the part after deep drawing (OP10) and after cutting (OP20).
Each HDF5 file carries named scalar root attributes matching the columns of process_parameters.csv (process_parameters.tab on the DaRUS UI), the central index that maps every experiment to its parameters, data availability flags, and a recommended ML train, validation, and test split. A sample.zip (one experiment per category, 18 files in total) is provided for fast preview without downloading the full dataset. A Croissant 1.1 manifest (metadata.json) describes every field for direct use with mlcroissant. The accompanying Python package includes an optional preprocessing step that reconstructs cleaned 3D point clouds and aligns them to the matching DDACS simulations. |
|
Subject
| Computer and Information Science; Engineering |
|
Keyword
| Deep Drawing http://www.wikidata.org/entity/Q1068711 (Wikidata)
Sheet Metal Forming
Machine Learning http://www.wikidata.org/entity/Q2539 (Wikidata)
Point Cloud https://cso.kmi.open.ac.uk/topics/point_cloud (cso)
Experimental Measurement http://edamontology.org/data_3108 (edam)
DP600
Springback http://id.loc.gov/authorities/subjects/sh2013001458 (LCSH)
Sim-to-Real |
|
Topic Classification
| Real2Sim
Computer Science, Systems and Electrical Engineering (DFGFO) https://w3id.org/dfgfo/2024/44 |
|
Related Publication
| Is Cited By: Baum, S., Heinzelmann, P., Clauß, P. et al. Statistical Analysis of Simulation-to-Reality Deviation in Deep Drawing with a Benchmark Dataset. Trans Indian Inst Met 79, 176 (2026). doi 10.1007/s12666-026-03870-5 https://doi.org/10.1007/s12666-026-03870-5 |
|
Language
| English |
|
Production Location
| Stuttgart, Germany |
|
Contributor
| Data Collector: Heinzelmann, Pascal
Data Curator: Baum, Sebastian |
|
Funding Information
| DFG: SPP 2422 - 500936349 |
|
Distributor
| Baum, Sebastian (University of Stuttgart) |
|
Distribution Date
| 2026-07-03 |
|
Depositor
| Baum, Sebastian |
|
Deposit Date
| 2026-06-01 |
|
Date of Collection
| Start Date: 2025-09-01; End Date: 2025-11-20 |
|
Data Type
| Metal Forming Experiments; Point Cloud; Time Series; Categorical; Images |
|
Related Material
| DDACS dataset (the simulation counterpart); RDDAC Python package; DDACS Python package; Croissant 1.1 specification |
|
Related Dataset
| Baum, Sebastian; Heinzelmann Pascal "Deep Drawing and Cutting Simulations (DDACS) Dataset", doi: 10.18419/DARUS-4801 |
|
Did it work?
| Yes |
|
Explanation
| Of the 9000 experiments, 10 lack 3D point cloud scans (2 concave, 8 convex) and 123 lack oil film measurements (105 concave, 18 convex). The has_pointcloud and has_oil columns of process_parameters.csv flag the affected experiments; all other modalities are present for every experiment. |