{"success":true,"data":{"author":"Ioan-Daniel Craciun, Technical University of Munich","bucketAccessType":"public","bucketName":"ioandanielc/sph_dataset","bucketPrefix":"","category":"Physics","cloudProvider":"huggingface","createdAt":"2026-11-18T00:06:05.882Z","description":"Single-track laser powder bed fusion (LPBF) melt-pool simulations of Ti-6Al-4V in argon, produced with the LAMAS smoothed-particle-hydrodynamics (SPH) solver by the Chair for Physics-Enhanced Machine Learning at the Technical University of Munich. 241 simulations sample a four-dimensional process window i.i.d.: laser power 77 to 250 W, scan speed 0.2 to 1.0 m/s, laser spot radius 45 to 90 micrometers and substrate temperature 300 to 400 K, spanning conduction, transition and keyhole regimes. The data hold 65,472 frames rendered from three perspectives (196,416 images; top and front 256 x 256, side 512 x 256) with per-pixel phase labels (melt, solid, gas) and per-frame regime labels (initial emptiness, forming, conduction, keyhole) verified by two-round human review. Each experiments/<hash>/ folder has images/{top,side,front}, monitoring .dat scalar time series, labels.csv and metadata.json; index.json maps hashes to parameters. Licensed CC BY 4.0 as stated in the card; maintained by Ioan-Daniel Craciun (TUM).","displayTitle":"SPH-Simulated Laser Powder Bed Fusion Melt-Pool Dataset (Ti-6Al-4V, 241 Single-Track Simulations)","huggingFaceUrl":"https://huggingface.co/datasets/ioandanielc/sph_dataset","id":"e8bb8f22-8169-5df4-83cd-9cbc3f742d8a","license":"CC-BY-4.0","licenseEvidenceUrl":"https://huggingface.co/datasets/ioandanielc/sph_dataset","objectCountEstimate":100000,"probeStatus":"ok","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"ioandanielc/sph_dataset\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"ioandanielc/sph_dataset\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=ioandanielc, license:cc-by-4.0, size_categories:100K<n<1M, downloads=3641"],"format":"mixed","generatedAt":"2026-11-18T00:09:59.116Z","layout":{"example":"final-labels_all.csv","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"name,hash,P,VX,LS,ST,bug_free,correctly_finished,timestep,label_1,label_2,label_final\r\nP-101p127472137_VX-0p52394730996_LS-4p5e-05_ST-300_M-TI64_XI--0p0002_XF-0p0012_XL-0p0014_TE-0p00125_DT-3p56864695376e-06_H-b0e5aed98c,b0e5aed98c,101.127472137,0.52394730996,4.5e-05,300,1,1,0,Initial Emptiness,Initial Emptiness,Initial Emptiness\r\nP-101p127472137_VX-0p52394730996_LS-4p5e-05_ST-300_M-TI64_XI--0p0002_XF-0p0012_XL-0p0014_TE-0p00125_DT-3p56864695376e-06_H-b0e5aed98c,b0e5aed98c,101.127472137,0.52394730996,4.5e-0","encoding":"utf8","key":"final-labels_all.csv","truncatedAt":512}],"sampleObjects":["build_master_csvs.py","croissant_rai_sph_dataset.json","examples/example_classification.py","examples/example_generation.py","examples/example_regression.py"],"sidecars":[{"bytes":"---\nlicense: cc-by-4.0\nlanguage:\n  - en\npretty_name: SPH-Simulated LPBF Melt-Pool Dataset\nsize_categories:\n  - 100K<n<1M\ntask_categories:\n  - image-classification\n  - image-segmentation\ntags:\n  - additive-manufacturing\n  - laser-powder-bed-fusion\n  - smoothed-particle-hydrodynamics\n  - melt-pool\n  - keyhole\n  - physics-simulation\n  - scientific-machine-learning\n---\n\n# SPH-Simulated LPBF Melt-Pool Dataset\n\nSingle-track laser powder bed fusion (LPBF) melt-pool simulations for Ti-6Al-4V,\nproduced with the LAMAS smoothed-particle-hydrodynamics solver. 241 simulations\nsampled uniformly i.i.d. over a 4D process-parameter cube (laser power, scan\nspeed, laser spot radius, substrate temperature), spanning conduction, transition,\nand keyhole regimes.\n\nCompanion to the NeurIPS 2026 Evaluations & Datasets Track submission\n*A Simulation-Based Dataset for Melt Pool Dynamics in Single-Track Laser Powder\nBed Fusion*.\n\nSee examples folder in order to reproduce the results of the toy models.\n\nMaintained by Ioan-Daniel Crăciun, Chair for Physics-Enhanced Machine Learning,\nTechnical University of Munich.\n\n## Stats\n\n| | |\n|---|---|\n| Simulations | 241 |\n| Frames | 65,472 |\n| Images (3 perspectives) | 196,416 |\n| Top / front resolution | 256 × 256 px |\n| Side resolution | 512 × 256 px |\n| Material | Ti-6Al-4V |\n| Atmosphere | Argon |\n\n**Regime label distribution:** Initial Emptiness 5.0% / Forming 15.8% / Conduction 67.0% / Keyhole 0.8%.\nThe keyhole share is a lower bound — manual relabeling is ongoing.\n\n## Labels\n\nTwo label types per frame:\n\n- **Per-pixel phase** (image RGB): red = melt, green = solid substrate, blue = gas.\n- **Per-frame regime** (4 classes): Initial Emptiness, Forming, Conduction, Keyhole.\n\nRegime labels are seeded by a depth-based heuristic and verified through a\ntwo-round human review with majority-vote consolidation.\n\n## Process-parameter ranges\n\n| Parameter | Symbol | Min | Max | Unit |\n|---|---|---|---|---|\n| Laser power | P | 76.70 | 249.99 | W |\n| Scan speed | v | 0.204 | 0.998 | m/s |\n| Laser spot radius | r_s | 45.10 | 89.70 | µm |\n| Substrate temp. | T_s | 300.3 | 399.6 | K |\n\nFixed: particle spacing 4 µm, track length 1200 µm, end time 2.1 ms,\nlaser absorptance 0.2, wall temperature 300 K.\n\n## Layout\n\n```\nexperiments/<hash>/\n├── images/{top,side,front}/   # PNGs per timestep\n├── monitoring/*.dat            # scalar time series\n├── labels.csv                  # per-frame regime labels\n└── metadata.json               # process parameters\nindex.json                      # hash → parameters\n```\n\n## Usage\n\n```python\nfrom huggingface_hub import snapshot_download\npath = snapshot_download(repo_id=\"ioandanielc/sph_dataset\", repo_type=\"dataset\")\n```\n\n## Limitations\n\nSingle-track on bare substrate (no powder bed, no multi-track). Fixed laser\nabsorptance. 2D cross-sections, not full 3D volumes. Keyhole class sparse by\ndesign — the dataset is the Stage-1 initialization for an active-learning loop.\nNot yet validated against experimental measurements.\n\n## Citation\n\n\n## License\n\nCC BY 4.0.\n\n## Contact\n\nIoan-Daniel Crăciun · TUM · <email-or-HF-handle>\n","encoding":"utf8","key":"README.md","name":"README.md","size":3138,"truncated":false},{"bytes":"# Running the Examples\n\n**You do not need to download the full dataset.** Each script has a cap (`N_SUBSET` or\n`MAX_IMAGES`) that limits how many simulations or images are loaded. Download any small\nsubset of simulation directories and point `DATA_DIRS` at them — the scripts will use only\nwhat is available.\n\n## Requirements\n\n```bash\npip install numpy pillow scikit-learn matplotlib torch\n```\n\n## Expected data layout\n\nEach simulation directory must contain:\n\n```\n<sim_dir>/\n    parameters.json                        # process parameters\n    monitor/position-bounds_melt.dat       # melt-pool bounds (regression + classification)\n    frames.csv                             # frame labels (generation only)\n    frames/side/<frame>.png                # side-profile images (generation only)\n```\n\nPoint `DATA_DIRS` in each script at one or more parent directories that hold these\nsimulation folders.\n\n## Regression — Melt-Pool Width Prediction\n\n```bash\npython example_regression.py\n```\n\n- **Reviewer shortcut:** set `N_SUBSET = 30` (already the default) to use only the first\n  30 simulations found — ~1 min on CPU. Set `None` to use all.\n- Predicts steady-state melt-pool width from four process parameters.\n- Results (parity plots + feature importance) are saved to `runs/regression_<timestamp>/`.\n\n## Classification — Keyhole vs Conduction\n\n```bash\npython example_classification.py\n```\n\n- **Reviewer shortcut:** set `N_SUBSET = 30` (already the default) to use only the first\n  30 simulations found — ~1 min on CPU. Set `None` to use all.\n- Labels are derived automatically from simulation output — no manual annotation needed.\n- Results (confusion matrices + feature importance) are saved to `runs/classification_<timestamp>/`.\n\n## Generation — Unconditional VAE on Side Profiles\n\n```bash\npython example_generation.py\n```\n\n- **Reviewer shortcut:** set `MAX_IMAGES = 500` (already the default) to cap the number of\n  frames loaded — ~2 min on CPU. Set `None` to use all images.\n- Only frames labeled Keyhole or Conduction are used; Initial Emptiness and Forming Phase are excluded.\n- Results (sample grids, reconstructions, loss curve) are saved to `runs/generation_<timestamp>/`.\n","encoding":"utf8","key":"examples/README.md","name":"README.md","size":2199,"truncated":false},{"bytes":"{\n  \"sim_id\": 1,\n  \"original_folder_name\": \"P-101p127472137_VX-0p52394730996_LS-4p5e-05_ST-300_M-TI64_XI--0p0002_XF-0p0012_XL-0p0014_TE-0p00125_DT-3p56864695376e-06_H-b0e5aed98c\",\n  \"original_hash\": \"b0e5aed98c\",\n  \"material\": \"TI64\",\n  \"n_frames\": 352,\n  \"n_skipped_incomplete\": 0\n}","encoding":"utf8","key":"final_data_processed/sim_00001/metadata.json","name":"metadata.json","size":283,"truncated":false},{"bytes":"{\n  \"sim_id\": 2,\n  \"original_folder_name\": \"P-104p101541641_VX-0p916529700054_LS-8p82882842693e-05_ST-394p968412652_M-TI64_XI-0p0002_XF-0p0014_XL-0p0012_TE-0p0021_DT-5p2634634783e-06_H-6014a1f9c3\",\n  \"original_hash\": \"6014a1f9c3\",\n  \"material\": \"TI64\",\n  \"n_frames\": 390,\n  \"n_skipped_incomplete\": 0\n}","encoding":"utf8","key":"final_data_processed/sim_00002/metadata.json","name":"metadata.json","size":301,"truncated":false}],"sourceObjectKey":"final-labels_all.csv","starterCell":"import pandas as pd\ndf = pd.read_csv(\"hf://datasets/ioandanielc/sph_dataset/final-labels_all.csv\")\ndf.head()","starterSamplePath":"croissant_rai_sph_dataset.json","totalObjects":100000},"summary":"Single-track laser powder bed fusion (LPBF) melt-pool simulations of Ti-6Al-4V in argon, produced with the LAMAS smoothed-particle-hydrodynamics (SPH) solver by the Chair for…","title":"SPH-Simulated Laser Powder Bed Fusion Melt-Pool Dataset (Ti-6Al-4V, 241 Single-Track Simulations)","updatedAt":"2026-11-18T03:53:31.106Z","visibility":"public","websiteUrl":"https://huggingface.co/datasets/ioandanielc/sph_dataset"}}