|
Download README.md from abmorsego/tello-openfoam-study-data: direct link, hf CLI and curl.
- Browser
- Download file 5 kB
-
https://huggingface.co/datasets/abmorsego/tello-openfoam-study-data/resolve/main/README.md
- Command line
-
hf download hf://datasets/abmorsego/tello-openfoam-study-data/README.md
-
curl -L -o README.md https://huggingface.co/datasets/abmorsego/tello-openfoam-study-data/resolve/main/README.md
5 kB
| language: | |
| - en | |
| tags: | |
| - computational-fluid-dynamics | |
| - openfoam | |
| - rotor-interaction | |
| - reproducibility | |
| pretty_name: Tello-sized rotor OpenFOAM study — offline evidence | |
| # Tello-sized rotor OpenFOAM study: offline evidence | |
| These are the large reproducibility files kept outside the lightweight [source and results repository](https://github.com/abbiimg/tello-openfoam-study/tree/99a2858a67042320dccf4b08214194249a5ec209). The source commit is `99a2858a67042320dccf4b08214194249a5ec209`. The archives are research artifacts, not a trained machine-learning model or measurements of a physical drone. | |
| | File | Bytes | Contents | | |
| | --- | ---: | --- | | |
| | `setup_results.zip` | 1,200,933,912 | Exact setup and curated evidence snapshot, including the 90 external inputs omitted from the current Git tree. | | |
| | `cases.zip` | 3,286,551,687 | Full archived cases, including generated meshes and saved fields from the 84 completed CPU runs. | | |
| | `external_inputs_manifest.json` | — | Exact paths, byte counts and SHA256 values for the 90 external inputs. | | |
| | `archive_members_sha256.json` | — | Per-member SHA256/size/CRC catalogue of the 3,798-member setup/evidence ZIP. | | |
| | `SHA256SUMS` | — | Checksums for both archives and the two JSON catalogues. | | |
| The 90 external inputs total **2,641,001,310 uncompressed bytes**: 56 mapped warm-start fields, 30 completed CPU/reference validation fields, one canonical starting flux, and three pinned source/seed archives. The ZIP archives overlap intentionally: `setup_results.zip` is the compact exact-input/evidence package; `cases.zip` retains more simulation output. No complete WSL filesystem, home directories, credentials, SDK distributions, compiled build trees, caches or full raw private profiler databases are included. | |
| ## What was simulated | |
| The lift matrix contains **84 completed transient OpenFOAM Foundation 14 CPU cases**; 15 additional prepared cases remain unrun and deferred. The model uses interacting finite-volume actuator sources in a shared flow field, with 7.62 cm propellers, an assumed 80 g aircraft, and approximately 19.2923 W total inferred electrical input for either rotor count. Power is inferred from rated battery energy divided by advertised flight duration and an assumed loss map; it is not a measurement of Tello hover or peak power. | |
| The highest tested qualified fine-grid mean is approximately **1.311138 N**, for eight rotors on the square perimeter at 8 cm nearest hub spacing and outer edges raised 7.5 degrees (thrust directed inward/up). Nearby angles are close, so this supports a best-tested region, not a unique continuous or physical optimum. The clean equal-footprint count benchmark gives about 21.10% more modeled lift for eight rotors at 10 cm nearest spacing than four rotors at 20 cm. That quad control lies outside the 0–10 cm optimization range. | |
| Blades, synchronization phase, collision avoidance, actual motor curves, body/guards and added rotor/frame mass are not resolved. Numerical screens and qualifications are retained in the source reports. These are simulated/model-dependent results, not validated hardware payload ratings. | |
| ## GPU status | |
| The archives include setup recipes and selected real evidence for native stdpar, SPUMA, AmgX and profiler experiments. A real arithmetic positive control produced **eight attributable GPU kernel records**. The native full 922,576-cell rotor case failed with SIGSEGV in both profiled and unprofiled runs and remains **unqualified**. Short AmgX rotor compatibility is not a new lift or speed measurement. CPU remains the study's production default; no general GPU speedup or GPU lift result is claimed. | |
| ## Reproduce | |
| Download `setup_results.zip` and verify its SHA256 against `SHA256SUMS`. In the matching source checkout, use the supplied `tools/restore_external_inputs.py` and `external_inputs_manifest.json` before exact mapped-case reruns or the C29 offline replay. The helper defaults to verification only and restores selected inputs only with `--restore`. The source README gives the tested Linux/OpenFOAM commands and explains the distinction between exact replay and newly generated cold-start experiments. | |
| The setup archive passed every member SHA256 and full ZIP CRC check, preserving the original case records and modes. The guarded restore helper passed 22 inert safety checks and one read-only test against all 90 actual external inputs. These checks do not constitute a fresh-host CFD rebuild or hardware validation. | |
| ## Licenses and provenance | |
| These archives contain original study material and selected third-party source/patches. Existing source headers, license files, commit pins and attribution remain applicable to their respective components, including OpenFOAM's GPL terms and the licenses of other upstream projects. This dataset card does not relicense those components under a blanket permissive or Creative Commons license. Refer to the included source notices and provenance records before redistribution or reuse. | |