Datasets:

DeCoDELab
/

FLOATBench

license: cc-by-4.0
size_categories:
  - 100K<n<1M
task_categories:
  - other
pretty_name: 'FLOATBench: Wind Turbine Tower Damage'
tags:
  - floating-offshore-wind-turbine
  - tower-fatigue
  - 22-MW-wind-turbine
  - IEA-22-reference-turbine
  - tabular-dataset
  - benchmark-dataset
configs:
  - config_name: ref
    data_files:
      - split: train
        path: ref/train_damage.csv
      - split: test
        path: ref/test_damage.csv
  - config_name: opt1
    data_files:
      - split: train
        path: opt1/train_damage.csv
      - split: test
        path: opt1/test_damage.csv
  - config_name: opt2
    data_files:
      - split: train
        path: opt2/train_damage.csv
      - split: test
        path: opt2/test_damage.csv

FLOATBench: Wind Turbine Tower Damage

Paper | Project Page | GitHub

Tabular fatigue dataset for 22 MW floating offshore wind turbine (FOWT) towers. Contains 582,120 labelled tower section fatigue damage records across three tower geometries: the IEA-22 reference turbine baseline (ref) and two FLOAT-derived re-designs (opt1, opt2).

FLOATBench is the first FOWT fatigue benchmark for tabular surrogate modeling, offering an evaluation protocol that generalizes to engineering surrogates defined over physical operating envelopes.

Sample Usage

Using the `datasets` library

from datasets import load_dataset

# Load the IEA-22 reference turbine baseline
ds = load_dataset('DeCoDELab/FLOATBench', 'ref')
print(ds)

Using `pandas`

import pandas as pd

# Load from local CSVs (after downloading)
train = pd.read_csv("ref/train_damage.csv")
test  = pd.read_csv("ref/test_damage.csv")

# Evaluate on the worst-case wind+wave extrapolation cell
ex_ex = test[(test.wind_group == "Extrapolate") &
             (test.wave_group == "Extrapolate")]

Layout

FLOATBench/
├── ref/                      IEA-22 reference turbine baseline
│   ├── data.csv              194,040 rows × 16 cols (raw, no split/regime labels)
│   ├── train_damage.csv       51,840 rows × 18 cols (with regime labels)
│   ├── test_damage.csv       142,200 rows × 18 cols (with regime labels)
│   └── metadata.json         counts, split summary
├── opt1/                     FLOAT-derived re-design
│   └── ...                   same files
└── opt2/                     FLOAT-derived re-design
    └── ...                   same files

Schema

Columns appear in the order below. Each *_id grid index sits immediately before the value it indexes (wind_speed_id before wind_speed, wave_hs_id before wave_hs, wave_tp_id before wave_tp).

data.csv (16 cols):

sim_id, wind_speed_id, wind_speed, mean_wind_speed, std_wind_speed,
wave_hs_id, wave_hs, wave_tp_id, wave_tp, wind_seed_id,
section_id, section_height_m, section_radius_m, section_thickness_m,
damage_weight, damage

train_damage.csv / test_damage.csv (18 cols): same order, with wind_group, wave_group inserted right before damage_weight.

The tables below describe each column grouped by category.

Identifiers

Column	Type	Meaning
`sim_id`	int	Unique simulation identifier (ties the 30 sections of one run)
`section_id`	int	Tower section index ∈ {1,...,30}, 1 (base) to 30 (top)
`wind_speed_id`	int	Grid index ∈ {1,...,22}, ordered by `wind_speed` ascending
`wave_hs_id`	int	Grid index ∈ {1,...,7} within each `wind_speed`
`wave_tp_id`	int	Grid index ∈ {1,...,7} within each (`wind_speed`, `wave_hs`)
`wind_seed_id`	int	Turbulence seed index ∈ {1,...,6}

Environmental features

Column	Type	Meaning
`wind_speed`	float	Nominal hub-height wind speed (m/s)
`mean_wind_speed`	float	Realised 10-min mean hub-height wind speed (m/s)
`std_wind_speed`	float	Realised 10-min std of hub-height wind speed (m/s)
`wave_hs`	float	Significant wave height (m)
`wave_tp`	float	Wave peak period (s)

Tower section geometry

Column	Type	Meaning
`section_height_m`	float	Tower section midpoint height along tower axis (m)
`section_radius_m`	float	Tower section outer radius (m)
`section_thickness_m`	float	Tower section wall thickness (m)

Regime labels (only in train_damage.csv and test_damage.csv)

Column	Type	Meaning
`wind_group`	str	`In-train` / `Interpolate` / `Extrapolate` (all train rows are `In-train`)
`wave_group`	str	`In-train` / `Interpolate` / `Extrapolate` (all train rows are `In-train`)

Damage targets

Column	Type	Meaning
`damage`	float	Miner-summed fatigue damage at the section (dimensionless)
`damage_weight`	float	Probability of occurrence over the 25-year service life

Lifetime damage at a section is recovered as sum(damage_i * damage_weight_i) over all conditions.

Regime-aware split

The recommended train/test partition is regime-aware: an alpha-shape over the joint wind/wave operating envelope partitions test points into In-train / Interpolate / Extrapolate regimes on both the wind and wave axes, populating all nine cells of the 3×3 wind×wave regime grid. Per tower:

Subset	Rows	Conditions	Description
Train	51,840	288	All `In-train`/`In-train` cell
Test	142,200	790	Spans the remaining 8 wind×wave regime cells

Reproducing the split from grid IDs

The partition is fully determined by the integer grid IDs (wind_speed_id, wave_hs_id, wave_tp_id) shipped on every row. A row is in train iff its three IDs all fall in the train sets:

TRAIN_WS_IDS = {2, 3, 4, 5, 6, 7, 9, 10, 11, 12, 13, 14,
                16, 17, 18, 19, 20, 21}          # 18 of 22
TRAIN_HS_IDS = {2, 3, 5, 6}                      # 4 of 7
TRAIN_TP_IDS = {2, 3, 5, 6}                      # 4 of 7

is_train = (df.wind_speed_id.isin(TRAIN_WS_IDS)
            & df.wave_hs_id.isin(TRAIN_HS_IDS)
            & df.wave_tp_id.isin(TRAIN_TP_IDS))

Citation

@misc{ribeiro2026floatbenchdatasetbenchmarkfloating,
      title={FLOATBench: A Dataset and Benchmark for Floating Offshore Wind Turbine Tower Fatigue},
      author={João Alves Ribeiro and Bruno Alves Ribeiro and Francisco Pimenta and Sérgio M. O. Tavares and Faez Ahmed},
      year={2026},
      eprint={2605.25717},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2605.25717},
}

License

Released under CC-BY-4.0.