hpcforge/BubbleML_2

BubbleML 2.0:

BubbleML_2 is a high-fidelity dataset of boiling simulations in 2D for three fluids (FC-72, Liquid N2 and R515B). It provides paired time-series fields stored in HDF5 (.hdf5) files together with metadata (.json) and explicit train/test splits.

🚀 Quickstart

The current available dataset subsets are-

"single-bubble", "pb-saturated", "pb-subcooled", "fb-velscale", "fb-chf"

They are chosen for each individual forecasting task in our paper, viz. Single Bubble, Saturated Pool Boiling, Subcooled Pool Boiling, Flow Boiling- Varying Inlet Velocity and Flow Boiling- Varying Heat Flux.

from datasets import load_dataset

# Load the TRAIN split
ds_train = load_dataset(
    "hpcforge/BubbleML_2",
    name="single-bubble",
    split="train",
    streaming=True,            # to save disk space
    trust_remote_code=True,    # required to run the custom dataset script
)

# Load the TEST split
ds_test = load_dataset(
    "hpcforge/BubbleML_2",
    name="single-bubble",
    split="test",
    streaming=True,
    trust_remote_code=True,
)

Each example in dstrain / dstest has the following fields:

• —input NumPy array of shape (time_window=5, fields=4, HEIGHT, WIDTH)
• —output NumPy array of shape (time_window=5, fields=4, HEIGHT, WIDTH)
• —fluid_params List of 9 floats representing: Inverse Reynolds Number, Non-dimensionalized Specific Heat, Non-dimensionalized Viscosity, Non-dimensionalized Density, Non-dimensionalized Thermal Conductivity, Stefan Number, Prandtl Number, Nucleation wait time and the Heater temperature.

    [inv_reynolds, cpgas, mugas, rhogas, thcogas, stefan, prandtl, heater.nucWaitTime, heater.wallTemp]

• —filename HDF5 filename (e.g. Twall_90.hdf5)