PINNeAPPle-Labs/thermal-channel-cfd

Actuated heated-channel CFD, 2-D incompressible flow + heat (82 trajectories, 128 x 32, 101 snapshots)

Non-dimensional channel [0, 10] x [0, 1], Re 100, Pr 0.71: parabolic inlet with time-varying mean velocity U(t) in [0.6, 1.4] and inlet temperature 0, no-slip walls, adiabatic except a heater on the bottom wall at x in [2, 4] with flux -dT/dy = Q(t) in [0, 2], zero-gradient outlet. After a 15-unit warm-up each trajectory records u, v, p, T on the 128 x 32 cell centres every 0.4 time units over 40 units (101 snapshots), with the actuators and the quantities of interest (bulk outlet temperature, maximum temperature, pressure drop, heat in / out, stored energy). Training actuation is random holds, steps and ramps; two test splits use families never seen in training (chirped sinusoids, steps between the range edges), for digital-twin and forecasting models. Cell centres and all solver settings are in meta.json; fields[:, k] is the state at t = 0.4 k.

thermal_channel_cfd v1.0.0 (revision v1.0.0.post1) · licence cc-by-4.0 · part of the PINNeAPPle zoo

Configs and splits

Config	Split	Samples	Parquet rows	Parquet files	Content
default	train	48	4848	2	48 trajectories (random steps and ramps)
default	val	8	808	1	8 trajectories (same family)
default	test_same_family	12	1212	1	12 trajectories (same family, held-out seeds)
default	test_unseen_sinusoid	8	808	1	8 trajectories (chirped sinusoids, unseen family)
default	test_unseen_edge_steps	6	606	1	6 trajectories (steps between range edges, unseen family)

Split rule: fixed seeds per split (train 0-47, val 1000-1007, test_same_family 2000-2011, test_unseen_sinusoid 3000-3007, test_unseen_edge_steps 4000-4005)

Columns

Column	Unit	Type	Description
sample	-	int64	sample: the row of the split in pp.datasets.load this step belongs to
step	-	int32	step: index along the first axis of the stored arrays
Q	-	float32 (one step)	heater flux at the record times
Q_fine	-	float32[401]	heater flux every 0.1 time units (401 values)
U	-	float32 (one step)	inlet mean velocity at the record times
U_fine	-	float32[401]	inlet mean velocity every 0.1 time units (401 values)
family	-	string	actuation family
fields	-	float32[4, 128, 32] (one step)	u, v, p, T on the 128 x 32 cell centres at the 101 record times (trajectory x time x 4 x 128 x 32)
qoi_T_max	-	float32 (one step)	maximum temperature, heater wall included
qoi_T_out	-	float32 (one step)	bulk (flow-weighted) outlet temperature
qoi_dp	-	float32 (one step)	inlet-to-outlet pressure drop
qoi_energy	-	float32 (one step)	thermal energy stored in the channel
qoi_heat_in	-	float32 (one step)	heat input rate of the heater
qoi_heat_out	-	float32 (one step)	advected heat outflow rate at the outlet
seed	-	int64	actuation seed

Load

With the library (the npz files, SHA-256 checked; arrays with their full shape):

pip install "pinneapple @ git+https://github.com/PINNeAPPle-Labs/PINNeAPPle"

import pinneapple as pp

d = pp.datasets.load("thermal_channel_cfd", split="train")
print(d.n_rows, d.columns)
print(d["Q_fine"].shape)

With datasets (the Parquet files of this repo):

from datasets import load_dataset

ds = load_dataset("PINNeAPPle-Labs/thermal-channel-cfd", split="train", revision="v1.0.0.post1")
print(ds)
ds = ds.with_format("numpy")
print(ds[0]["Q_fine"].shape)

Each sample of this dataset is larger than the viewer can show in one row, so the Parquet files store one row per step of its first axis (101 steps): sample is the row of the split in pp.datasets.load and step the index along that axis; columns Q, U, fields, qoi_T_max, qoi_T_out, qoi_dp, qoi_energy, qoi_heat_in, qoi_heat_out hold one step, the others are repeated on every step. pp.datasets.load returns the samples whole; pinneapple_hub.zoo.parquet.read_parquet puts the Parquet rows back together the same way.

Provenance

python scripts/zoo/thermal_channel_data.py generate --out DIR; recipe of PINNeAPPle-Benchmark experiment 05 (Apache-2.0) on pinneapple_simulation.numerical_solvers.thermal_channel_2d (staggered MAC grid, Chorin projection, dt 0.01), seeds per trajectory in pinneapple_simulation.numerical_solvers.thermal_channel_datasets.PLAN

Verification

Solver checks (python scripts/zoo/thermal_channel_data.py verify, pinneapple_simulation.numerical_solvers.thermal_channel_datasets.verify): plane Poiseuille flow, max error of the velocity profile 4.8e-3 / 1.3e-3 / 3.5e-4 on 64 x 16 / 128 x 32 (the dataset grid) / 256 x 64 (second order), dp/dx within 1.2 % of -12/Re, max divergence 1e-7; fully developed Nusselt number for one wall at uniform flux and one adiabatic, 5.392 / 5.388 vs 5.385 exact (Shah & London 1978), global energy imbalance 2.7e-8. Three-grid study of an actuated run (64 x 16, 128 x 32, 256 x 64, dt halved with h): GCI of the dataset grid 3.7 % for the bulk outlet temperature (oscillatory convergence, observed order 1.3), 1.1 % for the maximum temperature (order 2.3), 1.3 % for the pressure drop (order 1.3). In every trajectory the change of stored energy matches the trapezoidal integral of the recorded heat in - out to 0.3 % (median, smooth actuation) and at most 4 % (steps, limited by the 0.4 sampling of the actuators). All fields finite; float32 storage.

Files

The npz files are the library's format and the source of truth: pp.datasets.load downloads them from this repo at the tagged revision and refuses any file whose SHA-256 differs from the manifest of the package. The Parquet files under data/ are derived from them by pinneapple_hub.zoo.parquet.write_dataset and hold the same numbers (a test round-trips npz -> Parquet -> arrays bit for bit); they are what the dataset viewer and load_dataset read.

File	Size	SHA-256
train-00.npz	53.3 MB	ce79b01b28b87164c111b949c9fd06865aca86944f470558b5730348eeb5212b
train-01.npz	52.9 MB	142de54e3ae9ed912e85996551131a0eaf0a2cac9de4fe2c94d449dddde634ef
train-02.npz	53.1 MB	17e77cbdaa91a6814ed2d6a84c5e68c1b105710ead75c5a466034e45e95e9eac
train-03.npz	53.0 MB	94df94f68e2581bb00ed4ee170c0f982a4b04a3c1ab46b9b880d4edd2c2af49b
val.npz	35.3 MB	6a8d6b9950cafc4f5e88bfdf325bdc31b09d88c65a7cda8bee96588656a0de56
test_same_family.npz	52.6 MB	c23cec4f69980c5307e778e815d2ac31c57ca120cff286329587e98d68cf0d49
test_unseen_sinusoid.npz	36.8 MB	f7e608698732c729454f7abaf94a9998f5ea0160ba10f2f15aec9f93e4faa5d1
test_unseen_edge_steps.npz	23.3 MB	5c9502396710a3c5cad7f16e9bd936c2bd697e79b3bc0acf9a1a277426f79e9c

Total 360.3 MB of npz.

Citation

PINNeAPPle-Benchmark experiment 05 (actuated heated channel). Data generated with PINNeAPPle (pinneapple_simulation.numerical_solvers.thermal_channel_2d); PINNeAPPle, Zenodo DOI 10.5281/zenodo.23343062.

Cite

If you use this dataset, please cite PINNeAPPle (software, Zenodo DOI 10.5281/zenodo.23343062; see CITATION.cff):

@software{pinneapple,
  author = {Barros, Yan and Schiassi, Enrico},
  title  = {PINNeAPPle},
  doi    = {10.5281/zenodo.23343062},
  url    = {https://github.com/PINNeAPPle-Labs/PINNeAPPle}
}