Wi-Re/diffSPH16K

Description

This dataset contains a set of example simulations using diffSPH. A dataloader and further information is available as part of diffSPH.

Simulation Setup

The general simulation setup:

• Domain Size: $1[m]\times 1[m]$ ($[-0.5, 0.5]^2$)
• Particle Count: $128 \times 128 = 16384$
• Particle size: $1/ (128\times 128) = 0.000061035[m^2]$
• Neighborhood Size: $45.228$ neighbors per particle
• Rest Density: $1 [kg/m^2]$
• Timestepping: Fixed $\Delta t = 1 [ms]$
• Integration Scheme: Symplectic Euler for WCSPH, RK2 for compressible SPH
• SPH Formulation: $\delta^+$-SPH for WCSPH, CRKSPH for compressible SPH
• Shifting Scheme: $\delta^+$-SPH for WCSPH only
• Boundary Handling Scheme: MDBC
• Viscosity: $\delta^+$-SPH, $\operatorname{Re}=2000$

Data Layout:

Each simulation is stored as a single hdf5 file with meta information available about the simulation configuration and parameters. Each timestep is a seperate group under the simulationData group. For more information on how to process the data, see diffSPH. To load the data using diffSPH you can simply:

import torch
import matplotlib.pyplot as plt
import warnings
from tqdm import TqdmExperimentalWarning
warnings.filterwarnings("ignore", category=TqdmExperimentalWarning)
from tqdm.autonotebook import tqdm
from diffSPH.plotting import visualizeParticles, updatePlot
from diffSPH.operations import sph_op
from diffSPH.kernels import getSPHKernelv2

from diffSPH.dataLoader import *
from BasisConvolution.convLayerv3 import BasisConvLayer

configuration = DataConfiguration(
    frameDistance=1,
    frameSpacing=1,
    maxRollout=1,
    historyLength=1,
    skipInitialFrames=0,
    cutoff=0
)
folder = './data/compressible/circleCase'


processed = processFolder(folder, configuration)

for file in processed:
    print(f'File: {file["fileName"]}, FrameCount: {len(file["frames"])}, Samples: {len(file["samples"])}, Style: {file["style"]}, Number of samples: {len(processed[0]["samples"])}, first sample: {processed[0]["samples"][0]}, last sample: {processed[0]["samples"][-1]}')

dataset, datasetLoader = getDataLoader(processed, 4, shuffle = True)
datasetIter = iter(datasetLoader)
nextData = next(datasetIter)

priorStates, currentState, trajectoryStates, domains, rotMats, configs, neighborhoods = loadAugmentedBatch(
    dataset, nextData, configuration, device = 'cuda', dtype = torch.float32,
    augmentAngle = False,
)

fig, axis = plt.subplots(1, len(nextData), figsize=(len(nextData) * 4, 5), squeeze = False)
plots = []
axes = axis.flatten()
for b in range(len(domains)):
    plot = visualizeParticles(fig, axes[b],
                              particles = currentState,
                              domain = domains[b],
                              quantity = currentState.densities,
                              which = 'both',
                              mapping = 'L2',
                              cmap = 'viridis',
                              visualizeBoth=True,
                              kernel = kernelNameToKernel(configs[b]['kernel']),
                              plotDomain = True,
                              gridVisualization=False, markerSize=2,
                              batch = b, streamLines = False)
    axes[b].set_title(f'Batch {b} {nextData[b]} - t={currentState.time[b]:.2f}s,')
    plots.append(plot)

fig.tight_layout()

diffSPH contains more examples on how to then use this data for training with both CConv networks and GNNs.

Dataset Overview

Compressible SPH

All simulations have 12 initial conditions available to them. For region cases the density and pressure are randomized per region, for the other cases there is a uniform octave noise across the domain to sample the density and pressure field, and the velocity field as divergence free

Circular Regions

Initial	Final

Dual Regions

Initial	Final

Quad Regions

Initial	Final

Density Pressure Noise

Initial	Final

Random Velocity Field

Initial	Final

Weakly Compressible SPH

Each simulaiton has 16 initial conditions available, except for the Taylor Green Vortex like initial conditions which are initialized for 2 and 4 vortices only.

Periodic Domains

BC	Initial	Final
NoSlip
FreeSlip

With an Obstacle

BC	Initial	Final
NoSlip
FreeSlip

Bounded Domain

BC	Initial	Final
NoSlip
FreeSlip

Both

BC	Initial	Final
NoSlip
FreeSlip

Taylor Green Vortex

x	Obstacle ?	Domain ?
-
-