lizeyouxiang/SFG-Wide

Dataset Card for Dataset Name

Dataset Card for SFG-Wide and SFG-Local. It has been generated using STAR-CCM+.

Dataset Details

Dataset Description

This dataset represents unsteady flooding flow fields in a ship compartment. The data is generated using CFD simulations under specific boundary conditions.

Each sample consists of:

• Spatial coordinates (x, y, t)
• Velocity components (u, v, p)

The dataset captures complex flow phenomena such as:

• Free-surface evolution
• Secondary flow structures
• Multi-scale turbulence interactions

It is suitable for tasks such as:

• Flow field prediction

• Physics-informed learning

• Spatiotemporal modeling

• Curated by: Ze Li, Harbin Engineering University

• Funded by [optional]: none

• Shared by [optional]: Ze Li

• Language(s) (NLP): Not applicable

• License: CC BY-NC 4.0

Dataset Sources [optional]

• Repository: [More Information Needed]
• Paper [optional]: Input-Decoupled Dual-Scale Attention Network for Flow Field Prediction in Complex Flows,2025

Uses

Direct Use

This dataset can be used for:

• Training neural networks for flow field prediction
• Physics-informed neural network (PINN) research
• Transformer-based spatiotemporal modeling
• Benchmarking data-driven CFD surrogate models

Out-of-Scope Use

This dataset is NOT suitable for:

• Real-time control applications without validation
• General fluid dynamics problems outside flooding scenarios
• High Reynolds number turbulence modeling without adaptation

Dataset Structure

Each data sample represents a spatial point in the flooding flow field and consists of:

• t: flooding time
• x: horizontal coordinate
• y: vertical coordinate
• u: velocity component in x-direction
• v: velocity component in y-direction
• p: pressure

Data Shape

• Each snapshot: (409600, 6)
• Grid resolution: 32 × 64

Data Organization

• The dataset is structured as multiple flow field snapshots over time
• Each snapshot is stored independently

Data Splits

• Training set: 80%
• Validation set: 10%
• Test set: 10%

Spatial Structure

• The domain is discretized into structured grids
• Vertical partitioning is applied for region-based modeling

Dataset Creation

Curation Rationale

This dataset is created to support research on data-driven modeling of ship compartment flooding.

Existing datasets mainly focus on general fluid dynamics problems and lack:

• Free-surface flooding scenarios
• Strongly nonlinear transient flow data
• Multi-scale flow structures in confined spaces

To address these limitations, this dataset provides high-resolution flooding flow field data under physically realistic boundary conditions.

Source Data

Data Collection and Processing

The data is generated using CFD simulations with structured grids.

Simulation Setup

• Governing equations: incompressible Navier–Stokes equations
• Boundary conditions: no-slip walls and inflow/outflow conditions

Processing Steps

• Extract spatial coordinates (x, y, t) and velocity components (u, v, p)
• Normalize velocity values to improve training stability
• Reshape data into structured grid format (32 × 64)

Domain Partitioning

• The spatial domain is divided into vertical regions
• Regions with similar flow characteristics are grouped to enhance model learning efficiency

Who are the source data producers?

The data is generated by numerical simulation systems developed by the authors.

All simulations are conducted by researchers at Harbin Engineering University.

Annotations [optional]

Annotation process

No manual annotation is involved.

All labels are automatically generated from CFD simulations based on physical governing equations.

Who are the annotators?

All annotations are automatically generated from CFD simulations.

No human annotators are involved. The velocity components (u, v) are directly obtained by solving the governing physical equations.

Personal and Sensitive Information

This dataset does not contain any personal, sensitive, or private information.

All data is generated from numerical simulations of physical processes and does not involve human subjects or identifiable information.

Bias, Risks, and Limitations

This dataset has several limitations:

• Simulation-based bias: The data is generated from CFD simulations and may not fully capture real-world uncertainties.
• Scenario limitation: The dataset focuses on ship compartment flooding and may not generalize to other fluid dynamics problems.
• Resolution constraints: The spatial resolution (32 × 64) may limit the representation of fine-scale turbulent structures.
• Boundary condition dependency: Model performance may be sensitive to the specific boundary conditions used in simulation.

Recommendations

Users should consider the following when using this dataset:

• Validate models on real-world or higher-fidelity simulation data when possible
• Avoid directly applying trained models to significantly different flow scenarios
• Consider incorporating additional physics constraints for improved generalization
• Use appropriate normalization and scaling strategies consistent with the dataset

Citation [optional]

APA:

Ze, Li, Yang Dongmei, and Yin Guisheng. "Input-Decoupled Dual-Scale Attention Network for Flow Field Prediction in Complex Flows." Expert Systems with Applications (2025): 130904.

Glossary [optional]

• x, y: Spatial coordinates
• u, v: Velocity components in x and y directions
• CFD: Computational Fluid Dynamics
• PINN: Physics-Informed Neural Network
• Free-surface flow: Flow with a deformable fluid interface