ChannelFlow-Tools: Open-Source Pipeline for Machine-Learning Datasets of 3D Obstructed Channel Flows
ChannelFlow-Tools has been unveiled by researchers as an open-source, configuration-based pipeline designed to create machine-learning-ready datasets from simulated three-dimensional obstructed channel flows, detailed in version 2 of arXiv preprint 2509.15236. This innovative tool overcomes the constraints of static datasets by facilitating the regeneration and customization of data to meet specific research requirements. The pipeline encompasses processes for generating obstacle geometries, transforming them into signed-distance-field voxelizations, executing Lattice-Boltzmann simulations, and organizing outputs into machine-learning tensors. Configuration files guarantee reproducibility with verified byte-identical results. This release offers the computational-fluid-dynamics community a versatile alternative to fixed datasets, enhancing adaptability for surrogate-modeling tasks. The arXiv entry is categorized under announcement type replace-cross.
Key facts
- ChannelFlow-Tools is an open-source, configuration-driven pipeline.
- It is described in arXiv preprint 2509.15236v2.
- The pipeline generates machine-learning-ready data for 3D obstructed channel flows.
- It integrates procedural obstacle geometry across six shape families.
- The workflow includes SDF voxelisation, lattice-Boltzmann simulation, and tensor packaging.
- Byte-identical reproducibility was verified for the geometry-generation stage.
- Evaluation includes a full-corpus mesh-integrity audit.
- The tool addresses the limitations of fixed, pre-generated datasets in surrogate modeling.
Entities
Institutions
- arXiv