A Validated LBM Dataset and Pipeline for Surrogate Modeling of Turbulent 3D Obstructed Channel Flows
A Validated LBM Dataset and Pipeline for Surrogate Modeling of Turbulent 3D Obstructed Channel Flows
Authors: Lukas Schröder, Shubham Kavane, Harald Köstler Date: 2026-06-15 Paper ID: openalex:2606.16765
Summary
This paper introduces a high-fidelity, validated lattice Boltzmann method (LBM) pipeline for generating training data for 3D turbulent channel flows across a range of Reynolds numbers. By ensuring physical accuracy through rigorous verification against experimental benchmarks like turbulent fluctuations and drag coefficients, the authors provide a standard for testing neural operator performance in fluid dynamics. The research framework is designed to evaluate surrogate models—such as FNOs and U-Nets—on complex tasks including forecasting and super-resolution, specifically measuring their ability to represent turbulent energy cascades.
Key Contributions
- Introduces a reproducible lattice Boltzmann (LBM) data generation pipeline for 3D turbulent channel flows at Reynolds numbers 1,000–10,000.
- Provides rigorous physical validation against experimental benchmarks, including Strouhal numbers, drag coefficients, and turbulent fluctuations.
- Establishes a standardized framework for evaluating neural surrogates (FNO, U-Net) across forecasting, super-resolution, and error-correction tasks in fluid dynamics.
Key Concepts
- lbm-turbulent-flow-pipeline: A verified pipeline for generating high-fidelity lattice Boltzmann simulation data of 3D turbulent channel flows for surrogate modeling evaluation.
Archivist Review
I have approved the LBM pipeline as a concept because it represents a reusable methodological framework for producing validated scientific datasets. I rejected the concept of energy cascade representation because it describes a physical phenomenon rather than a reusable algorithmic contribution or methodology. No specific datasets were approved as the authors did not name the specific generated dataset or repository in a way that suggests a standalone benchmark name beyond the general pipeline description.
Approved Concepts
- Lattice Boltzmann Method Pipeline for Turbulent Flows: Provides a standardized, validated data generation framework essential for training and benchmarking neural operators on complex, high-Reynolds number 3D turbulent flow problems.
Rejected Candidates
- [concept] Turbulent Energy Cascade Representation (
turbulent-energy-cascade-representation) - generic: This is a standard physical phenomenon rather than a specific neural operator mechanism or novel architectural concept.
Links
Metadata & Links
- url
- https://arxiv.org/abs/2606.16765
- paper_id
- 2606.16765
- paper_source
- openalex
- domain
- computer-vision
- tags
-
- architectures
-
- datasets
-
- concept_slugs
- lbm-turbulent-flow-pipeline
- dataset_slugs
-
- skill
- TimeSeriesSkill
- created_at
- 2026-06-18T06:38:10Z