BLASTNet: High-Fidelity Reacting Flow Simulations for ML

BLASTNet 2.0 provides 744 full-domain samples across 34 configurations, totaling 2.2 TB of reacting and non-reacting flow physics simulation data. The dataset was developed to foster open and fair machine learning development within the fluid dynamics community. It is designed for scientific engineering tasks in automotive, propulsion, energy, and environmental applications.

Use Cases

• Turbulent closure modeling based on high-fidelity simulation data
• Spatio-temporal modeling of reacting flow fields
• Inverse modeling for propulsion and energy system design
• Benchmarking machine learning algorithms on large-scale scientific simulation data

Strengths

• Contains 744 full-domain simulation samples, providing substantial data volume for model training
• Comprises 2.2 TB of high-fidelity data, indicating significant detail and resolution
• Covers 34 distinct configurations, offering variety in simulation scenarios
• Designed specifically for ML applications with community involvement and public accessibility

Limitations

• Column names and specific data structure are not described in the provided sources
• Conflicting metadata: some sources list 'Open Access (green)' license while others list 'None'
• Last update date and exact file formats are unknown, limiting assessment of freshness and usability

Provenance

Source

Wai Tong Chung, Stanford University

Collection Method

High-fidelity physics simulations