- 홈
- Research
Research
Research
ONGOING Sep 2026 – Aug 2031
Development of a CFD Method for Flow Domain Generation and Fast Prediction of Flow Characteristics from CAD Models
CFD analysis of internal flows in pipes, ducts, and valves still relies heavily on expert knowledge and high computational cost, and defining the flow domain in commercial preprocessing tools remains largely manual. AI-based approaches such as PINN and GNN have so far been applied mainly to simple geometries.
This project develops a Hybrid AI-CFD based end-to-end CFD method that takes only CAD geometry and essential boundary conditions as input and processes flow domain generation, meshing, CFD analysis, and result extraction in a single workflow. Conventional CFD solvers are retained for the physics, and AI is applied selectively at each stage.
End-to-end workflow
INPUT
CAD model + boundary conditions
STEP 01
Automatic flow domain generation
STEP 02
Mesh generation and quality control
STEP 03
Automated CFD analysis
STEP 04
Result extraction
STEP 05
Post-processing and visualization
Targets
70%+
Reduction in preprocessing time
<10%
Prediction error compared with conventional CFD
1,000+
Geometries in the AI training dataset
Key research topics
01 Flow domain generation
Recognition of internal flow regions from CAD models, semi-automatic inlet and outlet boundary setting, and geometry simplification to a user-defined level.
02 Mesh quality control
Mesh optimization based on aspect ratio and skewness, and selection of cell size and count considering the time step size.
03 Automated CFD analysis
Python-based solver setup, batch simulations, and automatic error handling and re-runs.
04 AI-based result extraction
Automatic extraction of key physical quantities, quantification of design indicators, and comparison of design alternatives.
05 Post-processing support
Result export in file formats compatible with commercial post-processing programs.
Conventional CFD vs. this project
| Item | Conventional CFD | This project |
|---|---|---|
| Scope of automation | Partial stages or speed-up only | Entire process from CAD to results |
| Users | CFD experts | Non-experts and small and medium-sized enterprises |
| Flow domain generation | Manual, experience-based | Automatic, based on CAD data |
| Use of results | Visualization-focused | Quantitative indicators and real-time operation optimization |
| AI application | Limited to simple geometries | Applicable to internal flows |
Five-year roadmap
YEAR 1
Basic algorithms and dataset
- Flow domain generation algorithm
- Inlet/outlet recognition and defeaturing
- AI training dataset
YEAR 2
Preprocessing prototype
- AI-based geometry recognition
- Mesh quality check functions
- Automated preprocessing prototype
YEAR 3
Mesh quality and solver coupling
- AI-based mesh quality control
- Mesh uniformity and solution stability
- CAD–mesh–solver workflow
YEAR 4
Design support and real-time results
- AI-based post-processing
- Design optimization support
- Real-time prediction
YEAR 5
System integration
- End-to-end platform
- Validation with industrial products
- User interface for non-experts
Applications
Product design support
Flow checks before prototyping, without a dedicated CFD specialist.
Digital engineering
Integrated CAD-to-analysis automation for digital twins and smart manufacturing.
Real-time operation
Flow prediction for system control, such as exhaust emission control of eco-friendly ships.
Other industries
Automotive, shipbuilding, aerospace, and energy, where design iterations are frequent.