MOO_Themal_Mechanical_Composites

Efficient Multi-Objective Optimization of Composite Microstructures for Thermal Protection Systems

This repository contains the implementation of the surrogate neural network model described in "Surrogate model based Thermo-mechanical optimization of honeycomb inspired composite RVE". The model predicts effective thermo-mechanical properties of periodic Representative Volume Elements (RVEs) based on their geometric parameters.

Input Parameters

The model accepts 10 geometric parameters that define the RVE architecture:

1. web [1.0 - 3.0]: Honeycomb web thickness (inputs[0])
2. iso [0.25 - 0.75]: Fraction of lightweight phase within the web (inputs[9])
3. theta1 - theta8 [0° - 180°]: Reinforcement angles for each of the 8 cells (inputs[1:9])

Output Properties

The model predicts 6 effective properties:

1. rho [kg/m³]: Effective density
2. E11 [GPa]: Elastic modulus in direction 1
3. E22 [GPa]: Elastic modulus in direction 2
4. G12 [GPa]: Shear modulus
5. K1 [W/mK]: Thermal conductivity in direction 1
6. K2 [W/mK]: Thermal conductivity in direction 2

Example

from model.model import create_model, run_model

# Create model
model = create_model()

# Example input parameters
inputs = [
    1.5,    # web thickness
    0.0,    # theta1
    45.0,   # theta2
    90.0,   # theta3
    135.0,  # theta4
    180.0,  # theta5
    135.0,  # theta6
    90.0,   # theta7
    45.0    # theta8
    0.5,    # iso fraction
]

# Get predictions
outputs = run_model(model, inputs)
print("Predicted properties:", outputs)

Repository Structure

your-repo/
├── model/
│   ├── nn_constants_parsed.json    # Model weights and constants
│   ├── model.py                    # Model implementation
│   └── example_usage.py           # Example usage script for multiobjective optimization
├── data/
│   ├── RVE_Geom.pkl               # RVE parameters data used for FEM simulations
│   ├── RVE_props.pkl              # Effective properties data from FEM simulations
├── README.md
└── requirements.txt

Data Availability

Citation

If you use this model in your research, please cite:

[Citation details will be added upon publication]

License

see the LICENSE file for details.

Contact

shmuliko@technion.ac.il