Machine-Learning-Based Characterization and Inverse Design of Metamaterials

Area of Science:

• Materials Science
• Computational Materials Science
• Machine Learning Applications

Background:

• Metamaterials possess unique structures yielding exceptional properties.
• Traditional characterization methods (experiments, FEM) are time-intensive for extensive structure exploration.
• Designing metamaterials with specific properties requires efficient predictive tools.

Purpose of the Study:

• To develop a machine learning-based approach for rapid prediction of effective metamaterial properties.
• To accelerate the discovery of microstructures with diverse and outstanding characteristics.
• To enable inverse design of metamaterials with multiple excellent performances.

Main Methods:

• Construction of 2D and 3D microstructures (porous, solid-solid, fluid-solid).
• Utilizing Finite Element Methods (FEM) for property determination.
• Applying Random Forest (RF) for property prediction and Aquila Optimizer (AO) for inverse design.

Main Results:

• The RF regression model achieved high accuracy (R² > 0.98, MAPE < 0.088, RMSE < 0.03).
• The AO method successfully designed an optimized structure with high Young's modulus and low thermal conductivity within 30 iterations.
• The machine learning approach accurately characterizes metamaterial properties and accelerates simulation.

Conclusions:

• The developed machine learning-based method accurately predicts metamaterial properties, significantly reducing design time.
• This approach facilitates the discovery of novel microstructures with tailored, multi-performance characteristics.
• The study provides a framework for designing advanced metamaterials for practical applications, including vibration energy absorbers.