Deep reinforcement learning for microstructural optimisation of silica aerogels.

Area of Science:

• Materials Science
• Computational Modeling
• Nanotechnology

Background:

• Silica aerogels possess valuable properties for aerospace and transportation.
• Characterizing aerogel microstructures is difficult due to nanoporous morphology and gelation randomness.
• Microstructural features critically influence thermal, mechanical, and acoustic performance.

Purpose of the Study:

• To develop a deep reinforcement learning (DRL) framework for optimizing silica aerogel microstructures.
• To enable inverse microstructure design for targeted material properties.
• To assess the efficiency and accuracy of DRL in aerogel design.

Main Methods:

• Utilized a deep reinforcement learning (DRL) framework.
• Modeled silica aerogel microstructures using the diffusion-limited cluster-cluster aggregation (DLCA) algorithm.
• Employed DLCA surrogate models in two environments for accelerated inverse microstructure design.

Main Results:

• The DRL framework effectively optimized silica aerogel microstructure morphology.
• Achieved material property accuracy correlated with environmental complexity.
• Demonstrated the capability to achieve desired material properties with sufficient DRL agent training.

Conclusions:

• The presented DRL methodology offers a resource-efficient approach to designing silica aerogels.
• This framework provides significant computational advantages compared to experimental methods or direct numerical simulations.
• Enables targeted design of aerogel microstructures for specific applications.