Reinforcement learning for automatic quadrilateral mesh generation: A soft actor-critic approach.

Area of Science:

• Computational Engineering
• Artificial Intelligence
• Numerical Simulation

Background:

• Mesh generation is crucial for Computer-Aided Design and Engineering (CAD/E) simulations.
• Current methods face challenges with computational complexity, mesh quality in complex geometries, and speed.
• Existing tools often require significant human expert input, limiting automation.

Purpose of the Study:

• To develop a fully automatic mesh generation system using reinforcement learning (RL).
• To address the limitations of existing semi-automatic mesh generation tools.
• To improve efficiency and mesh quality in numerical simulations.

Main Methods:

• Formulating mesh generation as a Markov Decision Process (MDP).
• Applying a state-of-the-art soft actor-critic reinforcement learning algorithm.
• Training the RL agent to learn optimal mesh generation policies through trial and error.

Main Results:

• The RL-based system achieved fully automatic mesh generation without human intervention.
• The system demonstrated effectiveness, scalability, and generalizability compared to commercial software.
• Eliminated the need for extra clean-up operations post-generation.

Conclusions:

• Reinforcement learning offers a viable approach for fully automated mesh generation.
• The proposed framework significantly advances the state-of-the-art in CAD/E simulation tools.
• This method provides a robust and efficient alternative to traditional semi-automatic techniques.