Deep Neural Network and Monte Carlo Tree Search applied to Fluid-Structure Topology Optimization

Area of Science:

• Computational Fluid Dynamics
• Mechanical Engineering
• Artificial Intelligence

Background:

• Topology optimization is crucial for designing efficient structures.
• Current methods like Cellular Automata, Adjoint, and Level-Set have limitations.
• Integrating AI can enhance optimization processes.

Purpose of the Study:

• To apply Deep Neural Network (DNN) algorithms to fluid-structure topology optimization.
• To introduce a novel, autonomous optimization strategy.
• To compare DNN-based optimization with classical methods.

Main Methods:

• Utilizing Deep Neural Network algorithms combined with Monte Carlo Tree Search.
• Representing the design space as a computational grid with fluid or solid states.
• Employing an incompressible fluid solver as the objective function, independent of the optimization process.

Main Results:

• The DNN approach demonstrated effective topology optimization.
• The system learned and optimized without human intervention.
• Results were comparable to traditional adjoint topology optimization codes.

Conclusions:

• Deep Neural Networks offer a powerful, autonomous approach to topology optimization.
• This AI-driven strategy can be integrated with existing methods.
• The technique shows promise for complex fluid-structure interaction problems.