Reinforcement Learning Guided Engineering Design: from Topology Optimization to Advanced Modelling

Applying reinforcement learning for topology optimization is a novel approach to engineering design. In this work, I produced 6 by 6 and 5 by 5 grid topologies by the PPO algorithm and 4 4 by the HRL algorithm in adequate compute wall-clock time. I have also addressed increasing the calculation speed for artificial intelligence-driven topology optimization by combining genetic algorithms and reinforcement learning in a straightforward sequential (one method after another). First, I apply genetic algorithms to get an outline of the topology, and then I 'fine-tune' or refine the obtained topology by reinforcement learning approach. This way, I can obtain more optimal topologies and reduce wall-clock time. In particular, I optimized topology for a 10 by 10 grid, which can be seen as an improvement over a 6 by 6 topology obtained by reinforcement learning alone. The genetic algorithm alone could not produce such an optimal topology as a combination of reinforcement learning and genetic algorithms in comparable wall-clock times.