STRUCTURAL MORPHOGENESIS OF GRID SHELLS BY USING AUTONOMOUS DECENTRALIZED SYSTEM AND REINFORCEMENT LEARNING

Abstract

This paper attempts to apply reinforcement learning to the structural morphogenesis problem of grid shells. The proposed method uses an autonomous decentralized system, which has low computational cost and high versatility because it does not require re-training when the analysis model is changed. Iterative calculations on nodal neighborhood models are performed for the structural analysis, and Q-learning is used to learn shape modification rules that reduce external force work. Numerical examples show the results of the agent generating grid shells with high stiffness. It is also shown that it is possible to generate shells with an unlearned number of grids.