Subjective-guided generation of tree-like structural forms

This study proposes a Monte Carlo Tree Search (MCTS)-based method for generating tree-like structures that integrates architects’ subjective judgment into structural form generation. Addressing the overreliance on automated optimization in digital design, the method establishes an interactive framework in which architects participate in real-time decision-making while structural performance constraints are evaluated in parallel. The approach embeds subjective guidance and objective performance evaluation through discretized node sampling, finite element analysis (FEA) based on bending moment thresholds at nodes, and the MCTS – UCT decision model. Tree-like structural growth is reformulated as a stepwise decision-making process, with structural feasibility validated using OpenSeesPy and 1:25 scaled physical loading experiments. The results indicate that: (1) the subjectivity-guided structure generated using Monte Carlo Tree Search (MCTS) achieves a material utilization ratio comparable to that of the structure generated by the hanging-chain method and significantly outperforms the BP neural network approach, demonstrating the effectiveness of the MCTS-based method; and (2) the results of the physical experiments closely match those of the numerical simulations, confirming the scientific validity and rationality of the proposed model.