Qichao Song, Siyi Chen, Huiling Zhang
The sample data were reliable, with strong intra- and inter-rater agreement. Significant differences appeared in foot take-off, knee up, and start leg flexion (p < 0.05; Cohen's d = 0.8). Conduct and aggregate criteria showed small to moderate negative correlations (r = 0.181; p < 0.05 and r = 0.396; p < 0.01). On the other hand, there was a perfect correlation between support leg position and attacking leg foot position.
Bio-inspired architectural design increasingly relies on generative artificial intelligence to expand early-stage concept exploration, yet current workflows often suffer from vague requirement definition, subjective proposal selection, and weak connections between user expectations, visual generation, and design evaluation. This study proposes a closed-loop human-AI decision-support framework for biomimetic architectural concept generation. The framework first uses Kansei-oriented requirement analysis and the KANO model to identify and classify stakeholder expectations concerning morphology, structural rationality, environmental integration, cultural narrative, visual novelty, interactivity, and sustainability. On the basis of 282 valid questionnaire responses, the most influential requirement categories are further translated into an Analytic Hierarchy Process (AHP) hierarchy, where expert judgement from a five-member specialist panel is used to derive criterion and sub-criterion weights. These weights are then converted into structured prompts-via a formally specified weight-to-language conversion strategy-to guide a diffusion-based image-generation system toward more targeted biomimetic concepts. Finally, TOPSIS is applied to rank generated design alternatives according to the same weighted criteria, thereby creating a traceable link from requirement discovery to generation and decision-making. Case studies involving eagle-, manta ray-, and cheetah-inspired architectural concepts indicate that the framework improves the explicitness of design objectives, supports more consistent comparison among alternatives, and reduces reliance on purely intuitive aesthetic judgement. An ablation comparison confirms that each stage of the framework contributes incrementally to the quality of final outcomes. This study proposes an integrated workflow that combines requirements modeling, multi-criteria evaluation, and AI-assisted visual design.