Ai-DFab：生成式人工智慧與多代理系統於設計與製造流程之應用

Ai-DFab: Applying Generative AI and Multi-Agent Systems in Design-to-Fabrication Workflows

生成式 AI 在設計發想階段為創作者提供了更為複雜、多元且具創新性的設計產出。本研究提出一套整合生成式人工智慧（Generative AI）與多代理系統（Multi-Agent System, MAS）的設計—製造整合框架 AI-DFab，以曲面木構建築為研究場域，探討如何透過語意推理與代理協作機制實現設計與製造流程的動態銜接。傳統設計常在製造階段出現資訊斷裂與效率低落問題，AI-DFab 旨在於早期設計階段即同時考量構造邏輯、材料特性與製造可行性。

框架由三層代理構成：語意分析與設計生成代理（Semantic & Generative Agents）負責解析語意與生成概念方案；知識推理與策略評估代理（Knowledge Reasoning & Evaluation Agents）整合跨領域模擬與效能分析以收斂策略；製造生成與規劃代理（Fabrication & Planning Agents）結合 MCP-Rhino 與 Manis 系統，將 2D 生成內容轉譯為 3D 幾何模型並自動生成接頭與 CNC 刀具路徑。以物件導向資料結構為核心，確保設計語意、幾何模型與製造資訊間的資料連貫與可追溯。

研究結果顯示，AI-DFab 能有效整合語意推理、策略決策與數位製造，實現設計—製造閉環資料流。系統不僅提升設計決策效率與製造精度，也重新定義 AI 於建築設計中的角色，從被動工具轉化為具推理與協作能力的設計夥伴，為人機共創（Human-AI Co-Design）奠定理論與技術基礎。

Generative AI broadens early-stage ideation by enabling creators to explore design outputs that are more complex, diverse, and novel. This paper presents AI-DFab, a design-to-fabrication integration framework that combines Generative AI with a Multi-Agent System (MAS), using curvilinear timber architecture as the testbed. AI-DFab explores how semantic reasoning and agent-based collaboration can establish a dynamic connection between design intent and fabrication constraints, addressing common discontinuities and efficiency losses that arise when conventional design workflows transition into manufacturing. By incorporating tectonic logic, material characteristics, and fabrication feasibility from the earliest design stage, the framework aims to reduce downstream rework and improve process continuity.

AI-DFab is structured as a three-layer agent architecture: (1) Semantic & Generative Agents parse natural-language intent and generate conceptual design proposals; (2) Knowledge Reasoning & Evaluation Agents integrate cross-domain simulations and performance evaluations to iteratively converge design strategies; and (3) Fabrication & Planning Agents connect MCP-Rhino with the Manis system to translate 2D generative outputs into 3D geometry while automatically generating joinery definitions and CNC toolpaths. An object-oriented data structure underpins the framework to ensure data consistency and traceability across design semantics, geometric models, and fabrication information.

Experimental results demonstrate that AI-DFab enables a closed-loop, data-driven workflow that effectively integrates semantic reasoning, strategic decision-making, and digital manufacturing. The proposed system improves both design decision efficiency and fabrication precision, and it reframes AI’s role in architectural design from a passive tool to a reasoning and collaborative partner. By operationalizing semantic-to-fabrication continuity through MAS, AI-DFab provides both a conceptual and technical foundation for Human–AI co-design in advanced timber construction.