建筑信息模型 (BIM) 中的人工智能

BIM is more than 3D drawing. Each element in a model, such as a wall, pipe or beam, carries properties like its material, fire rating, manufacturer and cost code. Authoring tools such as Autodesk Revit, Graphisoft Archicad and Tekla Structures create these models. IFC (Industry Foundation Classes), an open standard maintained by buildingSMART and published as ISO 16739, lets models move between different programs. Clash detection is often assumed to be AI, but at its core it is geometry. Tools such as Navisworks and Solibri test whether objects intersect or violate required clearances. A combined model from several disciplines can produce thousands of clashes, many of them trivial or repeated. Machine learning helps by grouping related clashes, predicting which ones coordinators usually treat as real, and ranking them by likely cost or schedule impact. Quantity takeoff counts and measures everything that must be bought and built. With a well-structured BIM model this is mostly a database query. AI is more useful when the model is incomplete or only 2D drawings exist, because computer vision can recognize symbols, rooms and dimensions on drawing sheets. Automated code checking turns regulations into rules a computer can check. Singapore's CORENET program was an early government effort. The hard part is that codes are written in natural language, full of exceptions and judgment calls. Language models are being tested to help convert code clauses into rules, but a person still has to interpret them. Generative design lets designers state goals and constraints while the software explores many options. It relies on optimization and search as much as on machine learning. For the handoff to the building owner, standards such as COBie organize the equipment and maintenance data owners need, and AI can check that data for gaps. A common misconception is that AI designs buildings on its own. In practice these tools are only as good as the data and modeling conventions the team follows.