Automated Quality Inspection of BIM Models Using Graph Neural Networks

Manual checking of building information models is time-consuming and error-prone. This paper formulates BIM quality inspection as a graph learning problem, where building components are nodes and spatial-topological relations are edges, and proposes a graph neural network that identifies missing, duplicated and mis-connected components.

问题提出

建筑信息模型的质量校验长期依赖人工审查，效率低且易遗漏。既有自动化方法多基于规则匹配，难以覆盖复杂多样的缺陷类型。

方法

本文将 BIM 质量校验建模为图学习问题：以建筑构件为节点、空间与拓扑关系为边，构建模型图；提出一种图神经网络，通过节点分类与边分类联合识别缺失构件、重复构件与错误连接。

实验与结论

在包含 240 个真实工程模型的数据集上，所提方法在缺陷识别任务上的 F1 值达到 0.89，较基于规则的方法提升约 21 个百分点，单模型平均检测时间由数十分钟缩短至秒级。

代码与数据

训练与推理代码已在项目主页开源，示例数据随代码发布。

@article{Y.Chen20262026-bim,
  title   = {Automated Quality Inspection of BIM Models Using Graph Neural Networks},
  author  = {Y. Chen and W. Li and M. Sun},
  journal = {Advanced Engineering Informatics},
  year    = {2026},
  doi     = {10.1016/j.aei.2026.100002},
}