Abstract Inspecting the diameter and spacing of rebar is an important task conducted by fabricators and site engineers during the manufacturing and construction stages. This is because the bearing capacity of reinforced concrete structures is affected by the size and position of the rebar, so installing rebar of the correct size and position should be ensured to safeguard the structural integrity of the structure. This study presents a new terrestrial laser scanning (TLS)-based method using machine learning to automatically classify rebar diameters and accurately estimate rebar spacing. To this end, a new methodology, named density based machine model, is proposed to improve classification accuracy. To validate the proposed method, experimental tests on laboratory specimens with rebars of seven different diameters are conducted. The results show that the prediction accuracy for large rebar diameters measuring D25-D40 are up to 97.2%, demonstrating great potential for the application of the proposed technique on manufacturing and construction sites. The key findings of the study are: (1) the proposed density-based modeling method for rebar diameter prediction is superior to the traditional machine learning approach; (2) scan density is one of the most important factors in the prediction results, especially in the small rebar diameter group; and (3) it was found that at least 10 points/cm2 is necessary to ensure accurate rebar diameter classification in small rebar diameters between D10-D20. It is expected that the proposed rebar diameter and rebar spacing technique will be useful in providing autonomous and accurate rebar inspection in manufacturing factories and on construction sites.

中文翻译：

---

使用基于点云数据的机器学习自动钢筋直径分类

摘要 检查钢筋的直径和间距是制造商和现场工程师在制造和施工阶段进行的一项重要任务。这是因为钢筋混凝土结构的承载力受钢筋尺寸和位置的影响，因此应确保安装正确尺寸和位置的钢筋，以保障结构的结构完整性。本研究提出了一种新的基于地面激光扫描 (TLS) 的方法，使用机器学习自动分类钢筋直径并准确估计钢筋间距。为此，提出了一种新的方法，称为基于密度的机器模型，以提高分类精度。为了验证所提出的方法，对具有七种不同直径钢筋的实验室试样进行了实验测试。结果表明，测量 D25-D40 的大钢筋直径的预测精度高达 97.2%，表明该技术在制造和施工现场的应用潜力巨大。研究的主要发现是：（1）所提出的基于密度的钢筋直径预测建模方法优于传统的机器学习方法；(2) 扫描密度是预测结果中最重要的因素之一，尤其是在小钢筋直径组中；(3) 发现至少需要 10 个点/cm2 才能确保 D10-D20 之间的小钢筋直径的准确钢筋直径分类。