NCCD-PF - A pre-failure narrow concrete cracks dataset for engineering structures damage classification and semantic segmentation

The NCCD-PF dataset was developed for the classification and semantic segmentation of narrow concrete cracks in engineering structures elements at the pre-failure state. It only includes cracks whose width is narrower than 0.3 mm, i.e. the limit value specified in EC 1992-1-1 for typical elements of engineering structures and environmental conditions.

This dataset is dedicated to the early crack detection at a stage when the serviceability limit state has not yet been exceeded and the failure of a structural element has not occurred. By implementing the early crack detection approach, it is possible to protect cracks in order to stop or slow down their propagation and thus to extend the structure's lifespan.

This dataset contains images of cracks appearing on various elements of engineering structures (bridges, viaducts, tunnels) made of reinforced concrete (including abutments, tunnel walls, concrete barriers, pillars). The images were captured on construction sites and during inspections of engineering structures, at different stages of the reinforced concrete structure's working conditions - from the construction stage (when the elements are loaded only by their own weight) to the structure's use stage (when the elements are loaded by most of the design loads). The images are also differentiated by the cause of the cracking (ex., thermal and shrinkage stresses in young concrete, excessive stresses). The images were acquired using fixed-focus cameras without prior conditioning in order to represent the real working conditions of a bridge engineer during structural inspections. The images are characterised by a high degree of complexity due to the quality of the concrete surface finish (e.g. presence of formwork marks, concrete trowel marks), which could potentially be recognised as cracks.

This dataset is dedicated to researchers working in the fields of computer vision, machine learning and deep learning. In particular, it contains domain knowledge in structural health monitoring, so that it can support the work of engineers in detecting cracks of concrete elements in a pre-failure state.

A detailed description of the dataset is presented in A pre-failure narrow concrete cracks dataset for engineering structures damage classification and segmentation (DOI: 10.1038/s41597-023-02839-z).

NCCD-PF数据集专为工程结构构件破坏前阶段的窄混凝土裂缝分类与语义分割任务开发。该数据集仅收录宽度小于0.3毫米的裂缝——该阈值符合欧标EC 1992-1-1针对典型工程结构构件及环境条件规定的限值。

本数据集面向服务极限状态尚未突破、结构构件尚未发生破坏的早期裂缝检测场景。通过开展早期裂缝检测工作，可对裂缝实施干预以阻止或减缓其扩展，进而延长结构的服役寿命。

本数据集涵盖钢筋混凝土工程结构各类构件上的裂缝图像，涉及桥梁、高架桥、隧道等场景，构件类型包括桥台、隧道壁、混凝土护栏、桥墩等。图像采集自施工现场及工程结构巡检过程，覆盖钢筋混凝土结构工作状态的全阶段：从仅承受自重的施工阶段，到承受绝大多数设计荷载的运营使用阶段。图像的裂缝成因存在多样性，例如新浇筑混凝土的热应力与收缩应力、超荷载产生的应力等。所有图像均采用定焦相机拍摄，未进行预处理，以还原桥梁工程师开展结构巡检时的真实工作场景。由于混凝土表面饰面质量存在差异（如存在模板痕迹、抹面痕迹），这些图像具有较高的识别复杂度，此类表面痕迹有可能被误识别为裂缝。

本数据集面向计算机视觉、机器学习与深度学习领域的研究人员。其蕴含结构健康监测领域的专业知识，可辅助工程师开展工程结构构件破坏前阶段的混凝土裂缝检测工作。

该数据集的详细说明见于论文"A pre-failure narrow concrete cracks dataset for engineering structures damage classification and segmentation"（DOI: 10.1038/s41597-023-02839-z）。