3D打印混凝土界面孔隙智能检测方法研究  

Research on intelligent detection method of 3D printed concrete interface pore

作　　者：曾妮 马宗方[1] 宋琳[1] 段明 ZENG Ni;MA Zongfang;SONG Lin;DUAN Ming(College of Information and Control Engineering,Xi'an University of Architecture and Technology,Xi'an 710055,China)

机构地区：[1]西安建筑科技大学信息与控制工程学院,陕西西安710055

出　　处：《工程设计学报》2025年第1期11-22,共12页Chinese Journal of Engineering Design

基　　金：国家自然科学基金面上项目(62276207)。

摘　　要：目前,3D打印混凝土领域仍存在诸多问题,严重制约了其大规模工业化生产与应用。其中,孔隙为最常见缺陷,亟须发展相关检测技术,以提高混凝土的打印质量。针对现有3D打印混凝土界面孔隙检测方法主要依赖人的主观经验,且存在耗时长、成本高和计算资源耗费量大等缺陷,引入基于深度学习的目标检测算法,提出了一种轻量级的孔隙智能检测方法。首先,利用传统图像处理算法对3D打印混凝土界面孔隙图像进行预处理,并构建适用于目标检测算法的孔隙图像数据集;同时,基于所构建数据集的特点对锚框计算方法进行优化,以获取更适合界面孔隙目标的锚框,从而提升检测准确度。然后,在检测方法的主干网络中,利用ShuffleNetv2网络进行多尺度特征提取,并去掉部分网络以降低网络深度和减少计算参数量,从而提高孔隙检测效率。最后,在特征提取网络中融合极化自注意力机制模块,在保持高分辨率的同时增强对孔隙目标的关注度,以提高检测精度。实验结果表明,所提出的方法能够有效完成3D打印混凝土界面孔隙的智能化检测,通过与多种检测算法对比,发现该方法的多个性能指标均有所提升,检测效率提升显著。研究结果可为后续混凝土的质量控制和性能评估提供一定的数据支持。At present,the 3D printed concrete field is still hampered by numerous issues that impede its large-scale industrial production and application.Among these,pores stand out as the most prevalent defect.Consequently,there is an urgent imperative to develop pertinent detection technologies for enhancing the printing quality of concrete.Aiming at the existing 3D printed concrete interface pore detection methods that mainly rely on subjective experience of individuals,and have disadvantages such as long-time consumption,high cost and large computational resource consumption,a lightweight intelligent pore detection method is proposed by introducing a deep learning-based object detection algorithm.Firstly,the traditional image processing algorithms were employed to preprocess the 3D printed concrete interface pore images,and the pore image dataset suitable for the target detection algorithm was constructed.At the same time,based on the characteristics of the constructed dataset,the anchor-box calculation method was optimized to acquire anchor boxes that were better suited to the interface pore targets,so as to improve the detection accuracy.Then,within the backbone network of the detection method,the ShuffleNetv2 network was utilized for multi-scale feature extraction,and part of the network was removed to reduce the network depth and the number of calculation parameters,thereby enhancing the pore detection efficiency.Finally,in order to improve detection precision,the polarized self-attention mechanism module was incorporated into the feature extraction network to enhance the attention to the pore target while maintaining high resolution.Experimental results demonstrated that the proposed method could effectively complete the intelligent detection of 3D printed concrete interface pores.Through comparing with various detection algorithms,it was found that multiple performance indicators of the method were improved,and the detection efficiency was significantly boosted.The research results can provide some data support f

关 键 词：3D打印混凝土 孔隙检测 ShuffleNetv2 自注意力机制 多尺度特征融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TP399[自动化与计算机技术—计算机科学与技术]