钢-混界面脱空是钢管混凝土结构中常见的结构缺陷，这类缺陷会导致钢管混凝土结构的组合截面的应力重分布，进而使结构承载能力下降和刚度降低，甚至引起结构失稳，影响结构的安全性和稳定性。现有的无损检测方法，比如超声探伤法和波传播法，依赖昂贵的设备和复杂的操作流程，检测成本高且定量判别存在局限性，难以在实际工程中广泛应用。因此本文拟基于机电阻抗法开发一种钢-混界面脱空智能定位及定量检测技术，本文的主要工作包括以下几个方面：

      （1）磁吸式传感器的制备及性能验证。首先，制备了一种简易磁吸式传感器。相比粘贴式传感器，虽然其灵敏度较低，但因其安装简单且可重复利用，在实际结构检测中更具优势。然后，利用无损钢管混凝土试件对制备的磁吸式传感器进行了检测，并计算了均方根误差和互相关系数。结果表明，制备的磁吸式传感器对钢-混凝土界面脱空缺陷的发展具有一定的敏感性，可用于钢-混凝土界面脱空的定量分析和定位研究。

      （2）钢-混结构界面脱空缺陷的智能定位研究。制备了含有脱空缺陷的钢管混凝土试件，并对试件表面进行标定，结合磁吸式传感器测得不同标定位置的导纳信号。随后，采用监督学习方法，结合综合采样方法，对比了决策树、随机森林、支持向量机、自适应增强算法、极端梯度提升算法在局部缺陷位置、缺陷边界位置以及局部无缺陷位置的定位效果。最后，采用不需要选取基线数据的无监督方法，即K均值聚类模型，并结合t分布随机邻域嵌入方法，实现了脱空区域的识别。

      （3）钢-混结构界面脱空尺寸的定量识别研究。设计了含不同大小脱空缺陷的钢管混凝土试件，并利用磁吸式传感器测得不同脱空程度试件的导纳信号。通过监督学习方法，分别对缺陷体积率、缺陷长度、缺陷厚度特征建立分类模型，使用训练集对模型进行训练，并使用验证集及测试集对模型进行测试。评价指标包括准确度、精确度、F1分数、召回率以及科恩卡帕值，针对决策树、随机森林、支持向量机、自适应增强算法及极端梯度提升算法进行了综合评价。此外，考虑到实际结构缺陷状态数据标签获取的复杂性，采用无监督学习方法来定量分析脱空缺陷，建立了K均值聚类模型，并使用轮廓系数、误差平方和等指标对缺陷聚类模型进行评价，结合主成分分析技术完成脱空缺陷的定量识别。

      本文采用自制的磁吸式传感器，结合机器学习算法，实现了对脱空缺陷的智能定位，并能够对脱空缺陷尺寸的发展情况进行定量的原位监测。与超声探伤法相比，该检测技术更加经济便捷，传感器可重复使用，具备进一步工程应用的潜力。

      Steel-concrete interfacial voiding is a kind of structural defect in concrete-filled steel tube (CFST) structures. These defects may cause stress redistribution within the composite cross-section, resulting in a reduction in the bearing capacity, the structural stiffness, and the overall safety. Current non-destructive testing methods, such as ultrasonic testing and wave propagation techniques, require expensive equipment and complex operational procedures, leading to high detection costs and limitations in quantitative assessment. These constraints hinder their widespread application in practical engineering. This work proposed an intelligent technique for the localization and quantitative detection of voids at the steel-concrete interface, utilizing the electromechanical impedance (EMI) method. The main contributions of this paper include the following aspects:

      (1) Preparation and performance verification of magnetic adhesion sensors. Initially, a simplified magnetic adhesion sensor was developed. Although it has lower sensitivity compared to adhesive sensors, it offers advantages in easy installation and reusability, making it more suitable for practical structural inspections. Subsequently, non-destructive testing was performed on steel-concrete specimens using the fabricated magnetic adhesion sensor. Key metrics, such as the root mean square deviation and mutual correlation coefficient, were calculated. The results demonstrate that the magnetic adhesion sensor possesses sufficient sensitivity to detect void defects at the steel-concrete interface, enabling quantitative analysis and localization of these voids.

      (2) Intelligent localization for void defects at the steel-concrete interface. Steel-concrete specimens with pre-defined void defects were prepared, and their surfaces were calibrated. Electromagnetic impedance signals were measured at different calibrated positions using the magnetic adhesion sensor. These signals were then analyzed using supervised learning methods combined with comprehensive sampling techniques. The positioning performance of several algorithms—decision trees, random forests, support vector machines, adaptive boosting, and extreme gradient boosting—was compared for local defect positions, defect boundary positions, and regions without defects. Finally, an unsupervised method, the K-means clustering model, was employed without the need for baseline data selection, combined with the t-distributed stochastic neighbor embedding (t-SNE) method for recognizing void regions.

      (3) Quantitative identification for void sizes at the steel-concrete interface. Steel-concrete specimens with void defects of varying sizes were designed, and their electromagnetic impedance signals were measured using the magnetic adhesion sensor. Supervised learning methods were used to establish classification models for defect volume rate, defect length, and defect thickness. These models were trained on training sets, and validated and tested on validation and test sets, respectively. The models' performance was evaluated using five metrics: accuracy, precision, F1 score, recall rate, and Cohen's kappa coefficient, across five algorithms: decision trees, random forests, support vector machines, adaptive boosting, and extreme gradient boosting. Considering the complexity of obtaining labels for actual structural defects, unsupervised learning methods were also employed for quantitative analysis. A K-means clustering model was established and evaluated using metrics such as the silhouette coefficient and sum of squared errors, combined with principal component analysis (PCA) to achieve quantitative identification of void defects.

      This study employed a self-fabricated magnetic adhesion sensor and machine learning algorithms for the intelligent localization and quantitative in-situ monitoring of void defects. Compared to ultrasonic testing methods, this approach is more cost-effective and convenient, featuring reusable sensors, and shows great potential for further engineering applications.

[1]胡彦孥. 钢管混凝土脱粘的振动特性及损伤定位研究[D]. 重庆: 重庆大学, 2020.

[2]苏俊臣. 钢管混凝土拱桥调查及其脱空问题研究[D]. 成都: 西南交通大学, 2012.

[3]超声法检测混凝土缺陷技术规程[S]. 北京: 中国城市出版社, 2000.

[4]王军文, 马少宁, 刘志勇, 等. 钢管混凝土脱空无损检测方法试验研究[J]. 石家庄铁道大学学报(自然科学版), 2021, 34(02): 38-45.

[5]杨科. 基于瞬态冲击的钢管混凝土拱桥钢管脱空检测技术研究[D]. 重庆: 重庆交通大学, 2012.

[6]陈劲, 陈晓东, 赵辉, 等. 基于红外热成像法和超声波法的钢管混凝土无损检测技术的试验研究与应用[J]. 建筑结构学报, 2021(S02): 042.

[7]刘夏平, 唐述, 杨作用, 等. 脱空钢管混凝土短柱偏心受压破坏形态试验研究[J]. 广州大学学报：自然科学版, 2010, 9(3): 5.

[8]刘夏平, 唐春会, 杨作用, 等. 脱空率对偏心受压钢管混凝土力学性能影响研究[J]. 中山大学学报：自然科学版, 2010(S1): 4.

[9]刘漳. 钢管混凝土结构脱空后的受力性能研究[D]. 杭州: 浙江大学, 2015.

[10]Chen S, Zhang H. Numerical analysis of the axially loaded concrete filled steel tube columns with debonding separation at the steel-concrete interface[J]. Steel & Composite Structures, 2012, 13(3): 277-293.

[11]张永宁, 张雪松. 管内混凝土脱空检测新方法及脱空对钢管砼拱桥力学性能影响的研究[D]. 重庆交通大学, 2013.

[12]汪旭, 邹中权, 王志美. 含缺陷钢管混凝土超声波特性试验研究[J]. 湖南工业大学学报, 2013, 27(2).

[13]杨金. 基于HHT的钢管混凝土缺陷特征提取研究与FPGA实现[D]. 湘潭: 湖南科技大学, 2016.

[14]Hongbing C, Xin N, Shiyu G, et al. Interfacial imperfection detection for steel-concrete composite structures using NDT techniques: A state-of-the-art review[J]. Engineering Structures, 2021, 245.

[15]艾德米. 基于压电传感机械阻抗的结构损伤识别方法研究[D]. 武汉: 华中科技大学, 2017.

[16]Shi Y, Luo M, Li W, et al. Grout Compactness Monitoring of Concrete-filled Fiber Reinforced Polymer Tube Using Electro-Mechanical Impedance (EMI)[J]. Smart Material Structures, 2018.

[17]付茂森. 基于卷积自编码神经网络的桥梁损伤识别研究[D]. 石家庄: 石家庄铁道大学, 2022.

[18]江永清. 基于深度学习的混凝土构件损伤检测与定位方法研究[D]. 济南: 山东建筑大学, 2022.

[19]Farrar C R, Baker W E, Bell T M, et al. Dynamic Characterization and Damage Detection in the I-40 Bridge Over the Rio Grande[J]. 1994.

[20]Sun F P, Chaudhry Z A, Rogers C A, et al. Automated real-time structure health monitoring via signature pattern recognition[J]. Proceedings of SPIE - The International Society for Optical Engineering, 1995, 2443.

[21]许斌, 李冰, 宋刚兵, 等. 基于压电陶瓷的钢管混凝土柱剥离损伤识别研究[J]. 土木工程学报, 2012, 45(7): 11.

[22]杨映泉, 许斌, 栾乐乐, 等. 基于表面压电波动测量的钢管砼剥离检测试验[J]. 压电与声光, 2018, 40(3): 5.

[23]许斌, 陈梦琦, 余地华, 等. 基于压电阻抗的钢管混凝土柱界面缺陷检测研究[J]. 施工技术, 2015(11): 5.

[24]Saravanan T J. Surface-Mounted Smart PZT Sensors for Monitoring Damage Using EMI-Based Multi-Sensing Technique[J]. Engineering Proceedings, 2021, 10.

[25]Moll, Jochen. Damage detection in grouted connections using electromechanical impedance spectroscopy[J]. Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science, 2018, 288: 947-950.

[26]朱健健. 基于圆形压电传感器机电阻抗耦合模型的结构健康监测技术研究[D]. 厦门: 厦门大学, 2020.

[27]冯磊. 基于深度置信网络的桥梁损伤识别研究[D]. 西安: 西安建筑科技大学, 2020.

[28]Mario D O, Andre M, Jozue V F. A New Structural Health Monitoring Strategy Based on PZT Sensors and Convolutional Neural Network[J]. Sensors, 2018, 18(9): 01-21.

[29]Demi A, Fang M, Feng Y, et al. Electromechanical impedance-based concrete structural damage detection using principal component analysis incorporated with neural network[J]. Journal of Intelligent Material Systems and Structures, 2022, 33(17).

[30]Ai D, Mo F, Han Y, et al. Automated identification of compressive stress and damage in concrete specimen using convolutional neural network learned electromechanical admittance[J]. Engineering Structures, 2022, 259: 114176.

[31]Alazzawi O, Wang D. Deep convolution neural network for damage identifications based on time-domain PZT impedance technique[J]. Journal of Mechanical Science and Technology, 2021, 35(4).

[32]姜世宇. 基于压电阻抗和卷积神经网络的风电塔筒法兰节点健康监测研究[D]. 青岛: 青岛理工大学, 2022.

[33]Park S, Ahmad S, Yun C B, et al. Multiple Crack Detection of Concrete Structures Using Impedance-based Structural Health Monitoring Techniques[J]. Experimental Mechanics, 2006, 46(5): 609-618.

[34]朱宏平, 王丹生, 张俊兵. 基于压电阻抗技术的结构损伤识别基本理论及其应用[J]. 工程力学, 2008, 025(A02): 34-43.

[35]陈潜, 许斌, 陈洪兵. 基于表面波的钢管砼界面剥离检测模拟分析[J]. 压电与声光, 2019, 41(1): 5.

[36]温家宇. 基于压电阻抗法的组合结构缺陷检测模拟与试验研究[D]. 长沙: 湖南大学, 2018.

[37]Guo B, Chen D, Huo L, et al. Monitoring of Grouting Compactness in Tendon Duct Using Multi-Sensing Electro-Mechanical Impedance Method[J]. Applied Sciences, 2020, 10(6).

[38]郭惠勇, 王磊, 李正良. 基于改进PSO算法的两阶段损伤识别方法[J]. 西南交通大学学报, 2011, 46(6): 926-932.

[39]Ramezani H, Kazemirad S, Shokrieh M M, et al. Effects of adding carbon nanofibers on the reduction of matrix cracking in laminated composites: Experimental and analytical approaches[J]. Polymer Testing, 2020, 94: 106988.

[40]Chen D, Montano V, Huo L, et al. Depth detection of subsurface voids in concrete-filled steel tubular (CFST) structure using percussion and decision tree[J]. Measurement, 2020, 163: 107869.

[41]骆勇鹏, 谢隆博, 廖飞宇, 等. 基于时序分析理论的钢管混凝土脱空缺陷检测方法研究[J]. 工业建筑, 2019, 49(10): 6.

[42]Sf A, Fp B, Mb C, et al. Autonomous damage recognition in visual inspection of laminated composite structures using deep learning[J]. Composite Structures, 2021, 268.

[43]黎赫东. 基于重构压电导纳信号的结构损伤识别[D]. 武汉: 华中科技大学, 2021.

[44]Chen D, Shen Z, Huo L. Percussion-based quasi real-time void detection for concrete-filled steel tubular structures using dense learned features[J]. Engineering Structures, 2023, 274: 115197.

[45]Eltouny K, Gomaa M, Liang X. Unsupervised Learning Methods for Data-Driven Vibration-Based Structural Health Monitoring: A Review[J]. Sensors, 2023, 23(6).

[46]Ezugwu A E, Ikotun A M, Oyelade O O, et al. A comprehensive survey of clustering algorithms: State-of-the-art machine learning applications, taxonomy, challenges, and future research prospects[J]. Engineering Applications of Artificial Intelligence, 2022, 110: 104743.

[47]Yan B, Zou Q, Dong Y, et al. Application of PZT Technology and Clustering Algorithm for Debonding Detection of Steel-UHPC Composite Slabs[J]. Sensors, 2018, 18(9).

[48]马宏伟, 林逸洲, 聂振华. 利用少量传感器信息与人工智能的桥梁结构安全监测新方法[J]. 建筑科学与工程学报, 2018, 35(5): 15.

[49]Tibaduiza D A, Mujica L E, Rodellar J, et al. Structural damage detection using principal component analysis and damage indices[J]. Journal of Intelligent Material Systems and Structures, 2015.

[50]Joao Santos, Christian, Calado, et al. On-line unsupervised detection of early damage.[J]. Structural Control & Health Monitoring, 2016.

[51]A A M, B J S, C D R, et al. Online unsupervised detection of structural changes using train–induced dynamic responses[J]. Mechanical Systems and Signal Processing, 165.

[52]Alamdari M M, Rakotoarivelo T, Khoa N L D. A spectral-based clustering for structural health monitoring of the Sydney Harbour Bridge[J]. Mechanical systems and signal processing, 2017, 87(PT.A): 384-400.

[53]Fernandez-Navamuel A, Pardo D, Magalhães F, et al. Deep neural network for damage detection in Infante Dom Henrique bridge using multi-sensor data[J]. Structural Health Monitoring, 2024: 14759217241227455.

[54]Liu X, Athanasiou C E, Padture N P, et al. A machine learning approach to fracture mechanics problems[J]. Acta Materialia, 2020, 190: 105-112.

[55]Lever J, Krzywinski M, Altman N. Principal component analysis[J]. Nature Methods, 2017, 14(7): 641-642.

[56]Akiba T, Sano S, Yanase T, et al. Optuna: A Next-generation Hyperparameter Optimization Framework[J]. ACM, 2019.

[57]Giurgiutiu V, Rogers C A. ElectroMechanical (E/M) Impedance Method for Structural Health Monitoring and Non-Destructive Evaluation[J]. 1997.

[58]Park G, Sohn H, Farrar C R, et al. Overview of Piezoelectric Impedance-Based Health Monitoring and Path Forward[J]. The Shock and Vibration Digest, 2003(6): 35.

[59]Annamdas V G, Radhika M A. Electromechanical impedance of piezoelectric transducers for monitoring metallic and non-metallic structures: A review of wired, wireless and energy-harvesting methods[J]. Journal of Intelligent Material Systems and Structures, 2013, 24(9): 1019-1040.

[60]张婷. 基于压电陶瓷的钢管混凝土柱界面剥离损伤监测的实验研究[D]. 长沙: 湖南大学, 2011.

[61]杨霞. 基于压电陶瓷的GFRP管混凝土组合柱损伤监测试验研究[D]. 银川: 宁夏大学, 2021.

[62]曹平. 钢管混凝土密实性的非线性振动识别研究[D]. 重庆: 重庆大学, 2015.

[63]王军. 钢管混凝土脱粘影响非线性振动特性的机理研究[D]. 重庆: 重庆大学, 2017.

[64]邓海明. 基于压电陶瓷的钢管混凝土组合结构界面损伤监测研究[D]. 长沙: 湖南大学, 2015.

[65]赵玉栋. 基于MASW的钢-混凝土组合结构界面损伤检测技术[D]. 北京: 清华大学, 2021.

[66]庄志有. 基于外贴压电陶瓷的钢管混凝土内部缺陷检测方法研究[D]. 泉州: 华侨大学, 2019.