【作者】 胡航；

【导师】 许颖；

【作者基本信息】 哈尔滨工业大学 ， 土木工程， 2022， 硕士

【摘要】 钢结构作为土木工程中的常见结构,具有轻质、高强的特点,但同时存在着突出的耐久性问题,因此需要涂层的保护以延长钢结构的寿命。涂层涂装的质量在一定程度上影响了钢结构的耐久性能,但是目前对于钢结构涂层的检测手段较为缺乏,规范中规定的手持涂层测厚仪只能对涂层的总体厚度进行检测,并且只能对涂层进行随机取点进行检测,无法对于涂层整体厚度分布进行把控,并且对涂层的内部缺陷也无法进行有效识别。随着太赫兹技术的兴起,太赫兹对于非极性物质具有极高的穿透性以及其高信噪比的特点,使其具备对钢结构涂层的无损检测的潜力。目前太赫兹无损检测的难点主要是:(1)材料折射率的获取方法不够成熟;(2)涂层内部缺陷缺少参照,因此难以对太赫兹的检测结果进行检验。本文通过理论分析、实验研究和数值模拟相结合的方法,探究以太赫兹时域光谱技术为基础、以BP神经网络算法作为数据处理手段的钢结构涂层成像以及内部缺陷的无损检测技术。本文基于太赫兹传播理论,为实现钢结构涂层厚度及缺陷的无损检测提供理论基础,分别验证了传统方法获取折射率并计算涂层厚度的方法与神经网络拟合涂层厚度的方法的准确性,确定了神经网络相对于传统方法的优越性;以太赫兹传播理论为基础,在COMSOL中模拟太赫兹的激发、传播和反射路径,并与太赫兹反射实验进行对比,证明了电磁仿真计算的准确性;通过对不同波长的太赫兹波进行仿真计算,确定了用于钢结构涂层的最佳太赫兹波长,将仿真结果与太赫兹扫描实验进行对比,仿真计算的太赫兹扫描信号与太赫兹扫描实验信号一致,2.5-3.5 THz频段的范围的太赫兹波是最适合用于钢结构无损检测的波段,最终确定以3 THz作为仿真计算的太赫兹波进行电磁仿真计算。本文基于涂层缺陷的太赫兹扫描实验,通过时域谱成像、频域谱成像和吸收谱三种成像方式的对比,最终确定时域谱成像为最适合于钢结构涂层的成像方式;通过对太赫兹信号对不同涂层的反射特征分析,编写太赫兹时域反射信号处理算法,提取太赫兹时域谱中成像特征信息(太赫兹抵达涂层界面的时间、幅值),最终可以成功提取太赫兹时域信号中的成像特征信息;构建BP神经网络,通过太赫兹在涂层内的飞行时间、反射峰幅值和太赫兹在涂层间的飞行速度拟合涂层厚度信息,以黑色油漆涂层和白色油漆涂层进行太赫兹扫描实验,分别采取3600个训练集点位数据训练神经网络模型,以900个测试集点位测试神经网络模型,最终可达到98.6%的训练精度和97.5%的测试精度。本文用太赫兹扫描设备分别对单一涂层和复合涂层进行了扫描实验,通过神经网络模型对实验数据进行处理,最终得到钢结构单一涂层和复合涂层的厚度分布,与规范中的检测方法进行对照,以太赫兹技术为支撑,神经网络为数据处理手段的钢结构涂层厚度检测具有较高的准确率,单一涂层试件的准确率可达到96.5%的测试精度,复合涂层可达到96.9%的测试精度;通过对不均匀涂层的扫描实验,获取了不均匀涂层的厚度分布信息,通过引入误差距离作为涂层均匀性评价标准,最终发现误差距离大的涂层其均匀性较差,误差距离小的涂层均匀性较好,证明以误差距离作为参数可以评价钢结构涂层的均匀性好坏。本文对钢结构涂层缺陷研究,总结了钢结构涂层的四种缺陷(夹杂、漏涂、脱粘、气泡),根据涂层缺陷的产生机理,制作了四种涂层缺陷试件;通过对四种缺陷的内部特征研究,使用COMSOL对四种缺陷进行仿真计算,每个仿真模型提取3600个点位数据训练神经网络模型,最低可达到96%以上的准确率;对四种涂层缺陷进行太赫兹扫描实验,选取6400个点位数据作为测试集数据测试神经网络,最低可达到95%的准确率;通过对神经网络模型预测的涂层厚度信息进行可视化处理,可实现钢结构涂层缺陷的成像。更多还原

【Abstract】 As a common structure in geotechnical engineering,steel structure has the characteristics of light weight and high strength,but it also has outstanding durability problems,so it needs coating protection to extend the service life of steel structure.The quality of coating has affected the durability of steel structures to a certain extent,but there is a lack of inspection methods for steel structure coating at present.The hand-held coating thickness gauge specified in the specification can only detect the overall thickness of the coating,and can only detect the coating at random points.It can not control the overall thickness distribution of the coating,and can not effectively identify the internal defects of the coating.With the rise of terahertz technology,terahertz has the potential for nondestructive testing of steel structure coatings due to its high penetrability to non-polar substances and its high signal to noise ratio.At present,the main difficulties of THz nondestructive testing are:(1)the method of obtaining the refractive index of materials is not matu re enough;(2)The internal defects of the coating are difficult to detect,so it is difficult to inspect the terahertz detection results.Through theoretical analysis,experimental research and numerical simulation,this paper explores the coating imaging of steel structures and nondestructive testing technology of internal defects based on terahertz time-domain spectroscopy and neural network algorithm as data processing means.Based on the terahertz propagation theory,this paper provides a theoretical basis for the nondestructive testing of the coating thickness and defects of steel structures.It verifies the accuracy of the traditional method to obtain the refractive index and calculate the coating thickness and the neural network method to fit the coating thickness,respectively,and determines the superiority of the neural network over the traditional method;Based on the terahertz propagation theory,the excitation,propagation and reflection paths of terahertz are simulated in COMSOL,and compared with the terahertz reflection experiment,which proves the accuracy of electromagnetic simulation calculation;Through the simulation calculation of terahertz wave with different wavelengths,the optimal terahertz wavelength for steel structure coating is d etermined.The simulation results are compared with the terahertz scanning experiment.The simulation results are consistent with the terahertz scanning.The terahertz wave in the range of 2.5-3.5 THz frequency band is the most suitable band for non-destructive testing of steel structures.Finally,3 THz is determined as the terahertz wave for electromagnetic simulation calculation.Based on the terahertz propagation theory,this paper provides a theoretical basis for the nondestructive testing of the coati ng thickness and defects of steel structures.It verifies the accuracy of the traditional method to obtain the refractive index and calculate the coating thickness and the neural network method to fit the coating thickness,respectively,and determines the superiority of the neural network over the traditional method;Based on the terahertz propagation theory,the excitation,propagation and reflection paths of terahertz are simulated in COMSOL,and compared with the terahertz reflection experiment,which proves the accuracy of electromagnetic simulation calculation;Through the simulation calculation of terahertz wave with different wavelengths,the optimal terahertz wavelength for steel structure coating is determined.The simulation results are compared with the terahertz scanning experiment.The simulation results are consistent with the terahertz scanning.The terahertz wave in the range of 2.5-3.5 THz frequency band is the most suitable band for non-destructive testing of steel structures.Finally,3 TH z is determined as the terahertz wave for electromagnetic simulation calculation.In this paper,terahertz scanning equipment is used to conduct scanning experiments on single coating and composite coating respectively.The test data is processed through neural network model,and finally the thickness distribution of single coating and composite coating of steel structure is obtained.Compared with the detection method in the specification,the thickness detection of steel structure coating supported by terahertz technology and data processing means of neural network has high accuracy,The accuracy of single coating specimen can reach 96.5%,and that of composite coating can reach 96.9%;Through the scanning experiment of uneven coating,the thickness distri bution information of uneven coating is obtained.By introducing the error distance as the evaluation standard of coating uniformity,it is finally found that the coating with large error distance has poor uniformity,while the coating with small error dis tance has good uniformity.It is proved that the uniformity of steel structure coating can be evaluated using the error distance as a parameter.In this paper,the coating defects of steel structure are studied,and four kinds of defects(inclusion,holiday,debonding,bubble)of steel structure coating are summarized.According to the mechanism of coating defects,four kinds of coating defect specimens are made;By studying the internal characteristics of the four defects,COMSOL is used to simulate the fo ur defects,and 3600 point data are extracted from each simulation model to train the neural network model,with the minimum accuracy of more than 96%;The terahertz scanning experiment was carried out for four kinds of coating defects.6400 point data wer e selected as the test set data to test the neural network,and the minimum accuracy could reach95%;The coating defects of steel structures can be imaged by visualizing the coating thickness information predicted by the neural network model.更多还原