【作者】 黄靓；

【导师】 易伟建；

【作者基本信息】 湖南大学 ， 结构工程， 2008， 博士

【摘要】 混凝土结构由于设计、施工、荷载以及材料等方面的原因,在建造和使用过程中,不可避免地存在一些宏观损伤和缺陷,严重影响结构物的安全性和耐久性。新近引入的混凝土超声波层析成像技术,可用“图像”的方式定量化地反映层析面上混凝土的内部缺陷,是混凝土超声波探伤由无损检测阶段向无损诊断阶段发展的必然趋势。目前这一前沿课题尚处于研究的初步阶段,为此本文结合国家自然科学基金项目“混凝土超声波层析成像方法的基础研究”(No. 50678066),从混凝土超声波层析成像的理论基础、反演算法、反演策略、不完全射线反演和反演图像处理等五个方面进行较深入系统的研究,主要的工作内容如下:(1)广泛搜集整理国内外相关文献,扼要阐述层析成像数学基础,系统总结混凝土超声波层析成像的相关理论方法,为后续的研究工作奠定基础。(2)针对混凝土超声波层析成像的不适定性,提出了基于自然权的加权广义逆与杨文采正则化方法相结合的反演算法(NWGI算法),以适应混凝土超声波层析成像的特点。通过数值试验和混凝土试件试验,比较了本文NWGI算法与常规SIRT算法和CGLS算法的反演图像。对比结果表明,NWGI算法要优于常规算法,具有更高的分辨率和可靠性。通过2组共8个内部存在模拟缺陷的混凝土试件模型的成像试验,检验NWGI算法的有效性。成像结果表明,NWGI算法的反演结果能够较好地反映各个混凝土试件内部缺陷的位置和分布。(3)为获得与实际情况相吻合的真实解,研究了混凝土超声波层析成像的反演策略。①提出了定性反演方法,实现声阴影重叠法判别缺陷的计算机处理。数值模型和混凝土试件模型的试验表明,定性反演可以大致识别对象内部声速偏高、偏低和声速正常区域的位置和分布。②在获取定性反演先验信息的基础上,提出了两阶段反演的方法:第一阶段,区分优质像元和非优质像元,合理确定像元声速的迭代初值;第二阶段,区别对待优质像元和非优质像元的约束权重,定量反演最终的速度分布。2个混凝土试件的模型试验表明,两阶段反演能有效抑制伪像,具有比常规的一阶段反演更好的缺陷识别效果。③为了改善射线路径矩阵的数学性质,提出了自适应网格技术。通过合并那些只有少数射线甚至没有射线穿越的网格像元,使更新后的离散像元具有相对均衡的射线穿越长度,自主适应射线路径的几何分布,从根源上提高混凝土超声波层析成像的稳定性。数值试验表明,与常规的固定网格相比,基于自适应网格技术的层析成像表现出更好的抗噪能力和缺陷识别能力。(4)采用均匀试验设计方法布置不完全射线,研究了不完全射线布置的混凝土超声波层析成像的可行性。数值试验表明,在保证射线数量和射线代表性的前提下,不完全射线布置的层析成像可以达到甚至超过完全射线布置层析成像的缺陷识别效果。在不完全射线反演的基础上,还可以发展新的研究思路。对于布置不完全射线的试验设计方法进行了相关研究:①通过平衡深度搜索与广度搜索,改进了随机进化寻优算法,能够节省30～70%左右的计算时间。②提出了分步试验设计方法,通过化整为零地减小搜索空间,可比常规的一步试验设计方法节省40～60%左右的计算时间。对3个测试函数建立RBF神经网络替代模型,其误差分析表明,分步试验设计方法对于需要大规模试验样本点的替代模型是非常实用的。(5)研究了混凝土超声波层析成像的图像处理方法,增强缺陷识别能力。①采用中值滤波和基于SOFM神经网络的聚类分析进行反演图像的后处理。应用实例表明,这两种方法的联合使用可以有效地抑制伪像,凸显缺陷,提高层析图像的分辨率、可靠性和可读性。②采用基于小波变换的图像融合技术对不同反演参数下的层析图像进行信息融合。应用实例表明,图像融合可以有效综合被融合图像的信息,使融合后的层析图像具有较小的背景噪音和较好的缺陷识别效果。更多还原

【Abstract】 There exist inevitably some macroscopical injuries and defects in concrete structure in the process of building and using, because of the reasons in design, construction, loads and materials so on. These defects should affect the safety and durability of structures seriously. Ultrasonic computerized tomography for concrete introduced recently, might reflect the inner defects of concrete by images in tomography. That is to be inevitable development tendency in ultrasonic test for concrete from non-destructive test to non-destructive diagnosis. At present, the theoretical research in this field is still on its early stage. Under the supports of the program of National Natural Science Foundation“method study on ultrasonic computerized tomography for concrete”(No. 50678066), this dissertation has made a relatively deeply and systematically researches from several aspects of theoretical basis, inversion algorithm, inversion strategy, incomplete rays inversion and procession of inversion images. The main content is as follows.(1) The domestic and international literature concerned was extensively collected and summarized, and the mathematical basis of computerized tomography was described briefly. The relative theories and methods on ultrasonic computerized tomography for concrete were systematically summarized. These works have established a good basis for future research.(2) According to the ill-posed nature of ultrasonic computerized tomography for concrete, a inversion algorithm is put forward, which combined the weighted generalized inverses based nature weight matrix with regularization methods of Yang Wencai. The algorithm meets the characteristics of ultrasonic computerized tomography for concrete. The inversion images of NWGI algorithm, conventional SIRT algorithm and CGLS algorithm are compared in terms of numerical models and concrete specimen models. The results show that NWGI algorithm is more excellent than conventional algorithm and has higher resolution and reliability. The efficiency of this algorithm is verified through two groups with a total of 8 concrete specimens which simulated different inner defects. The images results indicate that NWGI algorithm and its inversion images can reflect the location and distribution of inner defects in concrete specimens very well.(3) In order to obtain the real solution coincident with practical case, the inversion strategy of ultrasonic computerized tomography for concrete was studied. ①Qualitative inversion method was proposed to realize computer processing of defect identification by acoustic shadow overlapping method. The results of numerical models and concrete specimen models show that qualitative inversion can identify the location and distribution of regions in which interior sound velocity is high, low or normal approximately.②Based on the prior information obtained by qualitative inversion, the two stages inversion method was put forward. That is, the first stage is that distinguishing the high-quality element and non- high-quality element, and determining the initial iteration value of element sound velocity reasonably. The second stage is that treating differentially the constraint weights of the high-quality element and non- high-quality element, and then quantitatively inversing the final sound velocity distribution. The results of two concrete specimens indicate that the two stages inversion method can inhibit artifacts effectively and has better defect identification effect than normal one stage inversion.③The self- adaptability grid technique was proposed to improve the mathematical property of ray path matrix. The grids which traversed by a few or few rays were merged, and then the updated grids had equilibrium length of ray traversing relatively. That might self-adapt the geometric distribution of ray paths and improve the stability of ultrasonic computerized tomography for concrete from origin. The results of numerical models show that the computerized tomography used self- adaptability grid technique has strong anti-noise ability and capability of defect identification compared with the conventional stationary grids.(4) Adopting the uniform experiment design distributes the incomplete rays, and the feasibility of ultrasonic computerized tomography for concrete by incomplete rays distribution was studied. The results of numerical models show that on the premise of ensuring quantity and representativeness of rays, the defect identification capability of the computerized tomography by incomplete rays distribution can reach or be superior to the computerized tomography by complete rays distribution. This has brought new thinking to the research of computerized tomography. The experiment design methods of distributing incomplete rays were researched as followed.①The stochastic evolution and optimizztion algorithm was improved by means of balancing depth search and width search. The calculation time can be saved about 30～70%.②The fractional steps experiment design method was proposed. With decreasing the search space by the " break up the whole into parts " method, the calculation time of this method can be saved about 40～60% compared with the conventional one step experiment design method. The substitution model of RBF neural network was made respectively by 3 different test functions. The error analyses indicate that the fractional steps experiment design method is very practical for the substitution model needed large scale test sample points.(5) The image processing method of ultrasonic computerized tomography for concrete was studied also.①The median filtering and clustering analysis based on SOFM neural network were introduced to the post-procession of inversion images. The application examples show that both of the two methods can inhibit artifacts effectively, explain the defects and improve the resolution, reliability and readability of tomography images.②Image fusion technology based on wavelet transform was used to fuse information of tomography images in different inversion parameters. The application examples show that this method can integrate information of fused images effectively and make the tomography images fused have less background noise and better defect identification effect.更多还原