【作者】 王立军；

【导师】 王铁成；

【作者基本信息】 天津大学 ， 结构工程， 2008， 博士

【摘要】 随着混凝土强度无损检测技术的发展,回弹、超声、后装拔出法等单一测强手段已在工程检测中得到应用,相应的技术规程也相继颁布执行。由于对测强精度要求的逐步提高,单指标测强方法已不能满足工程需要,这就使综合法测强曲线的建立显得尤为重要。由于采用规程推荐的统一曲线受地方材料和各地施工水平差异等因素影响存在较大的误差,因此建立地区无损测强曲线是提高无损测强精度的主要措施。本文采用张家口地区常用的地方材料,对C15、C20、C25、C30、C35、C40六种常用不同强度等级的混凝土标养试件,进行28d、42d及56d三个龄期的超声、回弹及后装拔出无损检测试验,并同时进行混凝土破损测强。在此基础上用数理统计方法对试验数据进行回归分析,建立回弹-超声、超声-拔出、回弹-拔出、回弹-超声-拔出综合法测强曲线和相关公式。并利用自然养护混凝土试件无损检测试验结果对回归公式进行修正。经实验和部分工程实践验证,本文所建立的综合测强曲线相关性较好、误差较小。人工神经网络是理论化的人脑神经网络的数学模型,是基于模仿大脑神经网络结构和功能而建立的一种信息处理系统,能够实现复杂的逻辑操作和非线性关系。人工神经网络对于难以建立精确的数学模型而又易于收集学习样本的问题非常适合。影响混凝土强度无损检测的因素较多,具有许多不确定性,用一种确定的表达式来描述存在困难,鉴于此,本文作者提出了基于人工神经网络的混凝土强度无损检测综合法评定研究。探索并应用了神经网络的一些改进算法,其中包括附加冲量法,自适应学习算法及S型函数输出限幅算法等,以保证建立的神经网络的快速有效。对试验数据进行成功训练,利用Matlab语言编制相应程序,建立了混凝土强度无损检测综合法的人工神经网络模型。通过检测样本训练和实际工程验证,建立的人工神经网络模型推测的混凝土强度相对误差小,可以对混凝土强度进行预测。与传统的回归算法相比,采用人工神经网络建立的混凝土强度评定模型推测出的混凝土强度具有精度高的特点。更多还原

【Abstract】 With the development of the non-destructive testing techniques of concrete strength, the single testing strength means such as repercussion with hammer,ultrasound,embeded-pull-out, have widely been applied in the engineering inspection,and the relevant technical specification have also been issued and carried out.But with the improvement of the precision of testing strength,the single-target testing strength means cannot meet the engineering requirement that it appears especially important to set up testing strength curves with comprehensive methods.This paper introduced non-destructive testing experiments,that is,using ultrasound, repercussion and embeded-pull-out experiments to evaluate the strength for concrete standardized-tests of six types of strength grades ( C15, C20, C25, C30, C35, C40 )for 28 day,42day and 56days the three-age-period. The all materials used for tests are from Zhangjiakou area,which are widely applied in engineering projects. The related specimen concrete compressive strength experiments were performed in lab. On this base, the author made regression analysis on the tested data with mathematical statistical method.In addition , this thesis set up the testing strength curves of spring-back-ultrasound, ultrasound-pull-out , repercussion-pull-out , repercussion-ultrasound-pull-out comprehensive methods and relative formulas. Meanwhile,the author employed the results of the non-destructive testing of natural curing concrete tests to revise the regression formulas.By testing and verifying in some projects, the relativity of the comprehensive testing strength curves built in the thesis is better and the error is minor compared with common method. It is able to be promoted and applied in construction engineering and its surrounding area as the testing strength curves of the non-desstructive testing comprehensive methods of concrete strength.Artificial neural network (ANN) is a mathematical model of neural network of brain,and is a information handling system that simulate architecture and function of brain.ANN can realize complicated ligic operation and nonlinear relation,and it is very suitable for such problems that are difficult to be established,mathematical model but easy to be collected learning set.The non-destructive testing techniques of concrete strength is a very complicated prolem being influenced by many variable factors,and it is difficult to be expressed with a mathematical formula precisely.Therefore,the author put forword the research of the non-destructive testing techniques of concrete strength based on ANN.In order to make BP network more efficient,there are some optimized BP algorithms being put forword,including momentum Bp algorithms,variable learning rat BP algorithms and limited output of s-funcion algorithms.The mathematical model is set up by means of BP-artificial-neural-network technology based on considerable data from the experiment.Furthermore, corresponsive MATLAB program had been compiled to analyze the strength of concrete samples.更多还原