【作者】 张文；

【导师】 黄声享；

【作者基本信息】 武汉大学 ， 大地测量学与测量工程， 2020， 博士

【摘要】 目前,土石方填筑碾压施工质量管理主要采取控制碾压参数和试坑法检测的“双控”制,即主要依靠人工控制施工碾压工艺参数和人工现场挖坑取样检测等手段。然而,依靠监理和施工人员人为控制这些碾压参数,受人为因素干扰大,管理粗放,且最终的质量评定用抽检的方式来以点代面,无法准确反映整体的情况,故难于实现对压实参数的精准控制,难以确保碾压施工质量。因此,国内外学者利用现代先进的空间定位技术、物联网技术等技术,研发了一些填筑碾压施工质量监控系统,对填筑碾压施工过程进行实时监测和反馈控制,让碾压施工质量始终处于真实受控状态。但是,这些监控系统的研究主要集中在数据的获取和系统的集成方面,较少的深入去研究如何从获取的监控数据中更快速、更准确、更详细的计算出碾压质量控制参数信息。本论文旨在为填筑碾压施工向智能化、无人化的发展奠定理论与实践基础,系统深入地研究了碾压施工质量控制中的控制参数的计算问题:针对碾压施工中碾压遍数这一重要参数,提出了一种基于图像处理的填筑碾压遍数计算方法;考虑碾压仓面的实际特征和监控数据的特点,提出了一种顾及粗差的径向基函数神经网络曲面拟合方法,建立了高精度的碾压仓面数字高程模型;利用梯度提升回归算法建立了碾压参数和最终压实质量指标间的数学模型,实现了一种实时监控下的全仓面碾压施工质量评估方法。论文的主要工作和贡献如下:1)针对现有监控系统中碾压遍数计算方法的不足,提出了一种基于图像处理中Alpha混合算法的填筑碾压遍数计算方法。实验结果表明,该方法计算结果的精度和准确性较网格法更高;同时新方法的计算速度很快,为获得相同分辨率的碾压遍数计算结果,新方法的计算耗时约为网格法的1/50。2)为满足碾压层厚度及压沉值的计算需要,分析了高程数据在施工过程中覆盖更新的问题,提出了对监控测量数据进行预处理获取碾压仓面表面高程数据的方法;考虑到土石方碾压仓面表面的实际特征,提出了基于径向基函数神经网络的碾压仓面高程拟合方法,实验结果表明,基于径向基函数神经网络方法拟合的碾压仓面高程的外符合精度为±0.023m,优于多项式回归分析法、反距离加权法、克里金插值法等一些常用方法。3)针对径向基函数神经网络对训练样本的粗差较为敏感的特点,提出了基于高斯差分提取特征点的抗差径向基函数神经网络碾压仓面拟合方法。实验结果表明,含粗差时径向基函数神经网络拟合的中误差为±0.189m,远低于无粗差时仓面的拟合中误差,而抗差径向基函数神经网络算法的仓面拟合中误差为±0.027m,与无粗差时仓面的拟合中误差较为接近。4)提出了基于碾压仓面数字高程模型来解算仓面上各点的压沉值的方法。实验结果表明,虽然压沉值计算结果的精度不是很理想,但压沉值与干密度依旧表现了相关性,相关系数为-0.54,压沉值越小,干密度越大,表明压实质量越好。5)为解决以离散试坑检测样本来进行仓面压实质量评估导致的准确性问题,经过实验对比分析,建立了基于梯度提升回归算法的碾压参数和最终压实质量指标间的数学模型,在此基础上,进一步提出了实时监控下的全仓面碾压施工质量的评估方法,为碾压监控系统实现“最终参数控制”提供了解决方案。6)结合本文的研究成果,研发了基于北斗的路基碾压施工质量控制系统和基于测量机器人的大坝填筑碾压施工质量监控系统,并在实际工程中得到成功应用。更多还原

【Abstract】 At present,the quality control methods of “dual controls” are mainly adopted in the quality management of earth-rock filling and rolling construction,of which one is manually controlling the roller compaction parameters including rolling times and driving speed of compaction machines,thickness of filling layer and smoothness of storehouse surface,the other is inspecting the test holes sampled manually in the working surface.However,the conventional method is difficult to ensure construction quality,because it is hard to accurately control these roller compaction parameters,interfered by human factors and extensive managements.Therefore,domestic and foreign scholars have used modern advanced spatial positioning technology,Internet of Things technology and other technologies to develop some construction quality supervisory systems for filling and rolling,which can supervise the filling and rolling construction process and feedback control.So that,the quality of rolling construction is always in a real controlled state.However,the research of these supervisory systems is mainly focused on data acquisition and system integration,and there is less in-depth research on how to calculate the rolling quality control parameter information from the acquired monitoring data more quickly,accurately and in more detail.This thesis aims to lay a theoretical and practical foundation for the evolution of filling and rolling construction to intelligent and unmanned,and systematically and deeply study the calculation of control parameters in the quality control of rolling construction.Aiming at rolling times,which is the important parameter in rolling construction,a fast calculation method of rolling times based on image processing is proposed.Considering the actual characteristics of the roller compaction surface and the characteristics of the monitoring data,a radial basis function neural network surface fitting method considering the gross error is proposed,and a high-precision digital elevation model of the roller compaction surface is established.The mathematical model between rolling parameters and compacting quality is established by using gradient boosting regression algorithm,and a evaluation method of rolling construction quality of the whole warehouse surface is realized.The main work and contributions of this thesis are as follows:(1)In order to overcome the shortcomings of the existing calculation methods of rolling times,a new calculation method of rolling times based on the Alpha hybrid algorithm is put forward.The experimental results show that the accuracy and accuracy of the calculation results of the new method are higher than that of the grid method.At the same time,the calculation speed of the new method is very fast.In order to obtain the calculation results of rolling times with the same resolution,the calculation time of the new method is only about 1/50 of that of the grid method.(2)In order to meet the calculation needs of roller compaction layer thickness and compaction value,the problem of coverage and update of elevation data during the construction process is analyzed,and a method for preprocessing monitoring measurement data to obtain surface elevation data of roller compaction is proposed.Considering the actual characteristics of the roller compaction surface,a radial basis function neural network surface fitting method is proposed.The experimental results show that the external coincidence accuracy of the roller compaction surface elevation fitted by the radial basis function neural network method is ±0.023 m,which is better than some common methods such as polynomial regression analysis method,inverse distance weighting method,and Kriging interpolation method.(3)Aiming at the characteristic that the radial basis function neural network is more sensitive to the gross errors of the training sample points,a robust radial basis function neural network fitting method based on the Gaussian difference extracting feature points is proposed.The experimental results show that the fitting mean error by radial basis function neural network when the data contains gross error is ±0.189 m,which is much lower than that the data does not contain gross errors..At the same time,the fitting mean error by the robust radial basis function neural network algorithm is ±0.027 m,which is close to that when the data does not contain gross errors.(4)A method based on the digital elevation model of the roller compaction surface to calculate the compaction value of each point on the surface is proposed.The experimental results show that although the precision of the calculation results is not very ideal,the correlation coefficient between the compaction value and dry density is-0.54.The smaller the settlement value,the higher the dry density,indicating that the better the compaction quality.(5)In order to solve the accuracy problem caused by using a limited number of discrete test pit test samples to evaluate the quality of silo surface compaction,after experimental comparison and analysis,a mathematical model between the rolling parameters and the compaction quality indicators based on the gradient boosting regression algorithm was established.On this basis,a evaluation method of rolling construction quality of the whole warehouse surface is further proposed,which provides a solution for the rolling supervisory system to achieve "final parameter control".(6)Based on the research results of this paper,a series of software for dam filling and rolling construction quality monitoring and a series of software for highway subgrade rolling quality monitoring have been developed and successfully applied.更多还原