水泥加固红土的力学特性及神经网络模型研究

【作者】 杨玉婷；

【导师】 黄英；

【作者基本信息】 昆明理工大学 ， 水利水电工程， 2010， 硕士

【摘要】 本文在综合分析国内外关于土体加固及加固机理、微结构、神经网络模型应用等研究现状的基础上,针对存在的问题并结合云南广泛应用红土的力学特性不能完全满足工程要求的情况,提出了“水泥加固红土的力学特性及神经网络模型研究”的问题。针对云南典型红土,选用水泥作为强固剂,通过对水泥加固红土宏微观试验的对比研究,明确了水泥加固红土的力学特性及微结构特征；运用图像处理技术,提取分析了水泥加固红土的微结构特征参数；结合水泥加固红土的力学特性、微结构特征及特征参数,从微结构的角度阐明了水泥加固红土的加固机理；最后运用神经网络理论,建立起水泥加固红土抗剪强度的神经网络模型。水泥加固红土的击实、抗剪强度、压缩、渗透等各个力学特性是在考虑水泥加入比例和试样养护时间的基础上,通过击实、直剪、压缩、渗透等宏观试验进行对比分析来确定。宏观试验结果表明：水泥加固红土的最大干密度增大、最优含水率减小,抗剪强度及抗剪强度指标增大,压缩系数和渗透系数减小。随水泥加入比例的增大和试样养护时间的延长,抗剪强度及抗剪强度指标逐渐增大,压缩系数和渗透系数逐渐减小,但抗剪强度、抗剪强度指标增大的程度和压缩系数、渗透系数减小的程度都逐渐变缓,粘聚力增大的程度大于内摩擦角增大的程度。水泥加入比例对红土各个力学特性的影响程度大于试样养护时间对红土各个力学特性的影响程度。水泥加固红土的微结构特征可以通过扫描电镜观察对土样击实前后、加固前后、养护前后、剪切前后及压缩前后等不同情况下获得的不同放大倍数下的微结构图像进行对比分析提炼。扫描电镜试验表明：水泥加固红土具有密实性、胶结性、填充性、包裹性、孔隙性等微结构特征；对水泥加固红土的微结构图像进行数字化处理,提取了水泥加固红土的孔隙率和颗粒率等微结构特征参数。提取结果表明,经过击实、加固、养护及压缩后,水泥加固红土的孔隙率减小、颗粒率增大；随水泥加入比例的增大和试样养护时间的延长,水泥加固红土孔隙率逐渐减小、颗粒率逐渐增大；剪切后,孔隙率增大、颗粒率减小。水泥加固红土力学特性的变化实质上在于其微结构的变化,而红土微结构的变化又取决于水泥与红土颗粒之间的相互作用。将水泥加固红土的力学特性、微结构特征及特征参数结合起来,通过水泥加固红土所体现出来的胶结作用、包裹作用和填充作用,从微结构的角度来解释水泥加固红土的机理。水泥加固红土力学特性的变化正是以上三种作用综合影响的结果。水泥加固红土最大干密度的增大和最优含水率的减小主要取决于包裹作用和填充作用；水泥加固红土抗剪强度和抗剪强度指标的增大及压缩性的减小主要取决于胶结作用；而水泥加固红土渗透性的减弱主要是由包裹作用和填充作用引起。根据水泥不同加入比例、试样不同养护时间所获得的直剪试验数据,运用神经网络理论建立了神经网络模型。模型的输入层向量确定为水泥加入比例和试样养护时间两个影响因素,模型的输出层向量确定为粘聚力和内摩擦角两个抗剪强度指标,并对选定的样本数据进行归一化处理；通过试算,模型的隐层传递函数确定为正切函数tansig,模型的输出层传递函数确定为对数函数logsig,模型的隐层神经元数确定为5。根据模型构建层次,建立起水泥加固红土抗剪强度的神经网络模型,模型预测结果总体上令人满意。更多还原

【Abstract】 This paper based on the problem of comprehensively analysis of soil reinforce、reinforcement mechanism、micro- structure、and neural network model, combining with the mechanical characteristics of laterite which is widely used in Yunnan can’t completely satisfied engineering requirements, the problem "study on mechanical characteristics and neural network model of laterite-cement" was presented. Aiming at the typical laterite in Yunnan, this paper choosed cement as reinforced agent, and confirmed the mechanical characteristics and the microstructure characteristics of the laterite-cement through comparative study of the macro and micro test; Extracted the microstructure characteristic parameters of the laterite-cement by using image processing techniques; Illuminated that reinforcement mechanism of laterite-cement from the perspective of microstructure, combining with mechanical characteristics, microstructure characteristics and microstructure characteristic parameters; At last, the neural network model of shear strength of the laterite-cement was established by using neural network theory.The mechanical characteristics of compaction、shear strength compression and permeability of the laterite-cement was confirmed through the contrastive analysis of compaction test、shear strength test、compression test and penetproportionn test, considering the cement proportion and the sample curing time. The macroscopic experiment result indicated:For the laterite-cement, the maximum dry density increased and the optimum moisture content decreased, the shearing strength and it’s parameters increased, the compressibility coefficient and the permeability coefficient decreased. With increasing the cement proportion and extended the curing time, shear strength and it’s parameters increased gradually, the compressibility coefficient and the permeability coefficient decreased gradually, but the extent will slow down gradually, the increased extend of cohesion force is greater than the extend of internal friction angle. For all kinds of mechanics feature of laterite, the incidence of cement proportion is greater than the incidence of the sample curing time.The microstructure characteristic of the laterite-cement will be obtained by SEM, measuring and analyzing the microstructure images of different magnification factor before and after the condition of compaction reinforcement、curing、shear and compression. The SEM experiment result indicated:the the laterite-cement have the characteristic microstructure of close-grained、agglutinating、filling、enwrapping and porosity; Extracted the porosity rate and granular rate atc microstructure characteristic parameter through digitize image digital processing the microstructure of reinforced laterite. The extract result indicated:For the laterite-cement, the porosity rate decreased and the granular rate increased after compaction、reinforcement、curing and compression; With increasing the cement proportion and extended the curing time, the porosity rate decreased gradually, the granular rate decreased gradually; After shearing, the porosity rate increased and the granular rate decreased.For the laterite-cement, the change of mechanical characteristics rest with the change of microstructure, and the change of microstructure rest with the interaction between cement and laterite granule. The reinforcement mechanism of laterite-cement was explained from the perspective of microstructure combining with mechanical characteristics, microstructure characteristics and microstructure characteristic parameters and through agglutinating、filling、enwrapping, the change of mechanical characteristics is the result of the combined influence of the three effects. For the laterite-cement, the increase of maximum dry density and the decrease of optimum moisture content rest with enwrapping and filling; the increase of the shearing strength and it’s indicators and the decrease of compressibility rest with agglutinating; The decrease of permeability rest with enwrapping and filling.Using the neural network theory, this paper established the neural network model according to the direct shear test’s data which obtained through control cement different proportion and the sample different curing time. Make sure the cement proportion and curing time two factors as the input layer’s vector quantity of model, make sure the cohesion force and internal friction angle two shear strength parameters as the output layer’s vector quantity of model, and to make unitary processing for the selected specimen. The transfer function of hidden layer of the model is tangent function tansig and the transfer function of output layer of the model is log function logsin, the number of the hidden layer neuron of the model is 5. According to the arrangement of the model, the neural network model of shear strength of the laterite-cement is established, on the whole, the model’s forecast result turn up trumps.更多还原