基于反向建模的细胞自动机水泥水化仿真方法研究

发布时间：2018-10-18 19:29

【摘要】：水泥在工业领域应用广泛,是世界基础设施建设的支柱。水泥的微观结构直接影响其强度,因此研究水泥水化的微观演化过程对提高水泥的性能具有非常重要的意义。由于水泥水化的极端复杂性,人们至今没能完全理解水泥水化的过程。传统的物理化学方法在研究水泥水化的过程中根据实验测定的样品指标,采用人工定义的水化规则正向推导进行建模,方法复杂且建立的模型准确度低。随着计算材料学的发展,采用计算机模拟水泥水化反应过程已成为材料科学发展的热点。本文采用反向建模的思想,建立了水泥水化三维微观结构演化的细胞自动机模型,从观测到的数据中采用多种方式提取出水化规则,并对不同模型的实验结果进行了模拟对比。本文主要从以下几个方面对水泥水化微观结构的演变过程模拟进行了研究分析:(1)特征提取及数据预处理提出了基于微米级显微断层扫描图像(μCT)的水泥三维颗粒特征提取方法。由于CT技术在扫描样本的过程中不会对样本本身造成太大的损害,因此能够保证样本数据的真实性。同时,μCT像素的灰度值能够代表的是该区域的平均线性衰减系数,该系数和水泥图像中的物相组成相关,这就能够保证μCT得到的图像像素的灰度值与相应位置所表示的物质组成存在很好的相关性。将二维水泥图片序列合成三维水泥颗粒,通过三维空间的位置确定周围26个像素点为邻域,基于邻域提取到和邻域相关的特征数据。该数据进行处理后可以作为各种模型的原始输入输出数据。(2)多基因表达式与粒子群算法混合编程为了适应水泥水化数据的特性,论文提出了采用多基因表达式编程(MEP)优化函数结构,粒子群算法(PSO)优化函数系数的MEP-PSO混合编程算法。采用该算法对UCI数据集中的回归问题进行了测试,并与其余常用算法进行了对比,表明该混合编程方法在解决一般函数发现问题中也有很好的表现。(3)水泥微观结构演化过程反向建模模型分析论文最后采用多种反向建模方法构建水泥水化微观结构的细胞自动机模型,对模型的原理结构以及实现进行了详细的分析,并对仿真结果的误差进行了说明。实验表明采用MEP-PSO混合编程的方法建立的模型在所有测试数据集上都能有很好的预测结果。本文同时采用不同水泥样本建立的模型在其余水泥样本上进行测试也取得了很好的表现,说明该方法建立的模型有着很好的泛化能力。
[Abstract]:Cement is widely used in the field of industry and is the pillar of infrastructure construction in the world. The microstructure of cement directly affects its strength, so it is very important to study the evolution process of cement hydration to improve the performance of cement. Due to the extreme complexity of cement hydration, the process of cement hydration has not been fully understood. In the process of studying cement hydration, the traditional physical and chemical method is based on the sample index determined by experiment, and the artificial hydration rule is used to model the model. The method is complex and the accuracy of the model is low. With the development of computational materials, computer simulation of cement hydration process has become a hot spot in material science. In this paper, the cellular automata model of the 3D microstructure evolution of cement hydration is established by using the idea of reverse modeling. The hydration rules are extracted from the observed data in many ways, and the experimental results of different models are simulated and compared. In this paper, the evolution process of cement hydration microstructure is studied and analyzed from the following aspects: (1) feature extraction and data preprocessing method based on 渭 CT is proposed. Because the CT technique can not damage the sample itself in the process of scanning the sample, it can guarantee the authenticity of the sample data. At the same time, the gray value of 渭 CT pixel can represent the average linear attenuation coefficient of the region, which is related to the composition of the phase in the cement image. This can ensure that there is a good correlation between the gray value of the image pixels obtained by 渭 CT and the material composition represented by the corresponding position. The 2D cement image sequence was used to synthesize the 3D cement particles, and the surrounding 26 pixels were determined as neighborhood by the position of 3D space, and the feature data related to the neighborhood were extracted based on the neighborhood. After processing, the data can be used as the original input and output data of various models. (2) the multi-gene expression and particle swarm optimization algorithm are mixed to program to adapt to the characteristics of cement hydration data. In this paper, a hybrid MEP-PSO programming algorithm using multi-gene expression to optimize (MEP) function structure and particle swarm optimization algorithm (PSO) to optimize function coefficients is proposed. This algorithm is used to test the regression problem in UCI dataset, and compared with other common algorithms. The results show that the hybrid programming method has a good performance in solving the general function discovery problem. (3) the reverse modeling model of cement microstructure evolution process is analyzed in this paper. Finally, several reverse modeling methods are used to construct cement hydration microjunction. Cellular automata model, The principle structure and realization of the model are analyzed in detail, and the error of simulation results is explained. The experimental results show that the model established by using MEP-PSO hybrid programming method has good prediction results on all test data sets. At the same time, the model established by different cement samples has been tested on the other cement samples, which shows that the model established by this method has good generalization ability.
【学位授予单位】：济南大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TQ172.1;TP18

【相似文献】