基于马尔可夫随机场的水泥水化建模及性能预测

发布时间：2018-04-23 08:42

本文选题：水泥建模 + 马尔可夫随机场模型　；参考：《济南大学》2017年硕士论文

【摘要】：由于水泥作为一种重要的工业基材料被广泛应用到生产生活的各个方面,所以促使人们不断研究探索其水化机理。但是由于水泥水化内部反应的极端复杂性,直到目前科学家也没有完全搞清楚内部的反应原理和反应过程,而且传统的分析方法需要消耗很多的时间也不具有时间连续性。随着计算机科学技术研究的发展使得对水泥水化的研究进入计算材料时代,虽然之后的研究也取得了一些公认的成果,比如Bentz的细胞自动机模型。但是一般这种建模都是建立在像素与特征的基础之上的,在建模之前必须进行复杂的水泥图像配准等操作,而且一个极小的像素偏差都可能对建模结果产生比较大的影响,不仅对设备要求很高,而且工作计算量都很大。这里我们主要根据马尔可夫随机场在图像中的广泛应用为基础,把它应用到水泥水化建模中来,由于利用马尔可夫随机场进行水泥水化建模只用考虑水泥微观图像的邻域概率分布特征,不用事先进行复杂的图像配准和特征选择,所以提高了准确度和易用性。本文主要分下面几个方面来进行研究分析。(1)适应水泥微观图像的马尔可夫模型参数估计方法我们提出了一个新的马尔可夫模型参数估计方法,加权最小平方差适应方法(WLS)。马尔可夫模型的参数估计问题对使用者一直都是非常大的挑战,目前常用的一些方法都存在一定的问题,为了使马尔可夫模型可以顺利地用在水泥水化建模中,需要一个精确度高、时间复杂度低又具有噪声鲁棒性的参数估计方法。我们提出WLS方法是一个完整的体系,其中包括主要的WLS参数估计方法、零值处理方法和参数适应值评价方法。并通过实验对比表明我们的WLS方法比最小平方差方法(LS)最有更高的精确度和噪声鲁棒性。(2)马尔可夫模型水泥水化建模由于这是第一次把马尔可夫模型应用到水泥水化中来,我们进行了可行性分析。提出了一个基于采样图像的水泥水化相似度适应值函数,解决了水泥水化马尔可夫建模的最大难点,然后通过与粒子群算法(PSO)优化结合起来,方便直观地建立水泥水化图像的马尔可夫模型。(3)建立水泥水化马尔可夫模型之后的后续利用论文最后给出了建立水泥水化马尔可夫模型之后广泛的应用前景和研究方法。可以通过训练人工神经网络模拟水泥水化马尔可夫模型参数的变化,然后根据模型参数的变化可以通过采样看出来水泥微观结构的变化,进而预测水泥水化过程。而且也为探究马尔可夫模型参数与水泥各项性能的关系提供了可能。
[Abstract]:As an important industrial base material, cement has been widely used in various aspects of production and life, so people are constantly studying and exploring its hydration mechanism. However, due to the extreme complexity of the internal reaction of cement hydration, up to now, scientists have not fully understood the internal reaction principle and reaction process, and the traditional analytical methods need to consume a lot of time and do not have time continuity. With the development of computer science and technology, the study of cement hydration has entered the era of computational materials, although the later research has also made some recognized achievements, such as Bentz's cellular automata model. However, this kind of modeling is generally based on pixels and features. Before modeling, complex operations such as cement image registration must be carried out, and a minimal pixel deviation may have a relatively large impact on the modeling results. Not only high requirements for equipment, but also a lot of work calculation. Here we apply Markov random fields to cement hydration modeling, based on the widespread application of Markov random fields in images. Because the cement hydration modeling based on Markov random field only takes into account the neighborhood probability distribution features of cement microscopic images and does not need to carry out complicated image registration and feature selection in advance the accuracy and ease of use are improved. In this paper, we study and analyze the parameter estimation method of Markov model which adapts to cement microscopic image. We propose a new Markov model parameter estimation method, weighted least square difference adaptation method. The problem of parameter estimation of Markov model is always a great challenge to the user. Some commonly used methods have some problems. In order to make Markov model can be used in cement hydration modeling smoothly. A parameter estimation method with high accuracy, low time complexity and noise robustness is needed. We propose that the WLS method is a complete system, including the main WLS parameter estimation method, zero value processing method and parameter fitness evaluation method. The experimental results show that our WLS method has higher accuracy and noise robustness than the least square difference method (LSs) for cement hydration modeling based on Markov model because it is the first time to apply Markov model to cement hydration. We carried out a feasibility analysis. A fitness function of cement hydration similarity based on sampling image is proposed, which solves the biggest difficulty of cement hydration Markov modeling, and then combines with particle swarm optimization (PSO) optimization. It is convenient and intuitionistic to establish the Markov model of cement hydration image. (3) the subsequent utilization after the establishment of cement hydration Markov model. Finally, the paper gives the extensive application prospect and research method after the establishment of cement hydration Markov model. The parameters of cement hydration Markov model can be simulated by training artificial neural network. According to the change of model parameters, the change of cement microstructure can be seen by sampling, and the cement hydration process can be predicted. It is also possible to explore the relationship between the parameters of Markov model and the properties of cement.
【学位授予单位】：济南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TQ172.1;TP391.41

【参考文献】