碳纤维凝固过程的动态数据建模与优化

发布时间：2018-02-09 22:52

本文关键词： 碳纤维凝固过程动态数据建模免疫优化滑动窗口多核支持向量机　出处：《东华大学》2017年硕士论文　论文类型：学位论文

【摘要】：碳纤维复合材料是一种力学性能优异的新材料,它具有强度高,模量大,密度小等特点,同时还具有较高的比强度和很高的比模量。由于碳纤维优异的性能与广泛的应用,其纺丝凝固过程也便备受国内外学者的关注。目前国内外对碳纤维凝固过程的研究大多停留在机理建模层面,采用数据建模的相对较少,尤其对凝固过程中的动态变化进行动态数据建模更是少有涉及。近几年,随着统计学习与机器学习的兴起,如何利用数据,建立更加高效的模型成为非常必要的研究课题。本文利用碳纤维凝固过程中多项指标的动态数据,充分结合纤维凝固成形的机理,设计了高效、准确的动态数据模型,进行准确、稳定的凝固过程浓度预测,从而可以实时地帮助提高碳纤维原丝生产的产品性能。本论文的主要贡献如下:(1)分析了碳纤维凝固过程中的各项参数指标的数据特征以及它们之间的相关关系,针对原丝内部溶剂浓度变化的数据特征多样性问题,提出了一种多核支持向量机数据模型(MKSVM)。相比于采用单核的传统支持向量机模型,MKSVM更加适用于具有多样数据特征的纤维内部溶剂浓度变化。仿真对比实验结果进一步证明所建立的模型的准确性和优越性。(2)碳纤维的凝固过程是一个随时间连续变化的动态过程,本论文对凝固过程中的各项动态数据进行聚类分析,利用聚类结果训练出核矩阵切换机(KSM),核矩阵切换机可以在每次窗口滑动时候,根据当前载入到窗口的数据,判断该组数据的类别,然后将核矩阵切换到最佳,这里提到的核矩阵是MKSVM利用类别中心的数据训练而得。接着,引入滑动窗口理论,提出了一种基于滑动窗口多核支持向量机的动态数据模型(SWMKSVM)。实验结果表明,在凝固过程的动态变化中,该模型仍然保持了令人满意的准确性。(3)为了使得动态模型在整个动态过程中的效果达到最优,本论文使用免疫算法优化SWMKSVM的参数,提出了一种免疫滑动优化多核支持向量机模型(ISAMKSVM),利用免疫算法的寻优机制,确保快速而有效地寻找到全局最优解,保证了滑窗在整个滑动过程中的整体最优,而不仅仅是是每次滑动的最优,并且相比采用传统方式优化核权重系数,可以明显提高效率。经实验仿真测试表明,利用ISAMKSVM的动态数据模型时,与传统粒子群优化参数方法的模型相比较,该模型能够快速而有效地寻找到最优参数,而且算法稳定性强,在模型准确率方面有明显的优势。最后,对全论文的研究工作进行了总结,指出了工作中存在的不足,并对有待进一步研究的方向和方法进行了展望。
[Abstract]:Carbon fiber composite is a new material with excellent mechanical properties. It has the characteristics of high strength, high modulus and low density, and also has high specific strength and high specific modulus. At present, most of the researches on the solidification process of carbon fiber stay at the level of mechanism modeling, and the data modeling is relatively few. In recent years, with the rise of statistical learning and machine learning, how to use data, It is necessary to establish a more efficient model. Based on the dynamic data of many indexes in the process of carbon fiber solidification and the mechanism of fiber solidification forming, an efficient and accurate dynamic data model is designed in this paper. Accurate and stable concentration prediction of solidification process, The main contribution of this paper is as follows: 1) the main contribution of this paper is to analyze the data characteristics of the parameters during the solidification process of carbon fiber and the correlation between them. Aiming at the diversity of the data characteristics of solvent concentration change in the filament, A multi-core support vector machine data model (MKSVMN) is proposed. Compared with the traditional support vector machine (SVM) model with single core, MKSVM is more suitable for the change of solvent concentration in fibers with various data characteristics. The simulation results are compared with the experimental results. It is proved that the solidification process of the carbon fiber is a dynamic process with continuous change over time. In this paper, the dynamic data in solidification process are analyzed, and the kernel matrix switching machine is trained by clustering results. The kernel matrix switching machine can load the data to the window according to the data loaded into the window every time the window slips. Determine the category of the group of data, then switch the kernel matrix to the best. The kernel matrix mentioned here is trained by MKSVM using the data from the class center. Then, the sliding window theory is introduced. A dynamic data model based on sliding window multi-core support vector machine (SVM) is proposed. The experimental results show that, during the dynamic change of solidification process, In order to optimize the effect of the dynamic model in the whole dynamic process, the immune algorithm is used to optimize the parameters of SWMKSVM. An immune sliding optimization multi-kernel support vector machine model (ISAMKSVMM) is proposed. By using the optimization mechanism of the immune algorithm, the global optimal solution is found quickly and effectively, and the overall optimum of the sliding window in the whole sliding process is ensured. It is not only the optimum of each sliding, but also the efficiency can be improved obviously compared with the traditional way of optimizing the kernel weight coefficient. The experimental simulation results show that, when using the dynamic data model of ISAMKSVM, Compared with the traditional Particle Swarm Optimization (PSO) model, the model can find the optimal parameters quickly and effectively, and the algorithm is stable and has obvious advantages in the accuracy of the model. This paper summarizes the research work of the whole paper, points out the shortcomings of the work, and looks forward to the direction and methods to be further studied.
【学位授予单位】：东华大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TQ342.742;TB332

【参考文献】