分形粗糙面的参数识别、接触建模分析及预测

发布时间：2018-03-04 08:39

本文选题：WM模型　切入点：小波变换　出处：《西安电子科技大学》2015年硕士论文　论文类型：学位论文

【摘要】：微波器件的无源互调问题已经对许多通讯系统的正常工作造成影响,究其原因主要可分为电学元器件的材料非线性与接触非线性两类。其中,接触非线性是指电学元器件之间因微观接触表面粗糙而引起不充分接触所导致的电学非线性效应,例如电子隧穿效应等。研究微波器件接触非线性问题首先需要分析器件微观接触表面的接触情况。因此,本文针对微观粗糙面接触的问题,研究了基于分形的微观粗糙面建模及分析方法,分析了微观接触中的接触应力、实际接触面积等要素的变化规律。本文主要研究内容为:1.研究基于分形的粗糙面几何建模及相关分形参数的识别方法。首先,研究了Weierstrass-Mandelbrot模型的自仿射分形特性及其建模方法,并通过调整各项分形参数分析了它们对粗糙面模型形貌特征的影响。然后,研究了利用功率谱密度识别粗糙表面分形维数的方法。最后,根据功率谱密度法的原理提出了一种基于小波变换的粗糙面分形参数识别方法,该方法利用了小波变换的滤波能力及能量有限的特性,提取了轮廓曲线中不同频率波纹的信息,可同时识别分形维数与分形粗糙度,且识别精度较高。2.研究分形粗糙面接触分析方法及接触载荷、实际接触面积等的变化规律。首先,通过建立有限元模型分析了微波连接件波导法兰的接触面上的宏观接触应力分布。然后,通过建立微观粗糙面接触有限元模型,分析了不同加载条件下接触应力、实际接触面积等的变化规律,同时研究塑性变形对粗糙面接触的影响。最后,结合宏观下接触应力的分布规律以及微观下粗糙面接触载荷与接触面积的变化规律估算了波导法兰的实际接触面积。3.研究分形粗糙面实际接触面积的预测方法。首先,研究了几种典型支持向量回归算法及其核函数算法。经过对比分析,选择最小二乘支持向量机及高斯径向基核函数用于预测建模。然后,以分形维数、分形粗糙度和接触载荷为训练样本的输入,实际接触面积为输出,训练支持向量机。训练过程中支持向量回归算法的超参数由优化算法所确定,该优化算法将k倍交叉验证法所得泛化误差的估计值作为优化目标函数,通过耦合模拟退火算法及网格搜索算法结合的二段优化方法求解得到超参数的全局最优解。最后,通过支持向量回归算法实现了实际接触面积的预测。
[Abstract]:The passive intermodulation problem of microwave devices has affected the normal operation of many communication systems. The main causes can be divided into two categories: material nonlinearity and contact nonlinearity of electrical components. Contact nonlinearity refers to the electrical nonlinear effect caused by insufficient contact between electrical components due to the roughness of the micro contact surface. For example, electron tunneling effect and so on. In order to study the nonlinear contact problem of microwave devices, it is necessary to analyze the contact condition of the micro contact surface. Therefore, the problem of micro rough surface contact is discussed in this paper. The modeling and analysis method of micro rough surface based on fractal is studied, and the contact stress in micro contact is analyzed. In this paper, the geometric modeling of rough surface based on fractal and the recognition method of related fractal parameters are studied. Firstly, the self-affine fractal characteristics of Weierstrass-Mandelbrot model and its modeling method are studied. The influence of fractal parameters on the morphology of rough surface model is analyzed by adjusting the fractal parameters. Then, the method of identifying fractal dimension of rough surface by power spectral density is studied. According to the principle of power spectral density method, a method for identifying fractal parameters of rough surface based on wavelet transform is proposed. The wavelet transform's filtering ability and limited energy are used to extract the information of different frequency ripples in the contour curve. Fractal dimension and roughness can be recognized at the same time, and the recognition accuracy is high. 2. The contact analysis method of fractal rough surface and the change law of contact load and actual contact area are studied. The macroscopic contact stress distribution on the contact surface of the waveguide flange with microwave connection is analyzed by establishing the finite element model, and the contact stress under different loading conditions is analyzed by establishing the contact finite element model of the micro rough surface. At the same time, the influence of plastic deformation on the contact of rough surface is studied. Finally, The actual contact area of waveguide flange is estimated by combining the distribution of contact stress in macroscopic and the change of contact load and contact area of rough surface in microcosmic. The prediction method of actual contact area of fractal rough surface is studied. Several typical support vector regression algorithms and their kernel function algorithms are studied. After comparative analysis, least square support vector machine and Gao Si radial basis kernel function are selected for prediction modeling. Fractal roughness and contact load are the input of the training sample, the actual contact area is the output, and the training support vector machine. The super parameters of the support vector regression algorithm in the training process are determined by the optimization algorithm. In this optimization algorithm, the estimate of the generalization error obtained by the k-fold cross-validation method is taken as the optimization objective function, and the global optimal solution of the superparameter is obtained by using the two-stage optimization method combined with the coupled simulated annealing algorithm and the mesh search algorithm. The actual contact area is predicted by support vector regression algorithm.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN61

【参考文献】