基于遗传神经网络的故障诊断算法研究

发布时间：2018-06-14 21:34

本文选题：故障诊断 + 免疫遗传算法　；参考：《辽宁大学》2012年硕士论文

【摘要】：大规模、高精度的现代化机械设备的各部件之间的联系、耦合相当紧密，当某一部件发生故障，整台设备、甚至是整条生产线都将受到影响。滚动轴承作为机械设备中最常见的零部件之一，其运行状态又直接影响到整台机器的性能。在对滚动轴承的诊断过程中，其诊断模式与特征向量之间是非常复杂的非线性关系，用振动信号的时域、频域分析方法很难全面地反映。人工神经网络在故障诊断领域显示出巨大的应用潜力。本文以精密电机轴承故障诊断问题的研究为背景，通过对其故障特征的分析，提取适合的特征参数，利用神经网络的强非线性映射能力和遗传算法的全局优化能力建立诊断模型，探讨了其在电机轴承故障诊断中的应用。在阅读分析了大量关于数据驱动的故障诊断方法的基础上，本文提出一种对遗传神经网络的故障诊断算法的改进：为了避免进化初期超个体的误导作用，对适应度函数做了调整；采用适合于浮点数编码方式的算术交叉策略和非均匀变异策略，并增添交叉次数和变异次数，以加快模型的收敛速度。又提出一种改进的免疫遗传神经网络故障诊断方法，以“自我调节”机制为理论基础，提出一种基于抗体期望繁殖率的自适应交叉变异概率调整方法，使与抗体自身息息相关。同时又受到精英策略的启发，对算术交叉算法中的组合系数做相应改善，使改进后的算术交叉算法，，更加偏向于适应度较高的个体进行线性运算。重新定义了基于欧氏距离的相似度的计算方法，提出一种“3σ”计算方法，使得对门限的设置更简单直观。在MATLAB平台下进行仿真实验，实验结果表明，提出的二阶段数据预处理方法是有效可行的；门限值简单化的设定提高了该方法的泛化能力，建立的遗传神经网络（GA-BP）故障诊断算法和免疫遗传神经网络（IGA-BP）故障诊断算法具有更快的收敛速度，更强的自适应性，对故障类型具有较为准确的分类能力，诊断效果良好。
[Abstract]:The coupling between the components of a large scale and high precision modern mechanical equipment is very close. When a component fails, the whole equipment, even the whole production line, will be affected. As one of the most common parts in mechanical equipment, rolling bearing has a direct impact on the performance of the whole machine. In the diagnosis process of rolling bearing, the nonlinear relationship between the diagnostic mode and the eigenvector is very complex, so it is difficult to reflect the vibration signal in time domain and frequency domain analysis method. Artificial neural network (Ann) has shown great application potential in the field of fault diagnosis. In this paper, based on the research of bearing fault diagnosis of precision motor, through the analysis of its fault characteristics, the suitable characteristic parameters are extracted. Based on the strong nonlinear mapping ability of neural network and the global optimization ability of genetic algorithm, a diagnosis model is established, and its application in motor bearing fault diagnosis is discussed. On the basis of reading and analyzing a large number of data-driven fault diagnosis methods, this paper proposes an improved fault diagnosis algorithm for genetic neural networks: in order to avoid the misguided effect of superindividuals in the early stages of evolution, The fitness function is adjusted and the arithmetic crossover strategy and non-uniform mutation strategy suitable for floating-point coding are adopted and the crossover times and mutation times are added to speed up the convergence of the model. An improved immune genetic neural network fault diagnosis method is proposed. Based on the "self-regulation" mechanism, an adaptive crossover mutation probability adjustment method based on the expected reproduction rate of antibody is proposed, which is closely related to the antibody itself. At the same time, inspired by the elite strategy, the combination coefficients in the arithmetic crossover algorithm are improved accordingly, so that the improved arithmetic crossover algorithm is more inclined to the individual with higher fitness for linear operation. This paper redefines the similarity calculation method based on Euclidean distance, and proposes a "3 蟽" calculation method, which makes the setting of threshold more simple and intuitive. The simulation results on MATLAB platform show that the proposed two-stage data preprocessing method is effective and feasible, and the simplified threshold value improves the generalization ability of the method. The genetic neural network fault diagnosis algorithm and the immune genetic neural network IGA-BP-based fault diagnosis algorithm have faster convergence speed, stronger self-adaptability, more accurate classification ability and good diagnosis effect.
【学位授予单位】：辽宁大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TH165.3;TP18

【参考文献】