基于改进LMD和PNN神经网络的通风机轴承故障诊断研究

发布时间：2018-06-02 10:51

本文选题：通风机轴承 + 特征频率　；参考：《中国矿业大学》2017年硕士论文

【摘要】：通风机是一种典型的机械设备,运行状态直接影响经济发展和日常生产。轴承作为维持通风机持续稳定旋转的关键零部件,对其进行状态监测和故障诊断研究具有非常重要的意义。论文以通风机轴承为研究对象,采集了正常、轻度内圈故障、重度内圈故障、轻度滚动体故障、重度滚动体故障、轻度外圈故障和重度外圈故障7种状态的振动信号,对信号特征提取与故障诊断分类等问题进行研究。论文介绍了局部均值分解(LMD)算法,通过对仿真信号分析,证明LMD在处理非平稳信号优于EMD和传统时频分析方法;针对LMD存在模态混叠的问题,引入总体局部均值分解(ELMD)算法;针对ELMD分解完备性差,采用改进补充总体局部均值分解(ICELMD),不仅解决模态混叠问题,同时具有较高的完备性;使用ICELMD对通风机轴承不同状态振动信号分解,并提取能量熵和峭度熵作为其特征值,为故障识别奠定了基础。最后,采用概率神经网络(PNN)辨识故障类型。针对PNN的模式层结构复杂,采用主元分析法(PCA)对输入样本降维;针对PNN的平滑因子σ难以确定,采用粒子群算法(PSO)对σ的优化,提高了分类精度;再针对PSO算法易陷入局部极值和收敛速度慢的缺点,分别采用惯性权重凹函数减小策略和适应度值稳定作为迭代终止条件的优化策略。实验结果表明,PCA和PSO优化的PNN既保证了较快的训练速度,又获得了更高的故障分类正确率。
[Abstract]:Ventilator is a kind of typical mechanical equipment, the running state directly affects the economic development and daily production. Bearing is the key component to maintain the steady rotation of ventilator. It is of great significance to study the condition monitoring and fault diagnosis of the bearing. The paper takes fan bearing as the research object and collects vibration signals in seven states: normal, mild inner ring fault, severe inner ring fault, mild rolling body fault, heavy rolling body fault, mild outer ring fault and heavy outer ring fault. The problems of signal feature extraction and fault diagnosis classification are studied. The local mean decomposition (LMD) algorithm is introduced in this paper. By analyzing the simulated signals, it is proved that LMD is superior to EMD and traditional time-frequency analysis method in dealing with non-stationary signals, and the total local mean decomposition (LMD) algorithm is introduced to solve the problem of modal aliasing in LMD. In view of the poor completeness of ELMD decomposition, the improved total local mean decomposition is used to solve not only the problem of modal aliasing, but also the high completeness, and the ICELMD is used to decompose the vibration signals of fan bearings in different states. Energy entropy and kurtosis entropy are extracted as eigenvalues, which lays a foundation for fault identification. Finally, probabilistic neural network (PNN) is used to identify fault types. In view of the complexity of the model layer structure of PNN, the principal component analysis method (PCA) is used to reduce the dimension of input samples, and the particle swarm optimization algorithm (PSO) is used to improve the classification accuracy because the smoothing factor 蟽 of PNN is difficult to determine. Aiming at the disadvantage of PSO algorithm which is easy to fall into local extremum and slow convergence rate, the inertial weight concave function reduction strategy and fitness stability are adopted as the optimization strategy of iterative termination condition respectively. The experimental results show that the PNN optimized by PCA and PSO can not only guarantee faster training speed, but also obtain higher accuracy rate of fault classification.
【学位授予单位】：中国矿业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TH43;TP183

【相似文献】