小波包分解与神经网络风机齿轮箱故障诊断  

Fault Diagnosis of Fan Gearbox Based on Wavelet Packet Decomposition and Neural Network

作　　者：刘谨言 买买提热依木·阿布力孜[1] 项志成 谢丽蓉[1] LIU Jin-yan;Maimaitireyimu Abulizi;XIANG Zhi-cheng;XIE Li-rong(College of Electrical Engineering,Xinjiang University,Urumqi Xinjiang 830017,China)

机构地区：[1]新疆大学电气工程学院,新疆乌鲁木齐830017

出　　处：《计算机仿真》2023年第10期123-129,共7页Computer Simulation

基　　金：国家自然科学基金项目(62163034)。

摘　　要：齿轮箱作为风机传动系统重要的机械部件长期工作在恶劣环境中易于发生故障。针对风机齿轮箱产生的振动信号具有非平稳、瞬态、不确定性、以及外界干扰因素等特点,提出了小波包与改进的BP神经网络相结合方法。首先提取原始振动信号的时域波形,利用小波包对于时域特征值进行分解,将小波包分解后的子频带能量进行归一化计算,并作为BP网络的输入特征向量,其次创建BP故障模型,通过改进的优化算法对神经网络的权值和阈值进行全局寻优,最终使得对故障类型进行精确的判断,提高了故障诊断效率。仿真结果比较了两种优化算法的性能指标,得出改进的蚁群算法(MMA)优化算法识别精度更高,误差相对较小,收敛速度更快,验证了方法的有效性和可靠性。As an important mechanical part of the fan transmission system,the gearbox is prone to failure when it works in bad environment for a long time.Aiming at the characteristics of non-stationary,transient,uncertain and external disturbance of vibration signal generated by fan gearbox,a method combining wavelet packet and improved BP neural network was proposed.First extract the original vibration signal in the time domain waveform,for the time domain characteristic value by using wavelet packet decomposition,wavelet packet decomposition of the sub-band energy after normalization,and as input feature vectors of the BP network;Then create BP fault model,through the improved optimization algorithm for neural network weights and threshold for global optimization;Finally,the fault type can be accurately judged and the fault diagnosis efficiency can be improved.Simulation results show that the improved ant colony algorithm(MMA)has higher identification accuracy,smaller error and faster convergence speed,which verifies the validity and reliability of the method.

关 键 词：小波包分解 神经网络 智能优化算法 风机齿轮箱 故障检测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]