WPT-MSE结合PNN的电机轴承故障诊断方法

摘要

Rolling bearing is an important element in the motors;under actual operating condition,the interference of background noise is serious,so effective feature extraction and accurate fault judgment for motor bearings are difficult to be achieved.In order to solve this problem,a diagnosis method based on wavelet packet transform(WPT),multiscale entropy(MSE) and probabilistic neural network (PNN)is proposed.Firstly,the acquired signal samples are processed using wavelet packet transform,the original signals are decomposed into the corresponding subband signal components,and the cross interferences between the complex components in the original signals could be reduced.Then,the multiscale entropy of each subband signal component obtained is calculated,and the extracted multiscale entropy is used to construct the feature vectors which can characterize the operating conditions of the motor bearings.Finally,the feature vectors are input into the probabilistic neural network,and the different fault types and injury degrees of the motor bearings can be identified automatically.The analysis results of the measured data show that the diagnosis method proposed can effectively identify the different operating conditions of the motor bearings,which provides a new reference for fault diagnosis of motor bearings and has a certain value for engineering applications.%滚动轴承是电机的重要组成零部件,在实际运行工况条件下,背景噪声干扰严重,难以实现电机轴承特征的有效提取及故障的准确判别.为了解决这一问题,提出了一种基于小波包变换(UPT)、多尺度熵(MSE)和概率神经网络(PNN)的诊断方法.首先,利用小波包变换对拾取的信号样本进行处理,并将原始信号分解为相应的子带信号分量,以减少原始信号中复杂成分之间的交叉干扰;然后,计算所得子带信号分量的多尺度熵值,并利用提取出的多尺度熵值构造能够表征电机轴承运行状态的特征向量;最后,将特征向量输入概率神经网络中,实现对电机轴承不同故障类型及损伤程度的自动识别.实测数据分析结果表明,所述诊断方法能够有效识别电机轴承的不同工作状态,从而为电机轴承的故障诊断提供了参考,具有一定的工程应用价值.