针对风电机组滚动轴承振动信号具有非平稳、非线性、强干扰的特点以及故障特征提取困难的问题,提出了一种基于奇异值分解(singular value decomposition,简称SVD)、S变换与多尺度卷积神经网络(multiscale convolutional neural network,简称MSCNN)的故障诊断方法.首先,将原始信号构造成Hankel矩阵,采用SVD对Hankel矩阵进行奇异值分解,再根据奇异值曲率谱选取有效奇异值进行信号重构;其次,对重构信号进行S变换生成特征图谱;最后,将其输入到MSCNN自适应提取特征进行故障识别.试验结果表明,利用SVDS-MSCNN方法进行风电机组滚动轴承故障诊断,其故障识别准确率达到97.5％,故障诊断效果优于其他深度学习算法.
Fault Diagnosis of Wind Turbine Rolling Bearing Based on SVDS-MSCNN
For the rolling bearing vibration signal of wind turbine,it occurs the characteristics of non-stationary,non-linear and strong interference and the difficulty in extracting fault features.Therefore,a fault di-agnosis method based on singular value decomposition(SVD),S-transform and multi-scale convolutional neu-ral network(MSCNN)is proposed.First,the original signal is constructed into a Hankel matrix,and the Han-kel matrix is singular decomposed by SVD.Second,the signal is reconstructed by selecting the effective singu-lar value according to the curvature spectrum of the singular value.Third,the S-transform is performed using the reconstructed signal to generate characteristic map..Finally,it is input into MSCNN to extract feature adaptively for fault identification.Through the test verification,the SVDS-MSCNN method is used for wind turbine roll-ing bearing fault diagnosis,and the fault recognition accuracy reaches 97.5％,which is better than other deep learning algorithms.
rolling bearingfault diagnosissingular value decompositionS-transformconvolutional neural networks
NETL
NSTL
万方数据
