CWNT时频尺度多步噪声抑制的轴承故障诊断

Bearing Fault Diagnosis of Multi-step Noise Suppression at CWNT Time-frequency Scale

实际工况中的轴承信号常带有不同程度的噪声干扰,不利于识别轴承的健康状态,且严重影响其故障诊断的稳定性。针对噪声信号的处理难题,提出了一种CWNT时频尺度多步噪声抑制的轴承故障诊断方法。通过不同时频尺度分解,利用连续小波变换(CWT)的带通特性,可变长度时窗截取变换定位含噪信号的集中波形;从全局感受野对故障信号滤波,深层卷积非线性提取时频图像特征;在通道维度增加注意力机制,加权归一化权重至每个通道的特征,拟合通道间的复杂关联,以此达到多步噪声抑制的诊断效果。为验证所提诊断方法,向凯斯西储大学轴承数据集(CWRU)添加功率比为4.987%、12.538%、31.650%的高斯白噪声,诊断得到验证集准确率分别为97.384%、96.701%、95.407%。结果表明,CWNT方法具有较强的抗噪能力,能够提高噪声背景下故障诊断的准确性。

In the actual working conditions,the bearing signal often has different degrees of noise interference,which is not conducive to identify the health state of the bearing,and affects the stability of the fault diagnosis seriously.For the problem of processing noisy signals,a bearing fault diagnosis method for multi-step noise suppression at CWNT time-frequency scale(CWNT)is proposed.The concentrated waveform of continuous wavelet transform(CWT)is captured by the window of variable length;the fault signal from the global perception field is filtered and the time-frequency image features are extracted;the attention mechanism is added to the channel dimension,the normalized weight to the characteristics of each channel are weighted,the complex correlation between channels are fitted to achieve the diagnostic effect of multi-step noise suppression.In order to verify the proposed diagnostic method,gaussian white noise(GWN)of 4.987%,12.538%and 31.650%is added to the Case Western Reserve University bearing data set(CWRU),and the accuracy of the validation set is 97.384%,96.701%and 95.407%,respectively.The results show that CWNT method has strong noise resistance and can improve the accuracy of fault diagnosis in noise background.