摘要
The rolling bearing vibration signal is non-stationary and is easily disturbed by background noise,so it is difficult to accurately diagnose bearing faults.A fault diagnosis method of rolling bearing based on the time-frequency threshold denoising synchrosqueezing transform(TDSST)and convolutional neural network(CNN)is proposed.Since the traditional methods of wavelet threshold denoising and wavelet adjacent coefficient denoising are greatly affected by the estimation accuracy of noise variance,a time-frequency denoising method based on the STFT spectral correlation coefficient threshold optimization is adopted,which is combined with a synchrosqueezing transform.The ability of the TDSST to reduce noise and improve time-frequency resolution was verified by simulated impact fault signals of rolling bearings.Finally,the CNN is utilized to diagnose the time-frequency diagrams obtained by the TDSST.The diagnostic results of the rolling bearing experimental data show that the proposed method can effectively improve the accuracy of diagnosis.When the SNR of the bearing signal is larger than 0 dB,the accuracy is over 95%,even when the SNR reduces to-4 dB,the accuracy is still around 80%.Moreover,the standard deviation of multiple test results is small,which means that the method has good robustness.
针对滚动轴承振动信号的非平稳性和易被背景噪声干扰导致故障难以被准确诊断的问题,提出了一种基于时频阈值降噪同步压缩变换(TDSST)和卷积神经网络(CNN)的滚动轴承故障诊断方法.由于传统的小波阈值降噪及小波相邻系数降噪方法受信号噪声方差估计精度影响大,因此采用了基于STFT谱相关系数阈值寻优的时频降噪方法,将其与同步压缩变换结合,并用滚动轴承模拟冲击故障信号验证了TDSST方法降噪及提高时频分辨率的能力.最后,利用CNN对TDSST方法处理得到时频图进行诊断,滚动轴承实验数据诊断结果表明了所提方法能够有效地提高诊断准确率,当轴承信号信噪比大于0 dB时,诊断准确率都达到了95%以上,即使信噪比降到-4 dB时,诊断准确率也维持在80%左右,并且多次测试结果的标准差较小,表明方法具有良好的鲁棒性.
