基于组合降噪的卷积神经网络轴承故障诊断方法

Fault Diagnosis of Bearing Based on Convolutional Neural Network with Combined Noise Reduction

针对滚动轴承振动信号包含不同频率复杂噪声,单一降噪方法难以去除多种噪声,导致最终故障诊断率低的问题。提出一种基于组合降噪的卷积神经网络的轴承故障诊断方法。首先,通过SVD分解方法,根据奇异值差分谱,一次降噪并重构信号,去除信号内能量较低的宽频率噪声;再根据降噪重构信号,自适应选取CEEMD分解的参数,并进行自适应CEEMD分解,利用线性相关系数与峭度交集法滤除相关程度低、故障特征信息少的IMF分量,同时二次降噪并重构信号;最后,构建一维卷积神经网络并进行轴承故障诊断。通过原信号和组合降噪后信号的时频分析,轴承故障诊断实验,验证了该方法的有效性。

As the vibration signal of rolling bearing contains complex noise of different frequencies,it is difficult to remove multiple noises by a single noise reduction method,which leads to the problem of low fault diagnosis rate in the end.This paper presents a bearing fault diagnosis method based on convolutional neural network with combined noise reduction.First,the SVD decomposition method is used to reduce the noise once and reconstruct the signal according to the differential spectrum of singular values,so as to remove the wide-frequency noise with lower energy in the signal.Then,according to the noise reduction and reconstruction signal,the parameters of CEEMD decomposition are adaptively selected and the adaptive CEEMD decomposition is conducted.The IMF component with low correlation degree and little fault feature information is filtered by the method of linear correlation coefficient and kurtosis intersection,and the secondary noise reduction signal is reconstructed at the same time.Finally,the one-dimensional convolutional neural network is constructed and the bearing fault diagnosis is performed.In this paper,the effectiveness of this method is verified by time-frequency analysis of the original signal and the combined de-noising signal for the bearing fault diagnosis.