基于振动信号图像特征的降噪残差网络轴承故障诊断

Bearing Fault Diagnosis Based on Adaptive Denoise Residual Network with Image Features of Vibration Signals

针对将一维原始轴承振动信号作为既有轴承诊断模型的输入所致训练效率、抗噪性欠佳的问题,提出一种基于振动信号图像特征的自适应降噪残差网络轴承故障诊断方法。首先将一维轴承振动信号进行截断、重叠采样后重构成信号矩阵,最后将其编码为图像得到振动信号图像;再对图像进行直方图处理,计算得到其灰度分布特征矩阵,并将振动信号图像和对应的特征矩阵作为算法模型的输入;同时,在提出的网络模型中在残差卷积映射的过程中插入基于通道注意力机制的降噪路径,通过自适应地获得阈值进行降噪,提高网络对含噪声样本的故障特征提取能力。最后通过对比实验证明:网络模型在加入灰度分布特征后有更好的性能表现,提出的自适应降噪残差网络模型在将含有噪声的振动信号作为输入的情况下仍具有较高的故障识别精度。

In order to solve the problem that the bearing diagnosis model has poor efficiency and anti-noise performance for the one-dimensional original bearing vibration signal input,a bearing fault diagnosis method based on adaptive denoise residual network with vibration signals′image feature was proposed.In this method,the one-dimensional bearing vibration signal was truncated and overlap-sampled,then reconstructed into a signal matrix,and finally encoded into an image to obtain a vibration signal image.The histogram processing was used to process the images to obtain a grayscale distribution feature matrix.The vibration signal image and the its grayscale distribution feature matrix were used as the input of the algorithm model.And a denoise path based on the channel attention mechanism was inserted into the process of residual convolution mapping in the proposed model,and the threshold for denoising was obtained adaptively.Finally,the fault feature extraction performance of the network for noisy samples was improved.The comparative experiments show that the model after adding grayscale distribution feature has better performance;the proposed adaptive noise reduction residual network model still has high fault identification accuracy although the vibration signal containing noise is used as the input.