基于特征增强与深度学习的轴承故障诊断

Bearing fault diagnosis based on feature enhancement and deep learning

针对轴承故障诊断中数据集较小,现有诊断方法鲁棒性较低且易被噪声干扰的难题,提出了基于特征增强和卷积神经网络故障识别方法。首先对振动采样信号进行短时傅里叶变换(STFT)与小波变换处理,获取时频图,然后对时频图进行卷积操作,获取故障信号特征图。最后,将获得的特征图通过通道注意力机制模块,再通过卷积神经网络,实现对轴承故障的分类。结果表明,该方法在西储大学数据集添加-40 dB噪声的情况下,故障准确率达97%,在西储大学数据集以及江南大学离心风机轴承数据集上识别准确率分别为99.8%和100%。

In order to solve the problems of small dataset in bearing fault diagnosis, as well as, low robustness and easy to be disturbed by noise of existing diagnosis methods, a fault identification method based on feature enhancement and convolutional neural network was proposed. Firstly, the vibration sampling signal was processed by short-time Fourier transform(STFT) and wavelet transform to obtain the time-frequency map, then the timefrequency map was convoluted to obtain the fault signal feature map. Finally, the obtained feature map was passed through the channel attention mechanism module, and then the convolutional neural network to achieve the classification of bearing faults. The results showed that the method achieved 97% fault accuracy with the addition of-40 dB noise to Case Western Reserve University dataset, and 99.8% and 100% recognition accuracy on Case Western Reserve University dataset and Jiangnan University centrifugal fan bearing dataset, respectively.