基于轻量级CNN的电机轴承故障诊断研究

Research on fault diagnosis of motor bearing based on lightweight CNN

针对深度学习在嵌入式或移动端设备中用于故障诊断时,受限于有限的硬件资源而又需要有足够的效率和精度的应用需求,提出基于轻量级卷积神经网络的电机滚动轴承故障诊断方法。首先对滚动轴承的振动信号数据集进行连续小波变换生成固定尺寸的时频图,并以此方式生成数据集输入网络进行训练。使用测试集进行测试,结果表明,所生成的故障诊断预测网络模型具有较高的识别精度和识别速度,准确率达到99%。通过验证噪声对网络的影响,表明所使用的网络具有较好的鲁棒性和泛化能力。

Aiming at the application requirements of limited hardware resources and sufficient efficiency and accuracy when deep learning is used for fault diagnosis in embedded or mobile devices,a fault diagnosis method of motor rolling bearing based on lightweight convolutional neural network is proposed.Firstly,continuous wavelet transform is applied to the vibration signal data set of rolling bearing to generate the fixed size time-frequency diagram,and then the data set is generated and input into the network for training.The test results show that the network model has high recognition accuracy and speed,and the accuracy rate is 99%.By verifying the influence of noise on the network,it shows that the network has good robustness and generalization ability.