基于模型迁移学习的滚动轴承智能故障诊断方法

Rolling bearing intelligent fault diagnosis method based on model transfer learning

针对滚动轴承故障诊断中可用的故障数据较少,同时基于数据驱动的故障诊断模型在训练过程中需要耗费大量的时间和计算资源的问题,提出一种基于EfficientNet模型迁移学习的滚动轴承智能故障诊断方法。首先,利用信号转化图像的方法,将不同健康类别滚动轴承振动信号生成相应的训练集和测试集;然后,将经过预训练的EfficientNet模型,通过参数共享迁移到训练集上进行训练并微调,以达到模型最佳参数;最后,通过测试集验证模型的故障诊断能力。在双转子高速滚动轴承故障数据集上,对提出的故障诊断方法进行了验证。结果表明:在不同工况下,所提出故障诊断方法的准确率最高能达到99.48%,优于传统的数据驱动故障诊断方法,具有较好的应用前景。

Aiming at the problem that there are few fault data available in rolling bearing fault diagnosis,and the data-driven fault diagnosis model needs a lot of time and computing resources in the training process,a transfer learning intelligent fault diagnosis method for rolling bearings based on EfficientNet model was proposed.Firstly,the vibration signal data of rolling bearings in different health states were transformed into images to generate the corresponding training set and test set;then the pre-trained EfficientNet model was transferred to the training set through parameter sharing and fine-tuning to achieve the best parameters of the model;finally,the fault classification ability of the model was verified by the test set.The double-rotor high-speed rolling bearing fault diagnosis test bench was used to obtain the rolling bearing data sets of different health states under different working conditions,and the method was verified.The results show that the highest diagnostic accuracy of the proposed model under different working conditions can reach 99.48%,which is better than that of conventional diagnosis method and has a good application prospect.