基于ADASYN和Swin Transformer的滚动轴承故障诊断研究

针对实际工况下，正常样本丰富、故障样本稀缺的类别不平衡情形，导致基于深度学习的故障诊断模型诊断能力较差这一问题，提出一种基于自适应综合采样方法(ADASYN)和Swin Transformer的故障诊断模型。使用自适应综合采样方法，改善数据分布，解决实际工况中故障样本与正常样本类别不平衡问题；使用Swin Transformer网络模型代替CNN网络 ，并使用深度迁移学习方法，使Swin Transformer网络模型掌握判别滚动轴承故障所需的浅层权重，深层权重通过反向传播方法训练获得；之后，将模型用于轴承故障测试，并对其进行调试；最后，将模型用于轴承故障实测，检验其实际工况下的诊断能力。实验结果表明：所提模型具有97%的诊断准确率，能够很好地适用于类别不平衡情形下的滚动轴承故障诊断。

Study of Rolling Bearing Fault Diagnosis Based on ADASYN and Swin Transformer

To address the situation of class imbalance where normal samples are abundant and fault samples are scarce under actual working conditions leading to the problem that the fault diagnosis model based on deep learning has poor diagnostic capability,a fault diagnosis model based on adaptive synthetic sampling (ADASYN) and Swin Transformer was proposed.An adaptive synthetic sampling method was used to improve the data distribution and to address the imbalance between faulty and normal sample categories in actual operating conditions.A deep transfer learning method was used to equip the model with the shallow weights needed to discriminate rolling bearing faults,and the deep weights were obtained by back propagation training.Afterwards,the model was used for bearing failure testing and debugging.Finally,the model was used in a real-world test of bearing faults to check its diagnostic capability under actual operating conditions.The experimental results show that the proposed model has a diagnostic accuracy of 97% and is well suited to the diagnosis of rolling bearing faults in the case of category imbalance.