【摘要】 为了解决滚动轴承故障诊断中样本分布差异大、有效故障样本少以及不同故障样本数量不均衡所导致的诊断精度较低的问题；提出基于个性化联邦迁移学习(personalized federated transfer learning, PFTL)的滚动轴承故障诊断方法。在所提出的PFTL中，首先在预训练阶段，将不同分布的各类型故障样本作为联邦学习的各个客户端的输入，并引入贝叶斯层级模型对联邦学习的本地训练和聚合规则进行个性化调整，从而使得预训练模型在避免过拟合问题的同时具有较强的泛化能力；其次引入模型补丁，对预训练模型结构进行调整，并利用目标任务样本对模型进一步微调；最后在CWRU轴承数据集上进行故障诊断实验。实验结果证明所提方法的有效性。更多还原

【Abstract】 In order to solve the problem of low diagnostic accuracy caused by large sample distribution difference, few effective fault samples and unbalanced number of different fault samples in rolling bearing fault diagnosis; A fault diagnosis method for rolling bearings based on personalized federated transfer learning(PFTL) is proposed. In the proposed PFTL, firstly, in the pre-training stage, various types of fault samples with different distributions are taken as input to each client of federated learning, and a Bayesian hierarchical model is introduced to make personalized adjustment to the local training and aggregation rules of federated learning, so that the pre-trained model has strong generalization ability while avoiding overfitting problems. Secondly, the model patch is introduced to adjust the structure of the pre-trained model, and the target task sample is used to further fine-tune the model. Finally, the fault diagnosis experiment is carried out on the CWRU bearing data set. Experimental results prove the effectiveness of the proposed method.更多还原