针对迁移诊断中存在的源域和目标域分布差异导致的负迁移以及过分依赖源域样本带来的数据隐私问题，提出一种利用邻域信息优化伪标签监督训练的无源域自适应(Source-free domain adaptation, SFDA)迁移诊断方法以实现在无源域样本情况下的迁移诊断。首先，通过奇异谱分解(Singular spectrum decomposition, SSD)方法对数据进行降噪处理，使得样本具有更丰富的故障信息，然后，基于一维卷积神经网络构建特征提取器以提取域不变特征；其次，采用对比学习框架拉近同一类样本特征，利用聚合邻域信息精炼后的伪标签进行自监督学习；最后，基于智能诊断模型完成跨设备变工况下滚动轴承健康状态的识别。通过两个滚动轴承数据集间的跨设备迁移诊断验证所提方法的有效性。实验结果表明：所提方法能够充分挖掘不同设备间故障特征信息，提高无源无监督跨域条件下的迁移诊断精度。

In response to the challenges posed by distributional differences between the source and target domains leading to negative transfer in transfer diagnostics, as well as the data privacy concerns stemming from excessive reliance on source domain samples, this paper proposes a Source-free Domain Adaptation (SFDA) method for transfer diagnostics. This method leverages neighborhood information to optimize pseudo-label supervised training, enabling transfer diagnostics in the absence of source domain samples.Initially, the data undergoes denoising using Singular Spectrum Decomposition (SSD) to enrich the samples with more abundant fault information. Then, a one-dimensional convolutional neural network is employed to construct a feature extractor for extracting domain-invariant features. Subsequently, a contrastive learning framework is applied to bring closer the features of samples from the same class, utilizing refined pseudo-labels derived from the aggregation of neighborhood information for self-supervised learning. Finally, an intelligent diagnostic model is deployed to accomplish the recognition of the health status of rolling bearings under varying operational conditions across different devices. The effectiveness of the proposed method is validated through cross-device transfer diagnostics between two rolling bearing datasets. Experimental results demonstrate that the proposed method effectively uncovers fault feature information across different devices, thereby enhancing transfer diagnostic accuracy under source-free and unsupervised cross-domain conditions.