针对当前的无监督域自适应算法应用于滚动轴承故障诊断领域时，源域数据不平衡，且两个域之间存在域偏移，导致故障识别率低的问题，提出了一种基于多重流形标签传播的滚动轴承故障诊断方法，旨在将源域和目标域的数据多重投影到共同的子空间，减少域内以及跨域的差异，同时平衡样本分布，进而提高变工况轴承故障诊断的精度。首先，提出域内局部保持平衡映射方法，将源域和目标域数据映射到一重流形子空间，得到域内对齐后的样本数据，并对源域数据进行平衡处理，得到平衡后的源域数据；然后提出跨域流形结构细化对齐方法，将数据进一步映射到二重共享子空间，得到细化对齐后的样本数据；最后通过动态加权伪标签域适应传播方法，得到准确度高的伪标签。分别在CWRU和自建的轴承数据集上进行故障诊断试验，实验结果表明，所提方法不仅对多故障类型多故障尺寸、复合故障上有着较好的识别能力，且当标签样本稀少时，模型也表现出优秀的诊断效果。

In the field of rolling bearing fault diagnosis, current unsupervised domain adaptation algorithms often face challenges due to unbalanced source domain data and domain shifts between the source and target domains, leading to low fault recognition rates. To address these issues, we propose a rolling bearing fault diagnosis method based on multi-manifold label propagation. This method aims to project the data from both the source and target domains into a common subspace, reducing intra-domain and cross-domain differences and balancing the sample distribution. Consequently, this enhances the accuracy of fault diagnosis under variable operating conditions.Firstly, we introduce a locally balanced mapping method within the domain, which maps the source and target domain data into a manifold subspace, resulting in aligned sample data. This step also balances the source domain data. Next, we propose a cross-domain manifold structure refinement alignment method, further mapping the data into a double shared subspace to obtain refined aligned sample data. Finally, we employ a dynamic weighted pseudo-label domain adaptive propagation method to achieve highly accurate pseudo-labels.Fault diagnosis experiments were conducted on both the CWRU and self-built bearing datasets. The experimental results demonstrate that the proposed method not only effectively recognizes multiple fault types, fault sizes, and composite faults but also exhibits excellent diagnostic performance even when labeled samples are scarce.