Abstract: In practical industrial scenarios, the significant divergence in bearing state evolution caused by complex operating conditions imposes dual constraints on remaining useful life (RUL) prediction: incomplete mechanistic understanding and distribution shift in monitoring data. Although existing data-driven methods perform well under stable conditions, they suffer from strong reliance on annotated data and limited generalization capability under sudden changing conditions. To address these issues, this paper proposes a dynamic knowledge embedding-based method for bearing RUL prediction. The method innovatively integrates data-driven features and domain prior knowledge in a dynamic manner: first, a dynamic knowledge graph is constructed, where prior knowledge of bearing degradation is encoded into computable triplets, and a sliding window confidence mechanism is introduced to enable adaptive updating of domain knowledge; second, relational graph convolutional networks are used to extract physically meaningful embedding vectors, which are fused cross-modally with time-frequency features of vibration signals extracted by hierarchical convolutional networks; finally, a multi-head self-attention Transformer models dynamic interactions between features and knowledge, allowing the model to adaptively balance the contributions of data-driven features and mechanistic knowledge. Experimental results on the PHM2012 and Xi’an Jiaotong University datasets demonstrate that the proposed method significantly improves the accuracy of cross-condition bearing RUL prediction.

摘要： 在实际工业场景中，由于复杂工况导致轴承状态演化差异显著，使得寿命预测面临机制认知不完整与监测数据分布漂移的双重约束。现有基于数据驱动的方法虽在稳定工况下表现良好，但存在标注数据依赖性强、突变工况泛化能力不足的局限。针对上述问题，本文提出一种基于动态知识嵌入的轴承寿命预测方法。该方法创新性地动态融合了数据驱动特征与领域先验知识：首先，构建动态知识图谱，将轴承退化先验知识编码为可计算的三元组知识并引入滑动窗口置信度机制，实现了领域知识的自适应更新；其次，利用关系图卷积网络提取具有物理意义的嵌入向量，与层级卷积网络提取的振动信号时频特征进行跨模态融合；最后，通过多头自注意力Transformer实现特征与知识的动态关联建模，使模型能够自适应地平衡数据特征与机理知识的贡献。在PHM2012和西安交通大学数据集上的实验结果表明，该方法显著提升了跨工况的轴承寿命预测精度。