The intershaft bearing, a critical component of an aero-engine, is susceptible to failure. The restricted accessibility of labeled fault data across varying working conditions and fault types, poses a challenge to its intelligent diagnosis, which is referred to as ’missing training samples’. This issue requires methods to handle both domain generalization and open-set domain generalization, adding practical and technical complexity. To address this problem, this paper proposes a new self-improving fault diagnosis method for intershaft bearings. The framework is designed to accommodate the scenario of initial sample scarcity with continuous sample enrichment as diagnosis progresses. The key of our method is to enhance the capability of extracting domain-invariant and discriminative features, as well as the construction of the corresponding classifier. Firstly, guided by the information bottleneck principle, a feature extraction model named HybVAE is established. Subsequently, a metric learning strategy is employed to design the objective function for hyperparameter optimization. HybVAE is jointly optimized through loss function and hyperparameters to ensure generalization capability and robustness in diverse domains. Finally, a classifier utilizing the kernel trick is introduced to distinguish unknown fault categories and classify known ones. As new data are collected, the method undergoes a process of self-improvement to diagnose a broader range of fault types. Experiments conducted on two intershaft bearing datasets verified the effectiveness and advantages of the proposed method.

中文翻译：

---

一种缺失训练样本的轴间轴承自改进故障诊断方法


轴间轴承是航空发动机的关键部件，很容易出现故障。在不同工作条件和故障类型下，标记故障数据的可访问性受到限制，这对其智能诊断构成了挑战，这被称为“缺失训练样本”。此问题需要同时处理域泛化和开放集域泛化的方法，这增加了实际和技术复杂性。针对这一问题，本文提出了一种新的轴间轴承自改进故障诊断方法。该框架旨在适应初始样本稀缺的情况，并随着诊断的进展不断富集样本。我们方法的关键是增强提取领域不变和判别特征的能力，以及相应分类器的构造。首先，在信息瓶颈原理的指导下，建立了一种名为HybVAE的特征提取模型;随后，采用度量学习策略来设计超参数优化的目标函数。HybVAE 通过损失函数和超参数进行联合优化，以确保在不同领域的泛化能力和鲁棒性。最后，介绍了一种利用内核技巧的分类器来区分未知故障类别并对已知故障进行分类。随着新数据的收集，该方法会经历一个自我改进的过程，以诊断更广泛的故障类型。在两个轴间轴承数据集上进行的实验验证了所提方法的有效性和优势。