基于SIC变点识别与关系网络的轴承故障预警
Bearing Fault Warning Based on SIC Change-Point Detection and Relation Networks

为有效识别轴承振动异常并实现早期预警，本文提出一种融合SIC变点识别与关系网络的轴承故障预警方法。针对传统3σ准则对非正态分布数据适应性不足的问题，引入西沃兹信息准则(Schwarz information criterion, SIC)实现全局变点检测。针对轴承样本稀缺的实际情况，构建了一种基于关系网络的小样本状态评估模型。在模型训练阶段，首先采用SIC准则对历史轴承数据执行变点分析，据此构建支持集与查询集，通过关系网络学习不同状态样本特征间的非线性差异度量；在线监测阶段，实时采集轴承运行数据，计算其与健康样本的特征关系得分，作为健康状态指标。进一步，结合k均值聚类方法设定动态检测阈值，实现对轴承运行状态的实时判别与预警。与传统故障预警方法对比，该方法在小样本条件下仍具备良好的早期故障识别能力，为轴承的状态监测与故障预警提供了新的技术途径。
To effectively identify bearing vibration anomalies and achieve early warning, this paper proposes a bearing fault early warning method integrating Schwarz Information Criterion (SIC) change-point detection with relational networks. To address the inadequacy of the traditional 3σ criterion in handling non-normally distributed data, the Schwarz Information Criterion (SIC) is introduced for global change-point detection. Considering the practical scenario of limited bearing samples, a few-shot condition assessment model based on relational networks is constructed. During the model training phase, the SIC criterion is first employed to perform change-point analysis on historical bearing data, based on which support sets and query sets are established; the relational network subsequently learns the nonlinear discrepancy metric among feature embeddings of samples in different states. In the online monitoring phase, real-time bearing operational data are collected, and their feature relational scores with respect to healthy samples are computed as health indicators. Furthermore, dynamic detection thresholds are determined by incorporating k-means clustering to enable real-time state discrimination and early warning of bearing operation. Compared with conventional fault prognosis methods, the proposed approach maintains satisfactory early fault identification capability under few-shot conditions, offering a novel technical pathway for bearing condition monitoring and fault prognosis.