数据驱动的高速列车轴承振动信号故障诊断方法研究

Data-Driven Fault Diagnosis Method for High-Speed Train Bearing Vibration Signals

摘要: 在设备故障诊断领域，振动信号作为反映设备运行状态的核心载体，其数据质量与特征提取效率直接决定故障识别的准确性与可靠性。为构建后续故障诊断任务的坚实数据基础，本研究首先以源域试验台架为研究对象，对其采集的振动信号进行系统性的数据规整与样本构建工作。具体而言，通过信号去噪、时序同步对齐、异常值剔除及数据标准化等一系列预处理操作，有效提升了原始振动信号的纯度与一致性，最终形成了一个涵盖正常运行、轻度磨损、中度故障及严重失效四种典型健康状态的标准化数据语料库，为后续模型训练与验证提供了统一、可靠的数据源支撑。针对传统故障诊断方法依赖人工特征提取、主观性强、泛化能力弱的局限性，本研究以CWRU轴承故障数据集作为源域，结合部分高速列车试验台架数据作为目标域，构建了源域与目标域数据集x。通过信号重采样、滑动窗口分段及Z-score标准化等预处理操作，形成标准化样本语料库。提出了一种基于1D-CxNN的端到端数据驱动故障诊断方法，在源域数据集上实现了99.68%的平均分类准确率，为高速列车轴承智能故障诊断提供了方法学探索和初步验证。

Abstract: In the field of equipment fault diagnosis, vibration signals, as the core carrier reflecting the operating state of equipment, their data quality and feature extraction efficiency directly determine the accuracy and reliability of fault identification. To establish a solid data foundation for subsequent fault diagnosis tasks, this study first takes the source domain test bench as the research object and conducts systematic data regularization and sample construction on the collected vibration signals. Specifically, through a series of preprocessing operations such as signal denoising, time series synchronization and alignment, outlier elimination, and data standardization, the purity and consistency of the original vibration signals are effectively improved. Finally, a standardized data corpus covering four typical health states—normal operation, mild wear, moderate fault, and severe failure—is formed, providing a unified and reliable data source support for subsequent model training and verification. Aiming at the limitations of traditional feature engineering, which is time-consuming, subjective, and has weak generalization ability, this study takes the Case Western Reserve University (CWRU) bearing fault dataset as the source domain and partial high-speed train test bench data as the target domain. After preprocessing operations including resampling, sliding window segmentation, and Z-score standardization, a standardized sample corpus is constructed. A 1D-CNN-based end-to-end data-driven fault diagnosis method is proposed, achieving an average classification accuracy of 99.68% on the source domain dataset. This provides methodological exploration and preliminary validation for intelligent fault diagnosis of high-speed train bearings.