高铁传动系统关键轴承稀疏深度网络健康预测方法研究

Traction motor bearings and axle box bearings are the core components which exist widely and affect safety in the high-speed rail transmission system, bearing failure often occur under high-speed and heavy-load conditions, and the health prediction of these bearings is essential for ensuring safe traffic and lowering maintenance cost. In this project, we aim at conducting research on health prediction methods of bearings based on sparse convolutional neural networks, and breaking through the limitation of overreliance on expert experience and low accuracy of traditional fault diagnosis methods. Intensive researches will be conducted on three key problems: construction of sparse indicators, quantitation of warning threshold and prediction of bearing health. Firstly, aiming at the problem of low signal-noise ratio and under condition of more sources than sensors, a sparse construction method of bearing condition indicator will be researched to reveal the characterization mechanism of bearing health status. Secondly, aiming at the problem of quantification of multidimensional condition indicators of bearings in the traction system, an intelligent learning method for threshold determination will be proposed to clarify the mechanisms between the warning criteria and operating conditions; Thirdly, aiming at the problem of health condition prediction with multi-sensor on big data, a bearing health prediction method based on information integration and deep learning will be researched to improve the adaptability and accuracy of traditional methods; Experimental research and engineering verification of the key bearings in the traction system will be conducted. Through the project, a kind of health prediction method suitable for critical bearings components in the traction system of high-speed railway would be constituted, which may shed light on providing theoretical bases for safe traffic and health management of high-speed rail.

牵引电机轴承及轮对轴箱轴承是高铁传动系统中大量存在且影响运行安全的关键部件，在高速重载条件下经常发生故障现象，研究其健康预测方法对确保高铁安全运行与高效运维至关重要。本项目拟开展稀疏深度网络健康预测方法研究，突破传统智能诊断方法过度依赖专家经验及准确率低的问题。重点针对稀疏指标构造、定量预警以及健康预测方法三类核心问题展开研究。针对监测信噪比低及欠定问题，研究状态指标的稀疏构造方法，揭示状态指标表征机理；针对多维状态指标定量预警问题，提出状态阈值智能学习方法，厘清预警准则与运行工况的关联机制；针对多传感海量监测数据下轴承健康预测问题，研究基于信息融合与深度学习的健康预测方法，改善传统深度模型对新生故障的适应性，提升健康预测的精度；针对高铁传动系统关键轴承健康预测方法开展实验研究及工程验证。通过本项目的研究，可望探索出高铁传动系统关键轴承健康预测理论，为高铁运维及健康管理的研究提供理论依据。

本项目以高铁列车为研究对象，主要针对高铁传动系统中的关键轴承在复杂恶劣工况下安全服役问题，开展了健康预测方法的深入研究。重点针对状态指标构造、阈值学习方法、健康预测开展了系列化深入研究。主要研究内容包括：高铁传动系统关键轴承运行状态指标稀疏构造方法研究、状态阈值智能学习方法研究、健康预测方法研究与实验应用研究。研究创新成果主要包括研究出了一种轴承群故障健康预测理论与方法，并构建了基于注意力机制的深层学习模型，较大程度摆脱了对大量信号处理技术与诊断经验的依赖，实现大量数据下故障特征的自适应提取与健康状况的智能诊断，对于解决实际工程问题至关重要，准确率提升20%以上，并得到了初步装车工程验证。项目积累了高铁牵引电机轴承数据20T，包含牵引电机驱动端轴承(NU214轴承)和非驱动端轴承(6311轴承)的内圈、外圈、保持架和滚子的轻微、中等和严重故障，并在低载、中载和重载工况和低速、额定转速和高速工况下积累了珍贵故障数据，共三年的实验数据形成了核心数据集，为稀疏深度学习理论研究奠定了坚实的基础。通过本项目的研究，共发表标注青年基金项目的学术论文7篇，会议论文1篇，其中SCI论文7篇，申报发明专利18项，授权9项，项目形成的理论成果以发明专利转让的形式完成两项共54万的专利转让与两项应用证明，获得行业认可。