面向高速列车轴箱轴承寿命预测的数据特征知识迁移理论与应用研究

How to predict and even extend remaining life of axle box bearings is of great significance to ensure the safe and reliable operation of high-speed trains. However, the life cycle data of axle box bearings are difficult to be acquired in the operation condition, which seriously restricts the development and application of life prediction technology. Fortunately, the data collected from the laboratory bearings have sufficient performance state information. These data also have related feature knowledge to the collected data from the real-case bearings. Therefore, a new feature knowledge transfer method for predicting remaining life is proposed to solve the limited data problem. Firstly, a knowledge analysis method for monitoring data of axle box bearings is proposed to explain the influence mechanism of working conditions on the vibration characteristics of axle box bearings. The method also helps to trace and measure the distribution discrepancy of feature knowledge from the different data domains. Secondly, a new deep transfer learning method is built to construct a health status indicator that works well across data domains. Additionally, the generalization error of the method is theoretically analyzed, which reveals the cross-domain migration and reuse mechanism of monitoring data feature knowledge across axle box bearings. Thirdly, a life prediction model considering multiple uncertain factors is constructed to predict the remaining useful life of bearings in real time. This project not only enriches the basic theories and methods of feature knowledge transfer, but also solves the data constraint problem in life prediction technology, which helps to improve the intelligent level of predictive maintenance of high-speed train axle box bearings. It has important academic research significance and engineering practical value.

如何对轴箱轴承定寿甚至延寿对于保障高速列车的安全可靠运行具有重要意义。然而，在途运行轴箱轴承的全寿命周期数据不足，严重制约着其寿命预测技术的发展和应用。鉴于实验室环境中可获取充足的轴箱轴承数据，且此类数据与在途运行轴箱轴承的监测数据存在相关特征知识，本项目提出通过迁移特征知识解决上述数据制约问题。研究轴箱轴承监测数据特征知识解析方法，阐述工况条件对轴箱轴承振动特性的影响机制，从而溯源并度量数据特征知识分布差异；建立深度迁移学习模型并理论分析模型泛化误差，揭示轴箱轴承间监测数据特征知识的跨域迁移复用机理，构建跨域适配的健康指标；进一步建立考虑多层不确定因素的寿命预测模型，实现轴箱轴承寿命的准确在线预测。本项目工作不仅丰富了数据特征知识迁移的基础理论和方法，而且解决了寿命预测技术中的数据制约问题，有助于提升高速列车轴箱轴承预测性维护的智能化水平，具有重要的学术研究意义和工程实用价值。

本项目以高速列车轴箱轴承为研究对象，从动力学建模、故障信号分析与处理、不完备条件下的智能诊断、特征知识迁移下的寿命预测等方面开展研究。主要研究成果与创新点包括：（1）建立了轴箱轴承内部损伤激励动力学模型，阐明了不同故障模式、不同工况条件对关键部件动力学响应的影响规律，为轴箱轴承的动态信号分析、迁移诊断与预测奠定了理论基础；（2）提出了基于字典学习及谱平均的信号分解重构方法、基于故障特征阶数自动搜索的诊断信号解调技术、自适应瞬时相位估计和阶次跟踪方法等，实现了高噪声与变工况条件下的轴承故障特征信息表征；（3）构建了在样本标签缺失、样本类别不平衡、样本跨域等极端条件下的故障诊断网络，阐明了数据特征知识跨域迁移复用机理，实现了不完备条件下的轴承智能故障诊断；（4）设计了基于信号脉冲性与周期性的健康指标、基于自注意卷积自编码的健康指标等，形成了基于非线性随机模型参数优化的轴承剩余寿命预测方法。在本项目的资助下，共发表有本项目资助号标注的SCI期刊论文24篇、会议论文4篇；申请国家发明专利3项，申请软件著作权两项。培养博士研究生4名，硕士研究生6名。项目负责人获2022年四川省科技进步奖一等奖（排名第二）、2021年陕西省自然科学奖一等奖（排名第五）等奖励；获2022年四川省杰出青年基金、2021年四川省“天府峨眉计划”青年人才、西南交通大学“雏鹰学者”等荣誉。