高速动车组轴箱轴承故障演化特性与寿命预测方法研究

Axle bearing is one of the key rotating components in high-speed electrical multiple units (EMUs) bogie. Considering the factors of multi-source excitations, such as alternating load, track irregularity and wheel tread damage, the failure and abrasion state deviate of axle bearing increase obviously. In order to ensure the safe and reliable operation of high-speed EMUs, it is necessary to detect, identify and predict the serviceability and health status of axle bearing accurately. Firstly, an axle bearing dynamics model with the topography progresses evolution of localized fault defect is proposed. The vibration characteristic rules of bearing system under different failure damage stages are researched in this project. Then, inverse dynamics problem and life prediction of axle bearing are researched thereafter. The multi-channel and distributed monitoring network is adopted to collect information of bearing, and the preprocessing method of noise reduction, fault feature digging, extraction and fusion are studied for constructing parameter characteristic database which characterizes the service performance of bearing. Finally, the sparse self-coded deep neural network based on data-driven method is used to predict residual life of bearing. The research results will be helpful to improve the theoretical study level of fault detection, diagnosis and prediction, and provide theoretical foundation and technical support for developing the preventive maintenance of key equipment components of high-speed EMUs.

轴箱轴承是高速动车组走行部最重要的旋转部件之一，受交变载荷、轨道不平顺、车轮踏面损伤等多源激励影响，加剧了轴承故障产生及损伤状态偏离演变过程。准确监测、识别、预测轴箱轴承服役性能和健康状态对确保高铁安全、可靠运行具有重要意义。本项目拟建立含局部故障缺陷形貌特征演化的轴箱轴承系统动力学模型，揭示故障不同损伤阶段下轴承系统振动特性规律。在此基础上，进行轴箱轴承动力学反问题与寿命预测研究。通过多通道、分布式监测网络采集轴承系统状态信息，开展复杂数据降噪预处理、特征挖掘、提取及多源信息融合方法研究，构建描述轴承服役状态的特征信息数据库。搭建稀疏自编码深度神经网络，建立基于数据驱动的轴箱轴承剩余寿命预测方法。研究成果有助于提升高速动车组机械设备关键部件故障检测、诊断与预警水平，为制定高铁“视情”维修决策提供理论依据和技术支撑。

本项目主要以列车轴箱轴承为对象，从动力学建模分析与故障诊断正、反问题两个方面进行了研究，主要内容包括：①构建了含典型缺陷故障的高速列车轴箱轴承非线性动力学模型，通过仿真分析揭示了不同尺寸故障下轴承动力学行为响应规律，为后续开展轴承故障诊断方法研究奠定了理论基础；②开展了强背景噪声下轴承故障信号降噪滤波方法研究，提出了基于改进高斯拉普拉斯算子和基于自适应经验小波塔式分解的信号降噪预处理方法，探明了信号中噪声成分与故障冲击成分之间相互影响的耦合机制，在提高信号信噪比基础上，突出了信号中的有用成分；③进行了列车轴承复杂故障诊断相关方法研究，分别提出了基于自适应频率窗经验小波变换、自适应Morlet小波参数字典设计、故障冲击匹配参数字典设计、融合SVD降噪及自适应TMSST的轴承微弱故障特征提取方法，上述新方法从多个角度探讨了复杂轮轨激励下轴承信号中多种干扰分量与故障冲击成分的相互耦合作用关系，能够较为有效地用于列车轴承微弱故障特征的提取；④采用深度学习理论，建立了基于数据驱动的轴承剩余寿命预测与状态监测的网络模型，提出了基于多尺度特征融合与轻量化结构的深度神经网络模型并用于轴承故障智能分类识别研究，增强了传统神经网络复杂特征信息的学习能力，提升了网络模型的运行效率，同时提出了基于多尺度膨胀卷积网络模型的轴承寿命预测方法，可以较为准确地预测轴承剩余使用寿命。本项目的上述研究成果，具有较为广阔的应用前景，拓展并丰富了列车轴承故障诊断研究方法，降低走行部关键部件运维成本，为保障高速列车安全稳定运行提供理论指导和技术支持。