轮轨激励下高速动车组轮对轴承非线性振动机理与故障诊断

As a key component of the high-speed electric multiple unit (EMU), wheelset bearing’s operation plays an important role for its stability and safety. However, affected by the wheel-rail excitation and complex vibration transmission, it is difficult to accurately detect the incipient faults of the bearing with traditional methods. To solve this problem, a nonlinear bearing-wheelset model under typical wheel-rail excitation is firstly proposed to analysis its dynamic behavior to reveal the inherent relationship between the fault mechanism and fault feature. Then, some experiments will be conducted on two full-size wheelset bearing test benches to correct some model parameters and verify some nonlinear phenomena. At last, the dynamic model is utilized to simulate massive vibration responses of faulty bearings to train a deep convolution neural network, which can adaptively extract the weak fault features and accurately achieve qualitative/quantitative diagnosis under strong interference. The research results of this project can further deepen the significance of fault mechanism to fault diagnosis, solve the intelligent diagnosis problem with few failure samples and directly guide the design of on-board wheelset bearing monitoring system to provide supports for the running safety of high-speed EMU.

轮对轴承是走行部中的关键运动部件，对高速动车组的运行稳定性和安全性具有非常重要的影响，受复杂激励环境、振动传递路径等因素的影响，其早期微弱故障难以准确诊断。本项目拟从轮轨激励下轮对轴承的非线性振动机理出发，研究早期故障影响下的动力学特征和有效表征手段，揭示故障轮对轴承动力学行为的内在本质及特征规律；基于铁路轴承综合性能实验台和高速列车单轴滚振实验台开展系列针对性的实验研究，检验、修正动力学模型参数，验证理论分析结论；在此基础上，利用该模型模拟海量故障轮对轴承的振动响应作为训练样本，基于深度学习和仿真驱动的思想训练深度卷积神经网络，实现强干扰下轮对轴承微弱故障特征的自适应提取和早期故障的精准定性/定量诊断。本课题的研究成果可进一步深化故障机理对故障诊断的指导意义，解决小样本下的智能诊断问题，指导车载监测系统设计，保障高速动车组的运行安全。

轮对轴承是高速列车走行部中的关键旋转部件，在复杂轮轨激励的作用下极易出现由疲劳、过载等原因导致的失效，影响列车的行车安全，亟待进一步发展轮对轴承健康监测和诊断技术。针对该问题，本项目首先建立了轮对轴承-车辆系统耦合动力学模型，基于Simulink/UM联合仿真，研究了在轴承内、外圈存在局部缺陷时系统的动力学响应以及轨道不平顺、车轮不圆度对系统的振动响应的影响，并基于共振解调分析了轴箱上提取故障冲击的最佳测点位置。其次，应用铁路轴承综合性能实验台、高速列车单轴滚振实验台开展了轮对轴承的台架实验，并参与了某新型转向架的线路实验研究。这些实验结果一方面检验和修正了动力学模型中的参数，另一方面为研究轮对轴承故障诊断方法提供了数据支撑。最后，基于小波变换、解卷积等信号处理技术，提出了结合冲击性和循环平稳性的轮对轴承多目标故障特征提取方法，并针对基于深度学习的智能诊断问题，研究了网络结构优化、多源信息融合、基于边缘分布/联合分布自适应的迁移诊断等问题。项目研究成果为揭示轮轨激励下轮对轴承的故障机理、实现轮对轴承高精度的故障诊断提供了研究手段和技术支撑。