Abstract:  Wheel polygonal wear posed a significant challenge to the safety of high-speed trains in China， and the accuracy of current on-board wheel polygonal wear monitoring was inadequate. Traditional double integration methods for diagnosing wheel polygonal wears had limitations， primarily due to their inability to effectively mitigate interference from rail surface roughness and lacked of consideration for the impact of natural modal resonances in the wheel-rail system. Addressing these issues， a frequency response function-based quantitative diagnostic method was proposed for the wheel polygonal wears. A comb filter was utilized to mitigate the impacts of rail roughness， isolating the “pure” wheel polygon response components， then the frequency response function was employed to adjust for inherent modal influences， allowing for a quantitative assessment of polygonal order and roughness level from axle box vibrations. It is verified by simulations and measured data， compared with the double integration methods， the proposed method may achieve better diagnosis results across all polygonal orders， the average estimated errors of wheel roughness grade are less than 3.5 dB， which is significantly lower than the maximum estimated error of 14 dB for the quadratic integral methods. The results show that the proposed method has broader adaptability and enhanced accuracy， and has certain scientific and engineering values. 

Key words: wheel polygonal wear, axle box vibration acceleration, on-board diagnostic, frequency response function, comb filter

摘要： 车轮多边形磨耗问题对我国高速动车组安全运营造成困扰，目前车轮多边形磨耗车载监测的准确率仍远不尽如人意。传统二次积分方法在诊断车轮多边形时存在局限，这主要是因为它们未能有效地排除来自钢轨粗糙度的干扰，同时也没有充分考虑轮轨系统固有模态对测量结果的影响。提出一种基于频率响应函数的车轮多边形磨耗定量诊断方法，利用梳状滤波器滤除钢轨粗糙度的影响，留下更“纯”的车轮多边形响应成分，然后利用频率响应函数修正了结构固有模态的影响，从而由轴箱振动信号定量诊断多边形阶次和粗糙级。经过仿真与实测数据验证，相比于二次积分方法，所提方法面对各阶次时均能取得更佳的诊断效果，对车轮粗糙级的平均估计误差不高于3.5 dB，明显小于二次积分方法的最高估计误差14 dB。研究结果表明，所提方法适应范围更广、准确性更高，具有一定的科学和工程意义。

关键词: 车轮多边形磨耗, 轴箱振动加速度, 车载诊断, 频响函数, 梳状滤波器