Wheel-rail force is an essential indicator for evaluating rail vehicles’ operational safety. This study combined the traditional time domain identification method for dynamic loads and the machine learning method to rapidly identify the wheel-rail forces for rail vehicles and evaluate the operational safety of rail vehicles. Firstly, a vertical dynamic model of rail vehicles considering the nonlinearity of partial suspension parameters was established. The acceleration responses of the axle box and frame under random excitation of the track were calculated by the Runge–Kutta method, with the random noise simulating the actual observation noise as the input for the wheel-rail force identification model. Secondly, in response to the difficulty of identifying lateral wheel-rail forces for rail vehicles, a nonlinear autoregressive with exogenous inputs (NARX) model was introduced to construct relationships between vehicle component responses and lateral wheel-rail forces. Then, the influences of input responses and model parameters on the identification results of lateral wheel-rail forces were analyzed, and the lateral wheel-rail forces under straight and curved negotiation were trained and identified. Finally, a comprehensive identification method of wheel-rail forces for rail vehicles based on time domain and machine learning methods was proposed, which can identify the vertical and lateral wheel-rail forces of each wheelset’s left and right wheels. The operational safety indicators, derailment coefficient and wheel unloading rate were verified with a simulation model. The Pearson correlation coefficients were not less than 0.96 and 0.95 under straight-line conditions and not less than 0.76 and 0.85 under curved conditions.

中文翻译：

---

基于时域和机器学习方法的轨道车辆轮轨力综合辨识


轮轨力是评价轨道车辆运行安全性的重要指标。本研究将传统的动载荷时域辨识方法与机器学习方法相结合，快速辨识轨道车辆的轮轨力，评估轨道车辆的运行安全性。首先，建立了考虑部分悬架参数非线性的轨道车辆垂向动力学模型。采用龙格-库塔法计算了轴箱和车架在轨道随机激励下的加速度响应，并以模拟实际观测噪声的随机噪声作为轮轨力识别模型的输入。其次，针对轨道车辆轮轨侧向力识别的困难，引入外生输入非线性自回归（NARX）模型来构建车辆部件响应与轮轨侧向力之间的关系。然后，分析输入响应和模型参数对轮轨横向力识别结果的影响，并对直线和曲线通过情况下的轮轨横向力进行训练和识别。最后提出了一种基于时域和机器学习方法的轨道车辆轮轨力综合识别方法，能够识别各轮组左右车轮的垂直和横向轮轨力。通过仿真模型验证了运行安全指标、脱轨系数和车轮卸载率。直线条件下Pearson相关系数不小于0.96和0.95，曲线条件下不小于0.76和0.85。