Identification of structural parameters of ballastless track based on SSA-CNN
-
摘要:
为获取无砟轨道在服役过程中的原位结构参数，结合无砟轨道有限元模型与数据驱动的麻雀搜索算法(SSA)-卷积神经网络(CNN)提出了轨道结构参数识别方法；建立了无砟轨道有限元模型，在给定参数空间内计算不同参数下的频响函数，形成数据集，并将70%的数据作为训练集，剩下的数据作为测试集；训练集频响函数作为输入，轨道结构参数作为输出，训练SSA-CNN参数识别模型，并通过测试集进行了验证；开展无砟轨道锤击试验，将实测频响函数输入参数识别模型，识别扣件刚度、阻尼和CA砂浆弹性模量，并将识别出的参数值代入轨道有限元模型，计算相同激励下的振动响应，计算结果与测试结果吻合良好。分析结果表明：通过含10%高斯噪声的数据集训练提出的参数识别模型，扣件刚度、阻尼和CA砂浆弹性模量识别结果的平均绝对百分误差分别为7.90%、1.00%、3.03%，验证了参数识别模型的可靠性；得到的参数识别方法可以利用锤击试验数据来准确识别轨道结构参数；在识别扣件参数或扣件损伤时应选用钢轨的振动响应，而识别CA砂浆层弹性模量或脱空时则应选择轨道板的振动数据；提出的无砟轨道结构参数识别方法可为无砟轨道层间连接构件的服役状态的检测和评估提供有效的分析工具。
Abstract:To obtain the in-situ structural parameters of ballastless track in service, a structural parameter identification method was developed by combining the finite element model of ballastless track with the data-driven sparrow search algorithm (SSA)-convolutional neural network (CNN). The finite element model of the ballastless track was established, and frequency response functions (FRFs) under different parameters were calculated within the given parameter space to form a dataset. 70% of the data was taken as the training set, and the remaining data was used as the test set. By using the FRFs of the training set as inputs and track's structural parameters as outputs, the SSA-CNN parameter identification model was trained and verified by the test set. A hammer test on the ballastless track was carried out, and the measured FRFs were input into the parameter identification model to obtain fastener stiffness, damping, and the elastic modulus of CA mortar. The identified parameter values were substituted into the track's finite element model to calculate the vibration response under the same excitation. The calculated results were in good agreement with the test results. Research results show that when the parameter identification model is trained with a dataset containing 10% Gaussian noise, the average absolute percentage errors for identifying fastener stiffness, damping, and the elastic modulus of CA mortar are 7.90%, 1.00%, and 3.03%, respectively, verifying the reliability of the parameter identification model. The track's structural parameters can be captured accurately using the parameter identification method in this paper and the hammer test data. The vibration response of the rail is beneficial to identifying parameters or damage of fastener, while the vibration data of the slab is suitable for identifying the elastic modulus or delamination of the CA mortar layer. The developed parameter identification method for ballastless track is an effective analytical tool for detecting and assessing the service performance of interlayer connection components in ballastless tracks.
-
图 5 不同测点振动响应对扣件刚度和CA砂浆层弹性模量灵敏度系数
Figure 5. Sensitivity coefficients of vibration responses at different measuring points to fastener stiffness and elastic modulus of CA mortar layer
图 11 S2实测与计算响应数据对比
Figure 11. Comparison of measured and calculated response data for S2
图 12 工况1实测与计算振动加速度对比
Figure 12. Comparison between measured and calculated vibration accelerations in case 1
图 13 工况2实测与计算振动加速度对比
Figure 13. Comparison between measured and calculated vibration accelerations in case 2
表 1 SSA-CNN算法中的超参数
Table 1. Hyper-parameters in SSA-CNN algorithm
麻雀坐标 超参数 随机初始化范围 x1 学习率 0.001~0.01 x2 迭代次数 10~50 x3 Batch size 16~256 x4 Conv1核大小 1~3 x5 Conv1核数量 1~20 x6 Conv2核大小 1~3 x7 Conv2核数量 1~20 x8 FC1全连接层神经元数量 1~50 x9 FC2全连接层神经元数量 1~50 
下载: 导出CSV
表 2 轨道结构参数
Table 2. Parameters of track structure
结构 类别 数值 结构 类别 数值 钢轨 弹性模量/Pa 2.1×1011 CA砂浆层 泊松比 0.47 密度/(kg·m-3) 7 800 密度/(kg·m-3) 1 400 截面面积/m2 7.745×10-3 宽度/m 1.2 截面惯性矩/m4 3.217×10-5 厚度/m 0.03 轨道板 弹性模量/Pa 3.6×1010 底座板 弹性模量/Pa 3.4×1010 密度/(kg·m-3) 2 500 密度/(kg·m-3) 2 500 泊松比 0.2 泊松比 0.2 宽度/m 1.2 宽度/m 1.2 厚度/m 0.2 厚度/m 0.3
下载: 导出CSV
表 3 待识别参数的取值范围
Table 3. Value range of parameters to be identified
扣件刚度/ (kN·mm-1) 扣件阻尼/ (kN·s·m-1) CA砂浆弹性模量/ GPa 20~70 30~100 0.01~0.50
下载: 导出CSV
表 4 CNN的最佳结构
Table 4. Optimal structure of CNN
序号 名称 最佳结构 无噪声 10%噪声 1 输入层 366×1×1 366×1×1 2 卷积层1 2×1×10 3×1×10 3 激活函数 ReLU函数 ReLU函数 4 卷积层2 3×1×12 2×1×2 5 激活函数 ReLU函数 ReLU函数 6 全连接层1 37个神经元 36个神经元 7 激活函数 ReLU函数 ReLU函数 8 全连接层2 23个神经元 29个神经元 9 激活函数 ReLU函数 ReLU函数 10 输出层 3×1×1 3×1×1 11 学习率 0.004 0 0.005 8 12 迭代次数 41 49 13 Batch size 173 92
下载: 导出CSV
表 5 试验仪器参数
Table 5. Parameters of test instruments
名称 型号 数量 其他参数 力锤 PCB力锤 1个 灵敏度为0.23 mV·N-1，量程为25 kN 采集设备 便携式Head Acoustics 1台 钢轨加速度传感器 PCB 1A102E 3个 灵敏度分别为1.089 0、1.038 0、0.957 2 mV·(m·s-2)-1 轨道板加速度传感器 PCB 333B32 3个 灵敏度分别为101.00、100.40、100.06 mV·(m·s-2)-1 扭矩扳手 1个 量程为150 N·m 数据传输线 振动传输线 6条
下载: 导出CSV
-
[1] XU J M, WANG P, AN B Y, et al. Damage detection of ballastless railway tracks by the impact-echo method[J]. Proceedings of the Institution of Civil Engineers-Transport, 2018, 171(2): 106-114. doi: 10.1680/jtran.16.00146 [2] CHUDZIKIEWICZ A, BOGACZ R, KOSTRZEWSKI M, et al. Condition monitoring of railway track systems by using acceleration signals on wheelset axle-boxes[J]. Transport, 2018, 33(2): 555-566. [3] CANTERO D, BASU B. Railway infrastructure damage detection using wavelet transformed acceleration response of traversing vehicle[J]. Structural Control and Health Monitoring, 2015, 22(1): 62-70. doi: 10.1002/stc.1660 [4] 缪炳荣, 刘俊利, 张盈, 等. 轨道车辆结构振动损伤识别技术综述[J]. 交通运输工程学报, 2021, 21(1): 338-357. doi: 10.19818/j.cnki.1671-1637.2021.01.016MIAO Bing-rong, LIU Jun-li, ZHANG Ying, et al. Review on structural vibration damage identification technology for railway vehicles[J]. Journal of Traffic and Transportation Engineering, 2021, 21(1): 338-357. doi: 10.19818/j.cnki.1671-1637.2021.01.016 [5] HOELZL C, DERTIMANIS V, ANCU L, et al. Vold-Kalman filter order tracking of axle box accelerations for track stiffness assessment[J]. Mechanical Systems and Signal Processing, 2023, 204: 110817. doi: 10.1016/j.ymssp.2023.110817 [6] BERGGREN E G, KAYNIA A M, DEHLBOM B. Identification of substructure properties of railway tracks by dynamic stiffness measurements and simulations[J]. Journal of Sound and Vibration, 2010, 329(19): 3999-4016. doi: 10.1016/j.jsv.2010.04.015 [7] KAEWUNRUEN S, REMENNIKOV A M. Field trials for dynamic characteristics of railway track and its components using impact excitation technique[J]. NDT & E International, 2007, 40(7): 510-519. [8] OREGUI M, LI Z L, DOLLEVOET R. An investigation into the vertical dynamics of tracks with monoblock sleepers with a 3D finite-element model[J]. Proceedings of the Institution of Mechanical Engineers, Part F: Journal of Rail and Rapid Transit, 2016, 230(3): 891-908. doi: 10.1177/0954409715569558 [9] LI Q, THOMPSON D J, TOWARD M G. Estimation of track parameters and wheel-rail combined roughness from rail vibration[J]. Proceedings of the Institution of Mechanical Engineers, Part F: Journal of Rail and Rapid Transit, 2018, 232(4): 1149-1167. doi: 10.1177/0954409717710126 [10] SHEN C, DOLLEVOET R, LI Z L. Fast and robust identification of railway track stiffness from simple field measurement[J]. Mechanical Systems and Signal Processing, 2021, 152: 107431. doi: 10.1016/j.ymssp.2020.107431 [11] LAM H F, ALABI S A, YANG J H. Identification of rail-sleeper-ballast system through time-domain Markov chain Monte Carlo-based Bayesian approach[J]. Engineering Structures, 2017, 140: 421-436. doi: 10.1016/j.engstruct.2017.03.001 [12] LE PEN L, MILNE D, THOMPSON D, et al. Evaluating railway track support stiffness from trackside measurements in the absence of wheel load data[J]. Canadian Geotechnical Journal, 2016, 53(7): 1156-1166. doi: 10.1139/cgj-2015-0268 [13] SHEN C, ZHANG P, DOLLEVOET R, et al. Evaluating railway track stiffness using axle box accelerations: A digital twin approach[J]. Mechanical Systems and Signal Processing, 2023, 204: 110730. doi: 10.1016/j.ymssp.2023.110730 [14] 胡俊亮, 余晓琳, 郑恒斌, 等. 基于优化BP神经网络的梁结构有限元模型修正[J]. 华南理工大学学报(自然科学版), 2013, 41(8): 67-73.HU Jun-liang, YU Xiao-lin, ZHENG Heng-bin, et al. Finite element model updating of beam structures based on the optimized BP neural network[J]. Journal of South China University of Technology (Natural Science Edition), 2013, 41(8): 67-73. [15] 万周, 刘璟泽, 张大海, 等. 基于LSTM的航空发动机整机支承刚度识别方法[J]. 东南大学学报(自然科学版), 2021, 51(4): 672-678.WAN Zhou, LIU Jing-ze, ZHANG Da-hai, et al. Identification method for support stiffness of whole aero-engine based on LSTM[J]. Journal of Southeast University (Natural Science Edition), 2021, 51(4): 672-678. [16] ZHANG X J, DENG Z M, ZHAO Y L. A frequency response model updating method based on unidirectional convolutional neural network[J]. Mechanics of Advanced Materials and Structures, 2021, 28(14): 1480-1487. doi: 10.1080/15376494.2019.1681037 [17] 莫叶, 王佐才, 丁雅杰, 等. 基于VMD和DBN的非线性结构模型参数识别[J]. 振动与冲击, 2022, 41(9): 136-143.MO Ye, WANG Zuo-cai, DING Ya-jie et al. Parametric recognition of nonlinear structural model based on VMD and DBN[J]. Journal of Vibration and Shock, 2022, 41(9): 136-143. [18] ZHANG Y Q, MIYAMORI Y, MIKAMI S, et al. Vibration-based structural state identification by a 1-dimensional convolutional neural network[J]. Computer-Aided Civil and Infrastructure Engineering, 2019, 34(9): 822-839. doi: 10.1111/mice.12447 [19] MIAO R, GAO Y T, GE L, et al. Online defect recognition of narrow overlap weld based on two-stage recognition model combining continuous wavelet transform and convolutional neural network[J]. Computers in Industry, 2019, 112: 103115. doi: 10.1016/j.compind.2019.07.005 [20] CHEN D H, ZHANG Y S, WAN R Z, et al. Indirect identification of bridge damage based on coupled vehicle-bridge vibration and 2D-CNN[J]. Measurement Science and Technology, 2024, 35(5): 055019. doi: 10.1088/1361-6501/ad2ad5 [21] AZIMI M, PEKCAN G. Structural health monitoring using extremely compressed data through deep learning[J]. Computer-Aided Civil and Infrastructure Engineering, 2020, 35(6): 597-614. doi: 10.1111/mice.12517 [22] TONDO G R, RAU S, KAVRAKOV L, et al. Stochastic stiffness identification and response estimation of Timoshenko beams via physics-informed Gaussian processes[J]. Probabilistic Engineering Mechanics, 2023, 74: 103534. doi: 10.1016/j.probengmech.2023.103534 [23] LIANG R H, LIU W F, FU Y G, et al. Physics-informed deep learning for structural dynamics under moving load[J]. International Journal of Mechanical Sciences, 2024, 284: 109766. doi: 10.1016/j.ijmecsci.2024.109766 [24] KRICHEN M. Convolutional neural networks: A survey[J]. Computers, 2023, 12(8): 151. doi: 10.3390/computers12080151 [25] XUE J K, SHEN B. A survey on sparrow search algorithms and their applications[J]. International Journal of Systems Science, 2024, 55(4): 814-832. doi: 10.1080/00207721.2023.2293687 [26] 任娟娟, 赵华卫, 李潇, 等. 减振CRTS Ⅲ型板式无砟轨道结构谐响应分析[J]. 铁道工程学报, 2016, 33(3): 44-50.REN Juan-juan, ZHAO Hua-wei, LI Xiao, et al. Analysis of harmonic response of CRTS Ⅲ prefabricated slab track with anti-vibration structure[J]. Journal of Railway Engineering Society, 2016, 33(3): 44-50. [27] 崔晓璐, 尹越, 包鹏羽, 等. 山地城市地铁制动/轨道结构参数对钢轨波磨的影响[J]. 表面技术, 2023, 52(1): 103-111.CUI Xiao-lu, YIN Yue, BAO Peng-yu, et al. Effect of the structure and parameter of brake/track on rail corrugation in a mountain city metro[J]. Surface Technology, 2023, 52(1): 103-111. [28] 韦凯, 王丰, 赵泽明, 等. 弹性分开式扣件弹性垫板静刚度测试评价方法[J]. 铁道工程学报, 2018, 35(11): 32-36, 86.WEI Kai, WANG Feng, ZHAO Ze-ming, et al. The methodology research on the test and evaluation of static stiffness of elastic pads in elastic separated fastener[J]. Journal of Railway Engineering Society, 2018, 35(11): 32-36, 86. -
计量
- 文章访问数: 272
- HTML全文浏览量: 129
- PDF下载量: 52
- 被引次数: 0
