A Method for Estimating Damage Degree of Wheel Flat Scars Based on Time-frequency Energy Spectrum and VGG16

doi:10.3969/j.issn.1004-132X.2023.16.003

Abstract

Abstract: In order to accurate monitoring of the damage degree of wheel flat scars of vehicles in operation， a method for estimating the damage degree of wheel flat scars was proposed based on the combination of time-frequency energy spectrum and VGG16 convolutional neural network. This method might quantitatively estimate the damage degree of wheel flat scars in real time by analyzing and processing the vibration acceleration signals of axle box during vehicle operation. The dynamics models of rigid flexible coupling system of vehicle tracks and the mathematical models of wheel flat scars were established to simulate the vibration response of the vehicle axle box under different flat scar damage conditions. Using morphological filter and CEEMDAN-WVD time-frequency analysis method， the vibration acceleration signals of the axle box were filtered and reduced， and then expressed in the time-frequency energy spectrum. The VGG16 convolutional neural network models were constructed， and the training sets were constructed by using the time-frequency energy spectrum of a large number of wheel flat scar fault data to train the VGG16 models. Several wheel flat scar conditions were randomly simulated and the VGG16 models were tested and verified. The simulation tests show that the method combining time-frequency energy spectrum with VGG16 models may accurately estimate the damage degree of wheel flat scars of vehicles in operation， and the estimation errors are within 1.6 mm.

Key words: wheel flat scar, morphological filtering, complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), Wigner-Ville distribution（WVD）, visual geometry group 16（VGG16）, time-frequency energy spectrum

摘要： 为了实现对运营中车辆车轮扁疤损伤程度的实时精准监测，提出了一种时频能量谱与VGG16卷积神经网络相结合的车轮扁疤损伤程度估计方法，该方法通过对车辆运营中轴箱振动加速度信号的分析处理来实时定量估计车轮扁疤的损伤程度。建立了车辆轨道刚柔耦合系统动力学模型和车轮扁疤数学模型，仿真计算不同扁疤损伤工况下的车辆轴箱振动响应。运用形态学滤波器以及完全噪声辅助集合经验模态分解结合Wigner-Ville分布的时频分析方法，将轴箱振动加速度信号滤波降噪后表达在时频能量谱中。构造了VGG16卷积神经网络模型，通过大量车轮扁疤故障数据的时频能量谱构造的训练集来训练VGG16模型。随机仿真若干车轮扁疤工况，对训练完善的VGG16模型进行测试验证。仿真试验表明，运用时频能量谱与VGG16模型结合的方法能准确地估计运营中车辆的车轮扁疤损伤程度，估计误差在1.6 mm内。

关键词: 车轮扁疤, 形态学滤波, 完全噪声辅助聚合经验模态分解, Wigner-Ville分布, VGG16, 时频能量谱

CLC Number:

U279.2

LI Dazhu, NIU Jiang, LIANG Shuling, CHI Maoru. A Method for Estimating Damage Degree of Wheel Flat Scars Based on Time-frequency Energy Spectrum and VGG16[J]. China Mechanical Engineering, 2023, 34(16): 1907-1914.

李大柱, 牛江, 梁树林, 池茂儒. 时频能量谱与 VGG16结合的车轮扁疤损伤程度估计方法[J]. 中国机械工程, 2023, 34(16): 1907-1914.

References

［1］翟婉明. 车辆-轨道耦合动力学［M］. 北京：科学出版社， 2015.
ZHAI Wanming. Vehicle Track Coupling Dynamics［M］. Beijing：Science Press， 2015.
［2］翟婉明. 铁路车轮扁疤的动力学效应［J］. 铁道车辆， 1994（7）：1-5.
ZHAI Wanming. Dynamic Effect of Railway Wheel Flat Scar［J］. Railway Rolling Stock， 1994（7）：1-5.
［3］RAJKOMAR A， LINGAM S， TAYLOR A G， et al. High-throughput Classification of Radiographs Using Deep Convolutional Neural Networks［J］. Journal of Digital Imaging， 2017， 30（1）：1-7.
［4］CAI M， LIU J， CAI M， et al. Maxout Neurons for Deep Convolutional and LSTM Neural Networks in Speech Recognition［J］. Speech Communication， 2016， 77（3）：53-64.
［5］KRIZHEVSKY A， SUTSKEVER I， HINTONG E. Image Net Classification with Deep Convolutional Neural Networks［J］. Communications of the ACM, 2017, 60(6):84-90.
［6］李恒，张氢，秦仙蓉，等. 基于短时傅里叶变换和卷积神经网络的轴承故障诊断方法［J］. 振动与冲击， 2018， 37（19）：124-131.
LI Heng， ZHANG Qing， QIN Xianrong， et al. Bearing Fault Diagnosis Method Based on Short-time Fourier Transform and Convolutional Neural Network［J］. Vibration and Shock， 2018， 37（19）：124-131.
［7］孟强斌. 基于时频图和卷积神经网络的水电机组故障诊断研究［D］. 西安：西安理工大学， 2020.
MENG Qiangbin. Research on Fault Diagnosis of Hydropower Units Based on Time-frequency Diagram and Convolutional Neural Network［D］. Xian：Xian University of Technology， 2020.
［8］张雪嘉. 基于经验小波与改进卷积神经网络的风机故障诊断方法研究［D］. 北京：北京化工大学， 2021.
ZHANG Xuejia. Research on Fan Fault Diagnosis Method Based on Empirical Wavelet and Improved Convolutional Neural Network［D］. Beijing：Beijing University of Chemical Technology， 2021.
［9］李大柱，牛江，梁树林，等. 基于多尺度时频图与卷积神经网络的车轮故障智能诊断［J］. 铁道科学与工程学报，2023， 20（3）：1032-1043.
LI Dazhu， NIU Jiang， LIANG Shulin， et al. Intelligent Wheel Fault Diagnosis Based on Multi-scale Time-frequency Map and Convolutional Neural Network［J］. Journal of Railway Science and Engineering，2023， 20（3）：1032-1043.
［10］许文天. 铁道车辆车轮扁疤在线实时定量监测方法研究［D］. 成都：西南交通大学， 2020.
XU Wentian. Research on On-line Real-time Quantitative Monitoring Method for Railway Vehicle Wheel Flat Scar［D］. Chengdu：Southwest Jiaotong University， 2020.
［11］李奕璠，刘建新，李忠继. 基于Hilbert-Huang变换的列车车轮失圆故障诊断［J］. 振动. 测试与诊断， 2016， 36（4）：734-739.
LI Yichen， LIU Jianxin， LI Zhongji. Train Wheel out of Round Fault Diagnosis Based on Hilbert Huang Transform［J］. Vibration. Test and Diagnosis， 2016， 36（4）：734-739.
［12］张俊甲，马增强，王建东，等. 变分模态分解和形态学滤波在滚动轴承故障诊断中的应用［J］. 石家庄铁道大学学报（自然科学版）， 2018， 31（4）：52-57.
ZHANG Junjia， MA Zengqiang， WANG Jiandong， et al. Application of Variational Modal Decomposition and Morphological Filtering in Rolling Bearing Fault Diagnosis［J］. Journal of Shijia-zhuang Railway University（Natural Science Edition）， 2018， 31（4）：52-57.
［13］李兵，张培林，米双山，等. 机械故障信号的数学形态学分析与智能分类［M］. 北京：国防工业出版社， 2011.
LI Bing， ZHANG Peilin， MI Shuangshan， et al. Mathematical Morphology Analysis and Intelligent Classification of Mechanical Fault Signals［M］. Beijing：National Defense Industry Press， 2011.
［14］蔡艳平，李艾华，石林锁，等. 基于EMD-WVD振动谱时频图像SVM识别的内燃机故障诊断［J］. 内燃机工程， 2012， 33（2）：72-78.
CAI Yanping， LI Aihua， SHI Linsuo， et al. Internal Combustion Engine Fault Diagnosis Based on EMD-WVD Vibration Spectrum Time-frequency Image SVM Recognition［J］. Internal Combustion Engine Engineering， 2012， 33（2）：72-78.
［15］孙国栋，王俊豪，徐昀，等. CEEMD-WVD多尺度时频图像的滚动轴承故障诊断［J］. 机械科学与技， 2020， 39（5）：688-694.
SUN Guodong， WANG Junhao， XU Yun， et al. Rolling Bearing Fault Diagnosis Based on CEEMD-WVD Multi-scale Time-frequency Image［J］. Mechanical Science and Technology， 2020， 39（5）：688-694.
［16］张梅军，唐建，何晓辉. EEMD方法及其在机械故障诊断中的应用［M］. 北京：国防工业出版社， 2015.
ZHANG Meijun， TANG Jian， HE Xiaohui. EEMD Method and Its Application in Mechanical Fault Diagnosis［M］. Beijing：National Defense Industry Press， 2015.
［17］李中，卢春华，王星，等. 考虑滚动轴承故障位置与损伤程度的双分支卷积神经网络故障诊断方法［J］. 科学技术与工程， 2022， 22（4）：1441-1448.
LI Zhong， LU Chunhua， WANG Xing， et al. Double Branch Convolution Neural Network Fault Diagnosis Method Considering Fault Location and Damage Degree of Rolling Bearing［J］. Science， Technology and Engineering， 2022， 22（4）：1441-1448.
［18］黄晋，李保强，吕明燕，等. 基于VGG卷积神经网络的机场能见度预测［J］. 集成电路应用， 2022， 39（3）：58-59.
HUANG Jin， LI Baoqiang， LYU Mingyan， et al. Airport Visibility Prediction Based on VGG Convolutional Neural Network［J］. Integrated Circuit Applications， 2022， 39（3）：58-59.
［19］夏坚，周利君，张伟. 基于迁移学习与VGG16深度神经网络的建筑物裂缝检测方法［J］. 福建建设科技， 2022（1）：19-22.
XIA Jian， ZHOU Lijun， ZHANG Wei. Building Crack Detection Method Based on Transfer Learning and VGG16 Deep Neural Network［J］. Fujian Construction Technology， 2022（1）：19-22.

[1]	Liu Siyuan, Yang Mengxue, Wang Chuang, Zhang Wenwen. Morphological Filtering Method for Vibration Signals of Hydraulic Pump under Sinusoidal Variable Load Conditions [J]. China Mechanical Engineering, 2016, 27(19): 2557-2563.
[2]	LV Bo, ZHANG Yong-Xiang, KE Wei. #br# Research on Measure Parameter Optimization Method of a Difference Morphological Filtering Based on Genetic Algorithm [J]. China Mechanical Engineering, 2013, 24(3): 355-359.
[3]	Liu Wenyi;Tang Baoping;Chen Renxiang;Jiang Yonghua. Method to Suppress the Cross terms in Wigner-Ville Distribution Using Auto Terms [J]. J4, 2009, 20(21): 0-2531.
[4]	Cheng Fabin;Tang Baoping;Liu Wenyi. A Method to Suppress Cross-terms of Wigner-Ville Distribution and  Its Application in Fault Diagnosis [J]. J4, 2008, 19(14): 0-1763.