基于K-means++聚类分析的轮轨垂向力基线漂移预处理研究

PDF(4111 KB)

振动与冲击 ›› 2025, Vol. 44 ›› Issue (9) : 127-134.

振动理论与交叉研究

基于K-means++聚类分析的轮轨垂向力基线漂移预处理研究

施亦非，王锋*，石佳，黄宇峰

作者信息 +

Pre-processing of baseline drift of wheel-rail vertical force based onK-means++ clustering analysis

SHI Yifei, WANG Feng*, SHI Jia, HUANG Yufeng

Author information +

文章历史 +

摘要

采集轮轨垂向力等强冲击能量的振动信号时，受传感器特性和环境影响，测得信号中存在基线漂移，严重影响后续数据分析处理。曲线拟合和密度聚类是修正基线漂移的常见方法，通过选取特定信号区间作为基点进行拟合，可有效去除基线漂移；然而，由于基点选取极度依赖先验知识，限制了其应用范围。为解决该问题，提出一种基于K-means++聚类分析的轮轨垂向力基线漂移预处理方法。首先，选取基尼系数和方差，在欧氏空间准确表征载荷与无载荷数据段的差异，进而引导K-means++聚类；随后，基于K-means++聚类选取无载荷数据段，量化信号的基线漂移干扰；最后，以无载荷数据段为基点，拟合并修正基线漂移。经过仿真和实测数据分析，与最小二乘法 (LS) 、经验模态分解 (EMD) 和密度聚类 (DBSCAN) 相比，该方法在信噪比 (SNR) 、均方误差 (MSE) 、基线去除误差 (BDE) 和运行时间等方面均有一定优势。结果表明，基于基尼系数和方差的K-means++聚类分析，克服了密度聚类分析的先验知识依赖，可有效修正轮轨垂向力基线漂移，有望用于其他强冲击能量振动信号的数据预处理。

Abstract

When sampling vibration signals with strong impact energy, such as wheel-rail vertical force, vibration signals exhibit baseline drift due to sensor characteristics and environmental influence. This drift seriously affects the subsequent data analysis and processing. While curve fitting and density clustering algorithms are established methods for correcting baseline drift. By selecting specific signal intervals as fitting baselines, baseline drift is effectively removed. However, the selection of baselines heavily relies on prior knowledge, thereby limiting the applicability of these algorithms. To address this limitation, a novel preprocessing method based on K-means++ for the wheel-rail vertical force is proposed. Firstly, the Gini index and Variance were selected to quantify the distinctions between loaded and unloaded data segments in the Euclidean space, thereby guiding K-means++ clustering. Secondly, K-means++ clustering was approached to select the unloaded data segments, quantifying the baseline drift interference of the signal. Finally, the unloaded data segments were chosen as the reference points for correcting the baseline drift. Both the simulated and measured data show that this approach offers advantages in terms of Signal-to-noise Ratio (SNR), Mean Squared Error (MSE), and runtime in comparison to other methods like Least Squares (LS), Empirical Mode Decomposition (EMD), and Density-Based Spatial Clustering of Applications with Noise (DBSCAN). These results indicate that the K-means++ algorithm along with the Gini index and Variance, mitigates the reliance on prior knowledge inherent in density clustering analysis. The method effectively corrects the baseline drift in wheel-rail vertical force and is potentially applicable to preprocessing other high-energy vibration signals.

导出引用

施亦非, 王锋, 石佳, 黄宇峰. 基于K-means++聚类分析的轮轨垂向力基线漂移预处理研究[J]. 振动与冲击, 2025, 44(9): 127-134

SHI Yifei, WANG Feng, SHI Jia, HUANG Yufeng. Pre-processing of baseline drift of wheel-rail vertical force based onK-means++ clustering analysis[J]. Journal of Vibration and Shock, 2025, 44(9): 127-134

参考文献

[1] 王祉歆, 周伟, 王雅昭, 等. 轮轨作用力与垂向载荷位置的轨旁识别法研究[J]. 铁道科学与工程学报, 2023: 1-12.
WANG Zhixin, ZHOU Wei, WANG Yazhao, et al. Wayside identification method for railway wheel rail force and vertical load position[J]. Journal of Railway Science and Engineering, 2023: 1-12.
[2] 王明猛, 朱涛, 王小瑞, 等. 一种逆结构滤波法的轨道车辆轮轨力识别[J]. 振动工程学报, 2019, 32(4): 602-608.
WANG Mingmeng ,ZHU Tao, WANG Xiaorui, et al. An inverse structural filter method for wheel-rail contact forces identification of railway vehicles[J]. Journal of Vibration Engineering, 2019, 32(4): 602-608.
[3] 赵小龙, 马铁华, 范锦彪. 高g值加速度计在高冲击下的失效特性的研究[J]. 传感技术学报, 2012, 25(12): 1668-1672.
ZHAO Xiaolong, MA Tiehua, FAN Jinbiao. The Study of Failure Characteristic of High g Acceleration in High Shock[J]. Chinese Journal of Sensors and Actuators, 2012, 25(12): 1668-1672.
[4] 刘庆杰, 黄辉, 雷晓燕. 基于集合经验模态分解和小波变换的轮轨力应变信号降噪[J]. 城市轨道交通研究, 2016, 19(11): 26-29+37.
LIU Qingjie, HUANG Hui, Lei Xiaoyan. Wheel/Rail Strain Signals Denoising by Using EEMD and Wavelet Transform[J]. Urban Mass Transit, 2016, 19(11): 26-29+37.
[5] 赵凤遥, 张建成, 葛巍, 等. 基于多项式拟合初值的动载荷识别修正算法[J]. 振动与冲击, 2021, 40(11): 211-219.
ZHAO Fengyao, ZHANG Jiancheng, GE Wei, et al. Dynamic load identification correction algorithm based on polynomial fitting initial value[J]. Journal of Vibration and Shock, 2021, 40(11): 211-219.
[6] Randall R B, Antoni J. Why EMD and similar decompositions are of little benefit for bearing diagnostics[J]. Mechanical Systems and Signal Processing, 2023, 192: 110207.
[7] 李晨, 梁书锋, 刘传鹏, 等. 基于改进的集合经验模态分解的爆破振动信号趋势项消除方法[J]. 北京理工大学学报, 2021, 41(6): 636-641.
LI Chen, LIANG Shufeng, LIU Chuanpeng, et al. Tiend Removing Method of Blasting Vibration Signals Based on MEEMD[J]. Transactions of Beijing Institute of Technology, 2021, 41(6): 636-641.
[8] 闫鹏, 张云鹏, 侯善营, 等. 基于改进EMD-小波包的爆破振动信号降噪方法研究[J]. 振动与冲击, 2024, 43(11): 264-271+287.
YAN Peng, ZHANG Yunpeng, HOU Shanying, et al. Noise reduction method for blasting vibration signals based on improved EMD-wavelet packet[J]. Journal of Vibration and Shock, 2024, 43(11): 264-271+287.
[9] Wu D, Zhang G, Zhu S, et al. A baseline drift removal algorithm based on cumulative sum and downsampling for hydroacoustic signal[J]. Measurement, 2023, 207: 112344.
[10] 单豪良, 陈永高, 孙泽阳. 基于DBSCAN算法的改进确定-随机子空间模态参数识别算法[J]. 振动与冲击, 2022, 41(11): 156-163.
SHAN Haoqiang, CHEN Yonggao, SUN Zeyang. Improved CDSI algorithm based on DBSCAN algorithm[J]. Journal of Vibration and Shock, 2022, 41(11): 156-163.
[11] De La Selle T, Weiss J, Deschanel S. Acoustic multiplets detection based on DBSCAN and cross-correlation[J]. Mechanical Systems and Signal Processing, 2024, 211: 111149.
[12] 农汉彪, 曾巧妮. 基于聚类分析的心电信号基线漂移去除方法[J]. 中国测试, 2022, 48(9): 22-28.
NONG Hanbiao, ZENG Qiaoni. Baseline wander removing of ECG signal based on clustering analysis[J]. China Measurement and Test, 2022, 48(9): 22-28.
[13] 钟志贤, 马李奕, 蔡忠侯, 等. 基于VMD_MPE和FCM聚类的变转速工况下转子不平衡故障诊断方法[J]. 振动与冲击, 2022, 41(14): 290-298.
ZHONG Zhixian, MA Liyi, CAI Zhonghou, et al. Method of rotor unbalance fault diagnosis under variable-speed conditions based on VMD_MPE and FCM clustering[J]. Journal of Vibration and Shock, 2022, 41(14): 290-298.
[14] Ay M, Özbakır L, Kulluk S, et al. FC-Kmeans: Fixed-centered K-means algorithm[J]. Expert Systems with Applications, 2023, 211: 118656.
[15] 官松泽, 唐钰本, 蔡争, 等. 基于Kmeans++-Bi-LSTM的太阳辐照度超短期预测[J]. 太阳能学报, 2023, 44(12): 170-174.
Guan Songze, Tang Yuben, Cai Zheng, et al. Ultra-Short-Term forecast of solor irradiance based on K-Means++-BI-LSTM[J]. Acta Energiae Solaris Sinica, 2023, 44(12): 170-174.
[16] CERIANI L, VERME P. The origins of the Gini index: extracts from Variabilità e Mutabilità (1912) by Corrado Gini[J]. The Journal of Economic Inequality, 2012, 10(3): 421-443.
[17] ATTILI F. Uncovering Complexities in Horizontal Inequality: A Novel Decomposition of the Gini Index[J]. Social Indicators Research, 2024, 173(2): 351-376.
[18] YU Y, ZHAO X. Separation of fault characteristic impulses of flexible thin-wall bearing based on wavelet transform and correlated Gini index[J]. Mechanical Systems and Signal Processing, 2024, 209: 111118.
[19] CHEN B, CHENG Y, ZHANG W, et, al. Investigations on improved Gini indices for bearing fault feature characterization and condition monitoring[J]. Mechanical Systems and Signal Processing, 2022, 176: 109165.
[20] HU Y, ZHANG B, TAN A C. Acceleration signal with DTCWPT and novel optimize SNR index for diagnosis of misaligned cardan shaft in high-speed train[J]. Mechanical Systems and Signal Processing, 2020, 140: 106723.
[21] WANG Z, BADIU M A, COON J P. On the Value of Information and Mean Squared Error for Noisy Gaussian Models[J]. IEEE Communications Letters, 2022, 26(9): 2023-2026.
[22] 李国权, 李必禄, 林金朝, 等. 基于改进EWT的心电信号基线矫正算法[J]. 仪器仪表学报, 2020, 41(4): 156-166.
LI Guoquan, LI Bilu, LIN Jinzhao, et al. Baseline correction algorithm for ECG signals based on improved EWT[J]. Chinese Journal of Scientific Instrument, 2020, 41(4): 156-166.