摘要
针对轨道车辆压装车轴镶入部在定期超声波检测时噪声信号幅值较大、缺陷信号被淹没在噪声信号中无法直接识别的状况,基于超声波检测信号采集数据分析信号典型特征,结果表明有较高辨识度的缺陷信号具有能量集中、信噪比高且快速收敛的特性。采用小包波变换对采集的压装后的车轴镶入部超声波检测信号进行分解,提取区别于噪声信息的缺陷特征系数和缺陷幅值系数,以此重建信号波形;再采用基于logistic函数的自寻优阈值滤波改进算法提高缺陷信号信噪比,以检测淹没在噪声信号中的缺陷信号,并基于神经网络进行验证。结果表明:基于缺陷特征系数和缺陷幅值系数的重建信号波形,可以准确区分超声波检测数据中噪声信号和含有疲劳裂纹的缺陷信号,对于缺陷深度大于0.5 mm的裂纹缺陷,识别准确率为100%。
In view of the situation that the amplitude of noise signal is large and the defect signals are submerged in the noise signals and cannot be directly identified during the regular ultrasonic detection for the insertion part of the press-fit axle of track vehicle,the typical characteristics of the signal are analyzed based on the acquired data from the ultrasonic detection.Analysis results show that the defect signals with high recognition degrees have the characteristics of concentrated energy,high signal-to-noise ratio and fast convergence.The wavelet packet transformation is used to decompose the signals collected from the ultrasonic detection for the insertion part of the press-fit axle.The defect characteristic coefficients and defect amplitude coefficients,which are different from the noise information,are extracted to reconstruct the signal waveform.Then,an improved selfoptimizing threshold filter algorithm based on logistic function is used to improve the signal-to-noise ratio of reconstructed signals so that the submerged defect signals can be distinguish from noise signals,and is verified by neural network.Results show that the reconstructed signal waveform based on the defect characteristic coefficient and the defect amplitude coefficient can accurately distinguish the noise signals and the defect signals containing fatigue cracks in the ultrasonic detection data.For the crack defects with the depth greater than 0.5 mm,the identification accuracy is 100%.
作者
武冬冬
杨凯
彭朝勇
WU Dongdong;YANG Kai;PENG Chaoyong(Research and Develop Center,Chengdu Leading Software Technology Co.,Ltd.,Chengdu Sichuan 610091,China;School of Physical Science and Technology,Southwest Jiaotong University,Chengdu Sichuan 610031,China)
出处
《中国铁道科学》
EI
CAS
CSCD
北大核心
2021年第3期121-126,共6页
China Railway Science
基金
国家自然科学基金资助项目(61501381)。
关键词
无损检测
信息提取
压装车轴
裂纹缺陷
超声波
神经网络
Nondestructive test
Information extraction
Press-fit axle
Crack defect
Ultrasonic
Neural network
