基于模拟声辐射信号的桥上板式轨道脱空状态智能感知方法

Intelligent Perception Method for Delamination of Cement Emulsified Asphalt Mortar in Slab Tracks on Bridges Using Simulated Acoustic Radiation Signals

摘要:

提出了一种基于轨旁传声器采集结构声辐射信号的板式轨道脱空状态智能感知方法。建立了车-轨-桥耦合振动计算模型和声振耦合分析模型，模拟了列车动载激励下轨道板和桥梁结构的振动和声辐射响应，分析了轨道板脱空状态对结构振动和声辐射响应的影响规律，采用声辐射数值模拟数据和支持向量机（SVM）实现了对轨道板15种脱空状态的二分类和多分类识别。结果表明：相比于位移响应，加速度响应和声辐射响应对轨道板脱空状态的变化较为敏感；二分类SVM模型对于不同测点数据的分类效果有所差别，但准确率基本都能达到85 %以上；根据某测点声压数据训练出的二分类SVM模型对未知测点数据的分类准确率相比于自身测点数据下降10 %~30 %；多点位数据信息融合可以提高多分类识别准确率。

Abstract:

This paper proposes an intelligent perception method for detecting delamination of cement emulsified asphalt （CA） mortar in slab tracks based on the structure-borne sound signals collected by trackside acoustic sensors. A vehicle-track-bridge coupled vibration calculation model and an acoustic-vibration coupling analysis model are established to simulate the vibration and acoustic radiation response of the slab tracks and bridge structures under the dynamic loads caused by passing trains. The influence of CA mortar delamination on the vibration and acoustic radiation response is analyzed. By using simulated acoustic data and support vector machines （SVM）， binary and multi-class classification recognition of 15 types of CA mortar delamination are implemented. The results show that compared with displacement response， acceleration response and acoustic radiation response are more sensitive to CA mortar delamination. The classification performance of the binary SVM model varies for different measurement points， but the accuracy can generally reach over 85 %. The classification accuracy of the binary SVM model trained based on the sound pressure data at a specific measurement point decreases by 10 % to 30 % for unknown measurement points compared with that for the specific measurement points. The fusion of multi-point position data can improve the accuracy of multi-class classification recognition.