| 基于残差卷积网络的浮置板轨道钢弹簧损伤检测 |
| |
| 引用本文： | 朱胜阳,张庆铼,袁站东,翟婉明.基于残差卷积网络的浮置板轨道钢弹簧损伤检测[J].交通运输工程学报,2022,22(2):123-135. |
| |
| 作者姓名： | 朱胜阳 张庆铼 袁站东 翟婉明 |
| |
| 作者单位： | 西南交通大学 牵引动力国家重点实验室，四川 成都 610031 |
| |
| 基金项目： | 国家自然科学基金项目51978587国家自然科学基金项目11790283牵引动力国家重点实验室自主课题2019TPL-T16 |
| |
| 摘 要： | 针对传统故障诊断方法难以有效检测浮置板轨道钢弹簧损伤这一挑战性问题，提出了一种基于一维残差卷积网络的损伤检测方法；建立了车辆-浮置板轨道耦合动力学模型，得到了多种工况下列车通过导致的浮置板振动响应数据集；利用残差卷积网络对不同损伤情形下的振动响应进行特征提取和数据分类，实现了对损伤钢弹簧的准确定位；研究了残差卷积网络在不同传感器布置方案上的检测性能，分析了损伤钢弹簧和传感器之间的复杂位置关系对检测性能的影响规律，优化并确定了经济可靠的传感器布置方案。研究结果表明:传感器的位置越靠近浮置板中部，残差卷积网络对不同损伤情形下的数据分类准确性和鲁棒性越好；传感器的布置数量增多，损伤检测方法的性能也随之改善，但传感器过多地集中于浮置板中部并不会带来显著的性能提升；在浮置板中部的钢弹簧损伤比在浮置板端部的钢弹簧损伤更难识别；损伤检测方法在全覆盖式布置方案下达到了99.11%的分类准确率，对复杂多变的检测情景具有良好适应性，而优化后双传感器布置方案和三传感器布置方案的分类准确率分别达到了98.23%和98.96%，优化后传感器布置方案具有良好的检测性能，同时也保持了损伤检测方法对复杂情景的适应性。
|
| 关 键 词： | 车辆-轨道耦合动力学 浮置板轨道 损伤检测 卷积神经网络 残差学习 传感器布置 |
| 收稿时间： | 2021-10-21 |
Damage detection for floating-slab track steel-spring based on residual convolutional network |
| |
| ZHU Sheng-yang,ZHANG Qing-lai,YUAN Zhan-dong,ZHAI Wan-ming.Damage detection for floating-slab track steel-spring based on residual convolutional network[J].Journal of Traffic and Transportation Engineering,2022,22(2):123-135. |
| |
| Authors: | ZHU Sheng-yang ZHANG Qing-lai YUAN Zhan-dong ZHAI Wan-ming |
| |
| Affiliation: | State Key Laboratory of Traction Power, Southwest Jiaotong University, Chengdu 610031, Sichuan, China |
| |
| Abstract: | As traditional fault diagnosis methods can hardly effectively detect the steel-spring damage of floating-slab track (FST), a damage detection method based on the one-dimensional residual convolutional network was proposed. A vehicle-FST coupled dynamics model was built, and the data sets for the floating-slab vibration response caused by the passing vehicles under various conditions were generated. The residual convolutional network was utilized for the feature extraction and data classification of the vibration response under different damage scenarios to achieve the accurate positioning of damaged steel springs. The detection performance of the residual convolutional network on different sensor deployment schemes were studied. The influence of the complex positional relationship between the damaged steel springs and the sensors on the detection performance was analyzed, and the economic and reliable sensor deployment schemes were optimized and determined. Analysis results reveal that when the sensors are closer to the middle of the floating-slab, better classification accuracy and robustness of the residual convolutional network can be achieved on the data under different damage scenarios. As the number of sensors increases, the detection performance of the method also improves, but the excessive concentration of the sensors in the middle of the floating-slab will not bring about significant improvement on the performance. The damage of steel-springs in the middle of the floating-slab is more difficult to identify than that at the end of the floating-slab. The damage detection method achieves a classification accuracy of 99.11% on the full-coverage deployment scheme, boasting good adaptability to complex and changeable detection scenarios. The classification accuracies of the optimized two-sensor deployment scheme and three-sensor deployment scheme reach 98.23% and 98.96%, respectively. The optimized sensor deployment schemes have good detection performance and keep the adaptability of the damage detection method to complex scenarios. 4 tabs, 16 figs, 30 refs. |
| |
| Keywords: | |
|
| 点击此处可从《交通运输工程学报》浏览原始摘要信息 |
|
点击此处可从《交通运输工程学报》下载全文 |
|
