Train axle ultrasonic defect detection based on deep learning
-
摘要:
针对列车车轴超声检测中缺陷（特别是微小缺陷）检出率和检测效率低的问题,提出了一种基于深度学习的列车车轴缺陷超声检测方法,在YOLO v5s网络的基础上,改进特征提取层结构并加入SE注意力机制,采用实际车轴检测数据、CIVA仿真数据和GAN生成式数据构建了数据集,并进行了验证试验。试验结果表明,通过增加仿真数据和GAN生成式数据样本,所提方法在提高实际车轴超声检测缺陷检出率的同时,可有效检出微小车轴缺陷,其检出率可达99.25%,具有较高的应用价值和前景。
Abstract:Aiming at the low detection rate and slow efficiency of defects (especially minor defects) in train axle ultrasonic detection, a method based on deep learning was proposed. On the basis of YOLO v5s network, the feature extraction layer structure was improved and SE attention mechanism was added. The dataset was constructed using real axle detection data, CIVA simulation data and GAN generated data and validation experiments were conducted. The experimental results showed that by adding simulation data and GAN generated data samples, this proposed method can effectively improve the detection rate of the actual axle ultrasonic detection defects, and the detection rate reached 99.25%, which showed a high application value and prospect.
-
Keywords:
- deep learning /
- ultrasonic inspection /
- axle defect /
- data augmentation
-
-
Table 1 Backbone层网络参数
Stage（阶）i Operator（操作）Fi Resolution（分辨率）Hi×Wi Channels（通道）Ci Layers（层）Li 1 Conv3×3 640×640 32 1 2 EMblock,k3×3 320×320 16 1 3 EMblock,k3×3 320×320 24 2 4 EMblock,k5×5 160×160 40 2 5 EMblock,k3×3 80×80 80 3 6 EMblock,k5×5 40×40 112 3 7 EMblock,k5×5 40×40 192 4 8 EMblock,k3×3 20×20 320 1 9 Conv1×1&Pooling&FC 20×20 1 024 1 
下载: 导出CSV
Table 2 数据集扩增及分类结果
数据集 训练数据 测试数据 验证数据 Raw 120 100 20 Crop 1 200 1 000 300 GAN 2 400 0 600 Simulation 2 400 0 600 总计 6 120 1 100 1 520
下载: 导出CSV
Table 3 不同增强数据的识别准确率对比
数据集 准确率/% 召回率/% Real 94.49 91.38 Crop 98.73 98.94 Crop+Simulation 99.26 99.03 Crop+Simulation+GAN 99.31 99.73
下载: 导出CSV
Table 5 4种不同数据集在目标检测任务上的识别结果
项目 数据集 Real% Crop Crop+Simulation Crop+Simulation+GAN 检出率 96.00 97.75 98.00 99.25 漏检率 4.00 2.75 2.00 0.75 误报率 0 0.75 0 0
下载: 导出CSV
-
[1] WANG X Y ,LOU Z F ,WANG X D ,et al ．Prediction of stress distribution in press-fit process of interference fit with a new theoretical model[J]．Proceedings of the Institution of Mechanical Engineers,Part C：Journal of Mechanical Engineering Science,2019,233(8):2834-2846. [2] SHU Y L ,YANG G X ,LIU Z M ．Simulation research on fretting wear of train axles with interference fit based on press-fitted specimen[J]．Wear,2023,523:204777. [3] 彭朝勇,高晓蓉,王艾 ．车轴压装部相控阵超声波探伤的各向异性扩散去噪改进算法[J]．中国铁道科学,2017,38(3):77-82. [4] 周素霞,卢俊霖,吴毅,等 ．基于直流电位降的高铁车轴裂纹检测研究[J]．机械工程学报,2022,58(14):288-295. [5] 曹贞全 ．动车组空心车轴非接触式超声波检测设备的研究[J]．铁道机车与动车,2017(12):46-47,8. [6] MARSHALL M B ,LEWIS R ,DWYER-JOYCE R S ,et al ．Ultrasonic measurement of railway wheel hub–axle press-fit contact pressures[J]．Proceedings of the Institution of Mechanical Engineers,Part F：Journal of Rail and Rapid Transit,2011,225(3):287-298. [7] HE S Y ,HU D Y ,YU G ,et al ．Trackside acoustic detection of axle bearing fault using wavelet domain moving beamforming method[J]．Applied Acoustics,2022,195:108851. [8] ZHENG Z ,QI H Y ,ZHUANG L ,et al ．Automated rail surface crack analytics using deep data-driven models and transfer learning[J]．Sustainable Cities and Society,2021,70:102898. [9] LIN T Y ,GOYAL P ,GIRSHICK R ,et al ．Focal loss for dense object detection[J]．IEEE Transactions on Pattern Analysis and Machine Intelligence,2020,42(2):318-327. [10] LIU Z ,GU X Y ,WU W X ,et al ．GPR-based detection of internal cracks in asphalt pavement：a combination method of Deepaugment data and object detection[J]．Measurement,2022,197:111281. [11] YU H M ,LI Q Y ,TAN Y Q ,et al ．A coarse-to-fine model for rail surface defect detection[J]．IEEE Transactions on Instrumentation and Measurement,2019,68(3):656-666. [12] LEE H ,EUM S ,KWON H ．ME R-CNN：multi-expert R-CNN for object detection[J]．IEEE Transactions on Image Processing,2020,29:1030-1044. [13] QIAN Y ,LI X L ,ZHANG Q ,et al ．SPP-CPI：predicting compound-protein interactions based on neural networks[J]．IEEE/ACM Transactions on Computational Biology and Bioinformatics,2022,19(1):40-47. [14] RANI S ,GHAI D ,KUMAR S ．Object detection and recognition using contour based edge detection and fast R-CNN[J]．Multimedia Tools and Applications,2022,81(29):42183-42207. [15] KZLOLUK S ,SERT E ．Hurricane-Faster R-CNN-JS：hurricane detection with faster R-CNN using artificial Jellyfish Search （JS） optimizer[J]．Multimedia Tools and Applications,2022,81(26):37981-37999. [16] ZHANG R Y ,SONG Y ．Non-intrusive load identification method based on color encoding and improve R-FCN[J]．Sustainable Energy Technologies and Assessments,2022,53:102714. [17] HWANG Y J ,LEE J G ,MOON U C ,et al ．SSD-TSEFFM：new SSD using trident feature and squeeze and extraction feature fusion[J]．Sensors,2020,20(13):3630. [18] XU Z J ,SU J J ,HUANG K ．A-RetinaNet：a novel RetinaNet with an asymmetric attention fusion mechanism for dim and small drone detection in infrared images[J]．Mathematical Biosciences and Engineering：MBE,2023,20(4):6630-6651.
计量
- 文章访问数: 46
- HTML全文浏览量: 4
- PDF下载量: 21
