Please wait a minute...
浙江大学学报(工学版)
浙江大学学报(工学版)  2023, Vol. 57 Issue (5): 1009-1020    DOI: 10.3785/j.issn.1008-973X.2023.05.017
机械工程     
基于损伤区域融合变换的轴承鼓形滚子表面损伤检测方法
艾青林(),崔景瑞,吕冰海,童桐
浙江工业大学 特种装备制造与先进加工技术教育部/浙江省重点实验室，浙江 杭州 310023
Surface defect detection method for bearing drum-shaped rollers based on fusion transformation of defective area
Qing-lin AI(),Jing-rui CUI,Bing-hai LV,Tong TONG
Key Laboratory of Special Purpose Equipment and Advanced Manufacturing Technology, Ministry of Education and Zhejiang Province, Zhejiang University of Technology, Hangzhou 310023, China
 全文: PDF(2975 KB)   HTML
摘要：

针对鼓形滚子表面损伤检测过程中损伤零件数量少、检测准确率及查准率不高的问题，提出一种基于多样本关键损伤区域融合数据增强的轴承鼓形滚子表面损伤检测方法. 搭建鼓形滚子表面损伤特征样本检测装置，采集滚子的工作面侧面图像并将侧面图像作为研究对象，根据样本图像中滚子外圆角分布特点对样本核心检测区域进行裁剪. 基于聚类预选框及损伤评估函数，从数据集负样本中筛选合适大小与纵横比的关键损伤区域，对所选区域进行不同的仿射变换并与随机采样的正样本进行融合以生成新的增强样本数据. 将经过数据增强处理的图像，输入到检测神经网络中进行检测分类. 实验表明，采用数据增强的鼓形滚子表面损伤检测模型在测试集上的准确率为95.2%，查准率为97.7%，经过正样本阈值调整后模型最终的查准率为99.8%. 该方法融合不同样本的复杂损伤特征，有效地提升了检测准确率以及查准率，满足鼓形滚子表面损伤检测的实际生产需求.
关键词： 鼓形滚子;  表面损伤检测;  数据增强;  图像分类;  特征融合    
Abstract:

A surface defect detection method of bearing drum-shaped roller was proposed based on multi-sample defective area fusion data augmentation to aim at the problems of defective parts scarcity, low detection accuracy and low precision in process of drum-shaped roller surface defect detection. A drum-shaped roller surface defect sample collection system was built. The side images of the rollers were collected and used as research object. The core detection area was cropped according to distribution characteristics of the outer corner of roller in the sample image. Based on the clustering pre-selection box and the defect assessment function, the key defective area with appropriate size and aspect ratio was selected from the negative samples. The different affine transformations were performed on the selected area.The area was merged with randomly sampled positive samples to generate new data. The processed image was input into the inspection neural network for detection and classification. The experiments showed that accuracy rate and precision rate of the data-enhanced drum roller surface defect detection model were 95.2% and 97.7% respectively. After the threshold was adjusted, the final precision rate was 99.8%. The detection accuracy and precision improved effectively by integrating the complex defects characteristics of the different samples. The actual production requirements of drum roller surface defect detection can be met.
Key words: drum-shaped roller    surface defect detection    data augmentation    image classification    feature fusion
收稿日期: 2021-12-16 出版日期: 2023-05-09
CLC:  TP 399  
基金资助: 国家自然科学基金资助项目（52075488）；浙江省自然科学基金资助项目（LY20E050023）
作者简介: 艾青林（1976—），男，教授，博士. 从事机器视觉检测技术与智能机器人技术研究. orcid.org/0000-0002-9017-1916. E-mail： aqlaql@163.com
服务　　
把本文推荐给朋友
加入引用管理器
E-mail Alert
作者相关文章　　
艾青林
崔景瑞
吕冰海
童桐

引用本文:

艾青林,崔景瑞,吕冰海,童桐. 基于损伤区域融合变换的轴承鼓形滚子表面损伤检测方法[J]. 浙江大学学报(工学版), 2023, 57(5): 1009-1020.

Qing-lin AI,Jing-rui CUI,Bing-hai LV,Tong TONG. Surface defect detection method for bearing drum-shaped rollers based on fusion transformation of defective area. Journal of ZheJiang University (Engineering Science), 2023, 57(5): 1009-1020.

链接本文:

https://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2023.05.017    或    https://www.zjujournals.com/eng/CN/Y2023/V57/I5/1009

图 1  滚子面损伤图像样本采集装置
图 2  轴承滚子实物与样本对照图
图 3  滚子侧面样本不同损伤
图 4  原图核心检测区域与高斯滤波后图像
图 5  经二值化后图像与轮廓特征
图 6  上侧外圆角轮廓与添加黑边图像
图 7  损伤区域数据标注示例
图 8  预选框内侧临界情况与外侧临界情况
图 9  损伤拼贴对照图
图 10  经过损伤变换后生成的负样本
图 11  轴承鼓形滚子整体检测流程图
图 12  不同数据增强方法处理的结果图
$ \alpha $ ${P_{\rm{a}}}$ $ {P_{\text{c}}} $ $ {F_\delta } $
0.50 0.927 0.979 0.966
0.75 0.922 0.956 0.950
1.00 0.929 0.958 0.953
1.25 0.927 0.949 0.947
1.50 0.924 0.926 0.932
表 1  损失函数不同α值对模型指标的影响
组别 $ {P_a} $ $ {P_{\text{c}}} $
原图 0.762 0.926
J组 0.879 0.934
K组 0.905 0.945
表 2  原图与2种裁剪方法在测试集上结果对比
色彩抖动实验 清晰度实验
u v 训练集 $ {P_a} $ 测试集 $ {P_a} $ u v 训练集 $ {P_a} $ 测试集 $ {P_a} $
0 0 0.931 0.905 0 0 0.931 0.905
0.5 0.1 0.942 0.919 0.5 0.1 0.935 0.903
0.5 0.2 0.942 0.920 0.5 0.2 0.937 0.908
0.5 0.3 0.941 0.921 0.5 0.3 0.938 0.914
0.5 0.4 0.936 0.917 0.5 0.4 0.938 0.914
0.5 0.5 0.931 0.914 0.5 0.5 0.938 0.914
0.5 0.6 0.931 0.913 0.5 0.6 0.938 0.916
0.5 0.7 0.931 0.911 0.5 0.7 0.937 0.919
0.5 0.8 0.926 0.910 0.5 0.8 0.937 0.920
0.5 0.9 0.921 0.910 0.5 0.9 0.937 0.921
0.5 1.0 0.916 0.909 0.5 1.0 0.936 0.905
0.1 0.3 0.937 0.919 0.1 0.9 0.929 0.905
0.2 0.3 0.936 0.911 0.2 0.9 0.938 0.914
0.3 0.3 0.935 0.903 0.3 0.9 0.938 0.919
0.4 0.3 0.938 0.912 0.4 0.9 0.939 0.924
0.6 0.3 0.936 0.915 0.6 0.9 0.943 0.922
0.7 0.3 0.930 0.908 0.7 0.9 0.943 0.929
0.8 0.3 0.931 0.912 0.8 0.9 0.943 0.936
0.9 0.3 0.932 0.916 0.9 0.9 0.941 0.930
表 3  滚子样本色彩抖动和清晰度实验
u 训练集 测试集
0.1 0.919 0.915
0.2 0.925 0.924
0.3 0.928 0.926
0.4 0.941 0.908
0.5 0.934 0.913
0.6 0.947 0.904
0.7 0.961 0.905
0.8 0.950 0.901
0.9 0.955 0.891
表 4  滚子样本损伤变换实验
图 13  不同数据增强方法下的准确率与损失变化
数据增强方案 ${P_{\rm{a}}}$ $ {P_{\text{c}}} $ R $ {F_\delta } $
原图 0.762 0.926 0.736 0.881
目标裁剪 0.871 0.937 0.886 0.926
目标裁剪+旋转+填充（基准） 0.896 0.940 0.891 0.930
基准+色彩抖动 0.919 0.934 0.935 0.934
基准+清晰度 0.931 0.974 0.924 0.964
基准+损伤变换 0.926 0.960 0.928 0.953
基准+色彩抖动+清晰度+损伤变换 0.952 0.977 0.930 0.967
表 5  不同数据增强方法名性能测试结果
t $ {P_a} $ $ {P_{\text{c}}} $ t ${P_a} $ ${P_{\text{c}}} $
0.5 0.952 0.977 0.8 0.934 0.992
0.6 0.947 0.981 0.9 0.888 0.998
0.7 0.940 0.985 ? ? ?
表 6  不同样本阈值对准确率与查准率的影响
1 何斌, 戚佳杰, 黎明和 小波分析在滚动轴承故障诊断中的应用研究[J]. 浙江大学学报:工学版, 2009, 43 (7): 1218- 1221
HE Bin, QI Jia-jie, LI Ming-he Application and research of wavelet analysis in fault diagnosis of roller bearings[J]. Journal of Zhejiang University: Engineering Science, 2009, 43 (7): 1218- 1221
2 梁子千, 玄文博, 王婷, 等 基于二维阻抗特征的管道环焊缝缺陷涡流检测[J]. 仪器仪表学报, 2017, 38 (9): 2138- 2145
LIANG Zi-qian, XUAN Wen-bo, WANG Ting, et al Eddy current NDt for the cracks of girth welds of pipes based on 2D impedance characteristics[J]. Chinese Journal of Scientific Instrument, 2017, 38 (9): 2138- 2145
doi: 10.3969/j.issn.0254-3087.2017.09.006
3 HANSEN J The eddy current inspection method[J]. Insight, 2004, 46 (5): 279- 281
doi: 10.1784/insi.46.5.279.55563
4 刘旭, 夏金东, 弓乐, 等 射频信号在超声检测缺陷识别中的应用研究[J]. 机械工程学报, 2002, 38 (4): 84- 87
LIU Xu, XIA Jin-dong, GONG Yue, et al Study on the application of radio-frequency signal in flaw classification of ultrasonic testing[J]. Chinese Journal of Mechanical Engineering, 2002, 38 (4): 84- 87
doi: 10.3321/j.issn:0577-6686.2002.04.018
5 杨辰龙, 周晓军 复杂曲面工件的超声无损检测系统研制[J]. 中国机械工程, 2005, 16 (18): 1622- 1625
YANG Chen-long, ZHOU Xiao-jun Development of ultrasonic none-destructive test systems for complex curved surface parts[J]. China Mechanical Engineering, 2005, 16 (18): 1622- 1625
doi: 10.3321/j.issn:1004-132X.2005.18.007
6 HAYNES C, TODD M, NADABE T, et al Monitoring of bearing failure in composite bolted connections using ultrasonic guided waves: a parametric study[J]. Structural Health Monitoring, 2014, 13 (1): 94- 105
doi: 10.1177/1475921713510756
7 王姮, 卜燕, 张华, 等 基于分形维数的磁痕图像缺陷检测[J]. 计算机应用研究, 2015, 32 (2): 603- 605
WANG Heng, BU Yan, ZHANG Hua, et al Defect detection of magnetic particle indication image based on fractal dimension[J]. Application Research of Computers, 2015, 32 (2): 603- 605
doi: 10.3969/j.issn.1001-3695.2015.02.063
8 陈昊, 张奔, 黎明, 等 基于图像光流的轴承滚子表面缺陷检测[J]. 仪器仪表学报, 2018, 39 (6): 198- 206
CHEN Hao, ZHANG Ben, LI Ming, et al Surface defect detection of bearing roller based on image optical flow[J]. Chinese Journal of Scientific Instrument, 2018, 39 (6): 198- 206
doi: 10.19650/j.cnki.cjsi.J1803226
9 CHAN F W Y Reflective fringe pattern technique for subsurface crack defection[J]. NDT and E International, 2008, 41 (8): 602- 610
doi: 10.1016/j.ndteint.2008.06.003
10 段志达. 基于机器视觉的轴承缺陷检测[D]. 安徽: 安徽工程大学, 2020: 23.
DUAN Zhi-da. Bearing defect detection based on machine vision [D]. Anhui: Anhui Polytechnic University, 2020: 23.
11 尚鲁强. 基于机器视觉的复杂曲面工件表面缺陷检测技术与系统[D]. 江西: 华东交通大学, 2020: 31.
SHAGN Lu-qiang. Technology and system for detecting surface defects of complex curved workpieces based on machine vision [D]. Jiangxi: East China Jiaotong University, 2020: 31.
12 苏俊宏, 刘胜利 圆柱型高精密零件表面缺陷检测及形貌分析[J]. 激光与光电子学进展, 2014, 51 (4): 150- 154
SU Jun-hong, LIU Sheng-li Measure system of surface flaw and morphology analysis of cylindrical high precision parts[J]. Laser and Optoelectronics Progress, 2014, 51 (4): 150- 154
13 文生平, 周正军, 张啸言, 等 基于计算机视觉的轴承滚子表面缺陷在线检测系统[J]. 华南理工大学学报:自然科学版, 2020, 48 (10): 76- 87
WEN Sheng-ping, ZHOU Zheng-jun, ZHANG Xiao-yan, et al On-line detection system for surface defects of bearing rollers based on computer vision[J]. Journal of South China University of Technology: Natural Science Edition, 2020, 48 (10): 76- 87
14 PRAPPACHER N, BULLMANN M, BOHN G, et al Defect detection on rolling element surface scans using neural image segmentation[J]. Applied Sciences, 2020, 10 (9): 3290
doi: 10.3390/app10093290
15 XU X , ZHENG H, GUO Z, et al. SDD-CNN: small data-driven convolution neural networks for subtle roller defect inspection[J]. Applied Sciences, 2019, 9 (7): 1364
doi: 10.3390/app9071364
16 LE X, MEI J, ZHANG H, et al A learning-based approach for surface defect detection using small image datasets[J]. Neurocomputing, 2020, (408): 112- 120
17 HU J, YAN P, SU Y, et al A method for classification of surface defect on metal workpieces based on twin attention mechanism generative adversarial network[J]. IEEE Sensors, 2021, 21 (12): 13430- 13441
doi: 10.1109/JSEN.2021.3066603
18 FANG H, SUN J, WANG R, et al. InstaBoost: boosting instance segmentation via probability map guided copy-pasting [C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Korea: Seoul, 2019: 682-691.
19 VINCENT O R, FOLORUNSO O A. Descriptive algorithm for sobel image edge detection [C]// Proceedings of Informing Science and IT Education Conference (InSITE). USA: Macon, 2009, 40: 97-107.
20 BASU M Gaussian-based edge-detection methods-a survey[J]. IEEE Transactions on Systems, Man, and Cybernetics, Part C, 2002, 32 (3): 252- 260
doi: 10.1109/TSMCC.2002.804448
21 HARALICK R M, STERNBERG S R, ZHUANG X Image analysis using mathematical morphology[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1987, (4): 532- 550
22 SUZUKI S Topological structural analysis of digitized binary images by border following[J]. Computer Vision, Graphics and Image Processing, 1985, 30 (1): 32- 46
doi: 10.1016/0734-189X(85)90016-7
23 ZHENG Z , WANG P , LIU W , et al. Distance-IoU loss: faster and better learning for bounding box regression [C]// AAAI Conference on Artificial Intelligence. USA: New York, 2020, 34(7): 12993-13000.
24 XIE S, GIRSHICK R, DOLLAR P, et al. Aggregated residual transformations for deep neural networks [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. USA: Honolulu, 2017: 1492-1500.
[1] 张剑钊,郭继昌,汪昱东. 基于融合逆透射率图的水下图像增强算法[J]. 浙江大学学报(工学版), 2023, 57(5): 921-929.
[2] 杨帆,宁博,李怀清,周新,李冠宇. 基于语义增强特征融合的多模态图像检索模型[J]. 浙江大学学报(工学版), 2023, 57(2): 252-258.
[3] 周国华,卢剑伟,倪彤光,胡学龙. 层次型非线性子空间字典学习[J]. 浙江大学学报(工学版), 2022, 56(6): 1159-1167.
[4] 张雪芹,李天任. 基于Cycle-GAN和改进DPN网络的乳腺癌病理图像分类[J]. 浙江大学学报(工学版), 2022, 56(4): 727-735.
[5] 张娜,戚旭磊,包晓安,吴彪,涂小妹,金瑜婷. 基于优化预测定位的单阶段目标检测算法[J]. 浙江大学学报(工学版), 2022, 56(4): 783-794.
[6] 黄新宇,游帆,张沛,张昭,张柏礼,吕建华,徐立臻. 基于多分类及特征融合的静默活体检测算法[J]. 浙江大学学报(工学版), 2022, 56(2): 263-270.
[7] 于楠晶,范晓飚,邓天民,冒国韬. 基于多头自注意力的复杂背景船舶检测算法[J]. 浙江大学学报(工学版), 2022, 56(12): 2392-2402.
[8] 张云佐,郭威,蔡昭权,李文博. 联合多尺度与注意力机制的遥感图像目标检测[J]. 浙江大学学报(工学版), 2022, 56(11): 2215-2223.
[9] 李明,段立娟,王文健,恩擎. 基于显著稀疏强关联的脑功能连接分类方法[J]. 浙江大学学报(工学版), 2022, 56(11): 2232-2240.
[10] 徐泽鑫,段立娟,王文健,恩擎. 基于上下文特征融合的代码漏洞检测方法[J]. 浙江大学学报(工学版), 2022, 56(11): 2260-2270.
[11] 杨栋杰,高贤君,冉树浩,张广斌,王萍,杨元维. 基于多重多尺度融合注意力网络的建筑物提取[J]. 浙江大学学报(工学版), 2022, 56(10): 1924-1934.
[12] 陈雪云,黄小巧,谢丽. 基于多尺度条件生成对抗网络血细胞图像分类检测方法[J]. 浙江大学学报(工学版), 2021, 55(9): 1772-1781.
[13] 陈智超,焦海宁,杨杰,曾华福. 基于改进MobileNet v2的垃圾图像分类算法[J]. 浙江大学学报(工学版), 2021, 55(8): 1490-1499.
[14] 陈岳林,田文靖,蔡晓东,郑淑婷. 基于密集连接网络和多维特征融合的文本匹配模型[J]. 浙江大学学报(工学版), 2021, 55(12): 2352-2358.
[15] 陈雪云,夏瑾,杜珂. 基于多线型特征增强网络的架空输电线检测[J]. 浙江大学学报(工学版), 2021, 55(12): 2382-2389.