• Acta Photonica Sinica
  • Vol. 49, Issue 10, 1015002 (2020)
He SUN1、2, Wen-zhen ZHAO1, Wen-hui ZHAO1, and Zhen-yun DUAN1
Author Affiliations
  • 1School of Mechanical Engineering,Shenyang University of Technology,Shenyang 110870,China
  • 2School of Electrical and Information Engineering,Liaoning Institute of Science and Technology,Benxi,Liaoning 117004,China
  • show less
    DOI: 10.3788/gzxb20204910.1015002 Cite this Article
    He SUN, Wen-zhen ZHAO, Wen-hui ZHAO, Zhen-yun DUAN. Classification of Edge Distortion of Tooth Profile Image Based on Improved Twin Support Vector Machine[J]. Acta Photonica Sinica, 2020, 49(10): 1015002 Copy Citation Text show less
    Geometric relationship of points on the involute
    Fig. 1. Geometric relationship of points on the involute
    Segmentation of the tooth profile measurement area
    Fig. 2. Segmentation of the tooth profile measurement area
    Discrimination of edge distortion area & type
    Fig. 3. Discrimination of edge distortion area & type
    Schematic diagram of edge distortion type discrimination
    Fig. 4. Schematic diagram of edge distortion type discrimination
    Structure of gear vision measuring device
    Fig. 5. Structure of gear vision measuring device
    The working process of gear vision image acquisition
    Fig. 6. The working process of gear vision image acquisition
    Eigenvector analysis
    Fig. 7. Eigenvector analysis
    Test results on OCF-PBT-TWSVM algorithm
    Fig. 8. Test results on OCF-PBT-TWSVM algorithm
    Test results on PBT-SVM algorithm
    Fig. 9. Test results on PBT-SVM algorithm
    Normalization methodTest accuracyOptimal parameter selection of OCF-PBT-TWSVM
    No normalization78% (39/50)

    Population size N=20;Maximum number of iterations K=200;

    Best c1=46.52; c2=48.97; g=0.084

    [-1,1]Normalization96% (48/50)

    Population size N=20;Maximum number of iterations K=200;

    Best c1=6.544; c2=6.998; g=4.634

    [0,1]Normalization96% (48/50)

    Population size N=20;Maximum number of iterations K=200;

    Best c1=6.875; c2=6.529; g=16.004

    Table 1. Comparison of different normalization methods
    Choice of kernel functionTest accuracyOptimal parameter selection of OCF-PBT-TWSVM
    Linear54% (27/50)

    Population size N=20;Maximum number of iterations K=200;

    Best c1=9.93; c2=10.46; g=11.795

    Polynomial92% (46/50)

    Population size N=20;Maximum number of iterations K=200;

    Best c1=7.59; c2=8.92; g=14.234

    Radial basis function96% (48/50)

    Population size N=20;Maximum number of iterations K=200;

    Best c1=6.89; c2=7.42; g=13.931

    Sigmoid42% (21/50)

    Population size N=20;Maximum number of iterations K=200;

    Best c1=6.58; c2=7.39; g=17.242

    Table 2. Comparison of different kernel functions
    AlgorithmTest accuracyOptimal parameter selectionAverage test accuracy
    OCF-PBT-TWSVM97.87% (46/47)N=20; K=200; Best c1=15.38; c2=16.92; g=4.1896.96%
    97.22% (35/36)N=20; K=200; Best c1=20.85; c2=22.02; g=20.69
    96% (48/50)N=20; K=200; Best c1=6.89; c2=7.42; g=13.931
    PBT-SVM95.75% (45/47)Best c=5.66; g=494.06%
    94.44% (34/36)Best c=11.6; g=8
    92% (46/50)Best c=8; g=16
    Table 3. Comparison of test results of different algorithms
    Normal edge signal

    Ignore type of distorted

    signal

    Eliminate type of distorted signalCompensation type of distorted signal
    1606410
    1587312
    162648
    1568412
    164448
    1539513
    Table 4. Real-time statistics of visual measurement results of tooth profile edges
    He SUN, Wen-zhen ZHAO, Wen-hui ZHAO, Zhen-yun DUAN. Classification of Edge Distortion of Tooth Profile Image Based on Improved Twin Support Vector Machine[J]. Acta Photonica Sinica, 2020, 49(10): 1015002
    Download Citation