Infrared and Laser Engineering, Volume. 53, Issue 7, 20240176(2024)

Detection method for structural defects of railway clip fastener based on 3D line laser sensor

Xiaocui YUAN1,2, Yongtao WANG1, Baoling LIU1, Dibo HOU2, and Zonghui JIANG1
Author Affiliations
  • 1School of Electrical Engineering, Nanchang Institute of Technology, Nanchang 330099, China
  • 2College of Control Science and Engineering, Zhejiang University, Hangzhou 310027, China
  • show less
    Figures & Tables(27)
    (a) Fastener structure of WJ-8; (b) Structure and gap of clip
    (a) Diagram of railway fastener data acquisition system; (b) Physical image of railway fastener data acquisition system
    (a) Top view of loose fasteners; (b) Front view of loose fasteners; (c) Top view of fasteners that are too tight; (d) Front view of over tightened fasteners
    Point cloud of railway and fastener areas
    PointNet++ network architecture
    (a) Fastener point cloud; (b) Labeled point cloud
    (a) Segmentation results of fastener; (b) View of clip point cloud
    (a) Point cloud of clip; (b) Clip image
    (a) The dilated result of clip image; (b) 2D skeleton of clip
    (a) The center of the clip section; (b) The center of clip section after removing the error points
    (a) Skeleton division diagram; (b) Clip gap
    (a)Schematic of standard step block; (b) Schematic of step block imaging; (c) Point cloud of step block; (d) Measurement error of block
    PointNet++ network segmentation accuracy metric graph
    (a) 0.5 mm gap gauge; (b) Clip gap equaling to 9 mm
    Measurement results of clip gaps under different buckle pressures
    (a) measured clip gaps with different dx; (b) 3D skeleton and gap of clip dy
    (a) dy=0.56 mm; (b) dy=0.84 mm; (c) dy=1.12 mm; (d) dy=1.68 mm; (e) dy=2.24 mm; (f) dy=2.8 mm
    (a) Rusted fastener; (b) Stained fastener; (c) Partial rusted fastener; (d) Clip gap of rusted fastener; (e) Clip gap of stained fastener; (f) Clip gap of partial rusted fastener
    (a) Fasteners measured at night; (b) Clip point cloud; (c) Skeleton and gap of clip; (d) Fasteners measured at noon; (e) Clip point cloud; (f) Skeleton and gap of clip; (g) Measuring fasteners at dusk; (h) Clip point cloud; (i) Skeleton and gap of clip
    (a) Over-tightening fastener scatter plot of clip gap value; (b) Histogram distribution of clip gap
    (a) Normal fastener scatter plot of clip gap value; (b) Histogram distribution of clip gap
    (a) Looseness fastener scatter plot of clip gap value; (b) Histogram distribution of clip gap
    (a) Field measurement of WJ-8 fastener is 14.2 mm; (b) Field measurement of WJ-8 fastener is 15.2 mm; (c) Field measurement of WJ-2 fastener is 13.32 mm; (d) Field measurement of WJ-2 fastener is 15.67 mm
    • Table 1. Hyper-parameters for PointNet++ network training

      View table
      View in Article

      Table 1. Hyper-parameters for PointNet++ network training

      ItemParameterValue
      Network trainingSample number84
      Batch size16
      Epoch251
      Learning rate0.001
      Step_size20
      Lr_decay0.5
      Number point(npoint)2 048
      Adam algorithmMomentum0.9
      Betas0.999
      Eps1e-08
    • Table 2. Measurement results and comparison for different fasteners

      View table
      View in Article

      Table 2. Measurement results and comparison for different fasteners

      FastenerManual measurementThis paperError
      Fastener 1(WJ-2)33.12720.1272
      Fastener 2(WJ-2)33.14350.1435
      Fastener 3(WJ-8)32.9501−0.0499
      Fastener 4(WJ-7)6.96.9870.087
      Fastener 5(WJ-8)9.69.526−0.074
      Fastener 6(WJ-8)00.10.1
    • Table 3. Accuracy of fastener detection under different allowable errors

      View table
      View in Article

      Table 3. Accuracy of fastener detection under different allowable errors

      $ \mathbf{\tau }$/mmClasses of fastenerTotalTPFNDR
      $ \tau =0 $Over-tight2001703085.0%
      $ \tau =0.1 $196498.0%
      $ \tau =0 $Normal100076223876.2%
      $ \tau =\pm 0.1 $9901099.0%
      $ \tau =0 $Looseness2302121892.2%
      $ \tau =-0.1 $2201095.7%
    • Table 4. Precision and recall of three method

      View table
      View in Article

      Table 4. Precision and recall of three method

      DefectAccuracyMethod
      Reference[15]Reference[18]This paper
      Over-tightenedPrecision61.4%85.8%90.8%
      Recall80.7%78.5%80.5%
      LoosenessPrecision67.2%70.5%95.2%
      Recall92.2%93.5%90.5%
    Tools

    Get Citation

    Copy Citation Text

    Xiaocui YUAN, Yongtao WANG, Baoling LIU, Dibo HOU, Zonghui JIANG. Detection method for structural defects of railway clip fastener based on 3D line laser sensor[J]. Infrared and Laser Engineering, 2024, 53(7): 20240176

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category:

    Received: Apr. 22, 2024

    Accepted: --

    Published Online: Aug. 9, 2024

    The Author Email:

    DOI:10.3788/IRLA20240176

    Topics