Laser & Optoelectronics Progress, Volume. 61, Issue 23, 2312002(2024)

Near-Surface Defect Detection Using Convolutional Neural Network and Laser Ultrasound Testing

Mingze Guo, Xingyuan Zhang*, and Zhenyue Jin
Author Affiliations
  • School of Air Transport, Shanghai University of Engineering Science, Shanghai 201620, China
  • show less
    Figures & Tables(10)
    Time spectrum diagram based on db4 wavelet transform. (a) Defect depth 0.5 mm, angle 20°; (b) defect depth 1.5 mm, angle 40°; (c) defect depth 3.5 mm, angle 140°; (d) defect depth 10 mm, angle 100°
    Network structure diagram
    Meshing of the finite element simulation model
    Time-domain signal and locally enlarged image of reflected wave with different depth defects. (a) Time-domain signal; (b) locally enlarged image
    Time-domain signal and locally enlarged image of reflected wave with different angle defects. (a) Time-domain signal; (b) locally enlarged image
    Schematic diagrams of experimental platform. (a) Principle diagram of laser non-destructive testing; (b) physical image of laser ultrasonic testing system; (c) top view of metal sample; (d) detailed diagram of laser parameter settings
    Laser ultrasonic signals obtained on specimens with different defect widths, depths and angles and comparison of experimental and simulation data. (a)‒(c) Laser ultrasonic signals; (d) comparison of experimental and simulation data
    Defect depth and angle prediction of validation set and test set. (a) VGG19 network loss function change diagram; (b) ResNet101 network loss function change diagram; (c) DenseNet169 network loss function change diagram; (d) VGG19 model validation set prediction result diagram; (e) ResNet101 model validation set prediction result diagram; (f) DenseNet169 model validation set prediction result diagram; (g) VGG19 model test set prediction result diagram; (h) ResNet101 model test set prediction result diagram; (i) DenseNet169 model test set prediction result diagram; (j) prediction results of VRD+SVR model on simulation data; (k) prediction results of VRD+SVR model on experimental + simulation data
    Prediction results of different methods
    • Table 1. Table of thermal properties and elastic parameters of two-dimensional aluminum alloy simulation model

      View table

      Table 1. Table of thermal properties and elastic parameters of two-dimensional aluminum alloy simulation model

      ParameterUnitValue
      Poisson’s ratioν0.33
      Young’s modulusE /GPa70
      Intensityρ /(kgm-3)2700
      Coefficient of thermal expansionα /μK23
      Thermal conductivityk /(Wm-1K-1)238
      Constant pressure heat capacityC /(JKkg-1)900
    Tools

    Get Citation

    Copy Citation Text

    Mingze Guo, Xingyuan Zhang, Zhenyue Jin. Near-Surface Defect Detection Using Convolutional Neural Network and Laser Ultrasound Testing[J]. Laser & Optoelectronics Progress, 2024, 61(23): 2312002

    Download Citation

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

    Category: Instrumentation, Measurement and Metrology

    Received: Jan. 9, 2024

    Accepted: Mar. 29, 2024

    Published Online: Nov. 19, 2024

    The Author Email: Xingyuan Zhang (zxy_sues@163.com)

    DOI:10.3788/LOP240477

    CSTR:32186.14.LOP240477

    Topics