Journals Articles Conferences News About CLP
Submit a paper Email Alert
Search
Search
Articles
Journals
News
Advanced Search
Top Searches
2D MaterialsTransformation opticsQuantum PhotonicsSmall lasersSemiconductor UV PhotonicsSupersymmetric microring laser arrays

Chinese Journal of Lasers, Volume. 48, Issue 6, 0602119(2021)

Laser Welding Penetration State Recognition Method Fused with Timing Information

Tianyuan Liu1, Jinsong Bao1、*, Junliang Wang1, and Jun Gu2
Author Affiliations
  • 1Institute of Intelligent Manufacturing, College of Mechanical Engineering, Donghua University, Shanghai 201600, China
  • 2Shanghai Institute of Laser Technology, Shanghai 200235, China
    • show less
    AbstractGet PDF(in Chinese)
    Figures&Tables (11)
    Equations (0)
    References (23)
    Cited By (6)
    Tools
    Paper Information
    Topics
    Figures & Tables(11)
    Overall framework

    Fig. 1. Overall framework

    Download full size
    Structure of BiLSTM network

    Fig. 2. Structure of BiLSTM network

    Download full size
    Test system

    Fig. 3. Test system

    Download full size
    Images of backside of weld joint and molten pool/keyhole corresponding to different penetration states

    Fig. 4. Images of backside of weld joint and molten pool/keyhole corresponding to different penetration states

    Download full size
    Comparison of training processes of different methods

    Fig. 5. Comparison of training processes of different methods

    Download full size
    Comparison of CNN-BiLSTM method on training set and validation set

    Fig. 6. Comparison of CNN-BiLSTM method on training set and validation set

    Download full size
    Confusion matrices of classification results of different methods. (a) CNN;(b) CNN-LSTM; (c) CNN-BiLSTM

    Fig. 7. Confusion matrices of classification results of different methods. (a) CNN;(b) CNN-LSTM; (c) CNN-BiLSTM

    Download full size
    Training processes corresponding to different hyperparameters. (a) Different learning rates; (b) different optimizers

    Fig. 8. Training processes corresponding to different hyperparameters. (a) Different learning rates; (b) different optimizers

    Download full size

    • Table 1. Distribution and division of datasets

      View table

      Table 1. Distribution and division of datasets

      DatasetLPOPEPTotal
      Train12817128071297638600
      Valid36533641373611030
      Test1802187218485522
      Total18272183201856055152

    • Table 2. Test results of different methods on test set

      View table

      Table 2. Test results of different methods on test set

      ParameterCNNCNN-LSTMCNN-BiLSTM
      AACC /%92.9996.9999.26
      PPRE /%92.9997.0099.25
      RREC /%93.0397.0199.26
      FF1-score /%93.0197.0199.26
      Model size /MB4.3016.6633.11
      Training time /min3.157.938.68
      Test time /ms1.769.419.43

    • Table 3. Test results for different hyperparameters

      View table

      Table 3. Test results for different hyperparameters

      ParameterLROM
      10-510-410-2AdadeltaAdagradAdam
      AACC /%93.1499.1199.3597.6699.1899.31
      PPRE /%93.7099.1199.3597.6799.1999.31
      RREC /%93.1499.1199.3597.6799.1899.31
      FF1-score /%93.1499.1199.3597.6699.1999.31
    Tools

    Get Citation

    Copy Citation Text

    Tianyuan Liu, Jinsong Bao, Junliang Wang, Jun Gu. Laser Welding Penetration State Recognition Method Fused with Timing Information[J]. Chinese Journal of Lasers, 2021, 48(6): 0602119

    Download Citation

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

    Category: Laser Material Processing

    Received: Oct. 9, 2020

    Accepted: Nov. 12, 2020

    Published Online: Mar. 15, 2021

    The Author Email: Bao Jinsong (bao@dhu.edu.cn)

    DOI:10.3788/CJL202148.0602119

    Topics

    laser devices and laser physics
    Lasers and Laser Optics
    Laser physics
    laser manufacturing
    Instrumentation, Measurement and Metrology