Deep Learning for Quality Prediction in Dissimilar Spot Welding DP600-AISI304, Using a Convolutional Neural Network and Infrared Image Processing

  • Bárbara Dora Ross Veitía ,
  • Alejandro Espinel Hernández,
  • Louriel Oliveira Villarinho,
  • Mario Sánchez Orozco,
  • Angel Sánchez Roca,
  • Hipólito Carvajal Fals,
  • Emilio Jimenez Macias 
  • a,b National Center for Applied Electromagnetism, University of Oriente, Santiago de Cuba, Cuba. Ave Las América s/n. 90400
  • Faculty of Mechanical Engineering. Federal University of Uberlândia, Minas Gerais, Uberlândia, Brazil
  • d,e Faculty of Mechanical and Industrial Engineering. University of Oriente, Santiago de Cuba, Cuba. Ave Las América s/n. 90400. Santiago de Cuba. Cuba
  • Federal Technological University of Paraná. Rua Doutor Washington Subtil Chueire, 330 - Jardim Carvalho. CEP 84017-20 - Ponta Grossa. PR. Paraná. Brazil
  • Department of Electrical Engineering, University of La Rioja, Av. de la Paz, 93, 26006 Logroño, Spain
Cite as
Ross Veitía B.D., Espinel Hernández A., Oliveira Villarinho L., Sánchez Orozco M., Sánchez Roca A., Carvajal Fals H., Jimenez Macias E. (2020). Deep Learning for Quality Prediction in Dissimilar Spot Welding DP600-AISI304, Using a Convolutional Neural Network and Infrared Image Processing. Proceedings of the 32nd European Modeling & Simulation Symposium (EMSS 2020), pp. 393-399. DOI: https://doi.org/10.46354/i3m.2020.emss.057

Abstract

In face of the dizzying progress of Industry 4.0, the application of Artificial Intelligence in manufacturing processes is a challenging task. In this research, a Convolutional Neural Network (CNN) was implemented to determine the quality of dissimilar joints DP 600-AISI 304. Infrared images obtained during the process of resistance spot welding were processed. The idea of applying a CNN focuses on filtering infrared images before training the deep neural network, for the detection of certain hidden features in the data, as well as extracting patterns and classifying welded joints. For its implementation, open source tools such as Anaconda, libraries such as Tensorflow and Keras, high-level Application Programming Interface (APIs) were used to work with neural networks in Python language. After processing and training with infrared images, a neural model was obtained and the metrics obtained from the training were analysed. It was found that the use of deep learning and in particular CNN are techniques that can be considered as predictive methods for the classification of welded joints and computer vision supervision. Data processing was possible on very small timescales, facilitating optimization and efficiency improvement in manufacturing processes.

References

  1. Espinel Hernández, A., Sánchez Roca, A., Carvajal Fals, H., Antonio Ferraresi, V., & Oliveira Vilarinho, L. (2016). Influence of polarity on mechanical properties of dissimilar resistance spot welds of DP 600/AISI 304 steels. Science and Technology of Welding and Joining, 21(8), 607-613.
  2. Garrido, I., Lagüela, S., Otero, R., & Arias, P. (2020). Thermographic methodologies used in
    infrastructure inspection: A review —Post-processing procedures. Applied Energy, 266, 114857.
  3. Gavidel, S. Z., Lu, S., & Rickli, J. L. (2019). Performance analysis and comparison of machine learning algorithms for predicting nugget width of resistance spot welding joints. The  International Journal of Advanced Manufacturing Technology, 105(9), 3779-3796.
  4. Guo, Z., Ye, S., Wang, Y., & Lin, C. (2017, July). Resistance welding spot defect detection with
    convolutional neural networks. In International Conference on Computer Vision Systems (pp. 169-174). Springer, Cham.
  5. Gutiérrez, C.A., Deep Learning applied to Welding Diagnostics via Plasma Optical Spectroscopy. 2017.
  6. He, Z., & Liu, Q. (2020). Deep Regression Neural Network for Industrial Surface Defect Detection. IEEE Access, 8, 35583-35591.
  7. ISO, B. (2007). 14373: 2007." Resistance Welding: Procedure for Spot Welding of Uncoated and Coated Low Carbon Steels. British Standards.
  8. Martín, Ó., De Tiedra, P., & López, M. (2010). Artificial neural networks for pitting potential prediction of resistance spot welding joints of AISI 304 austenitic stainless steel. Corrosion Science, 52(7), 2397-2402.
  9. Mayr, A., Lutz, B., Weigelt, M., Gläßel, T., Kißkalt, D., Masuch, M., ... & Franke, J. (2018, December). Evaluation of machine learning for quality monitoring of laser welding using the example of the contacting of Hairpin windings. In 2018 8th International Electric Drives Production Conference (EDPC) (pp. 1-7). IEEE.
  10. Zuniga, S., & Sheppard, S. D. (1997). Resistance spot weld failure loads and modes in overload
    conditions. In Fatigue and Fracture Mechanics: 27th Volume. ASTM International, vol. 1296: pp.
    469-489.