Automatic classification of pores in aluminum castings using machine learning

  1. Chaves, Deisy 1
  2. Fidalgo, Eduardo 1
  3. Rodríguez-González, Pablo 1
  4. Fernández-Abia, A.I. 1
  5. Alegre, Enrique 1
  6. Barreiro, Joaquín 1
  1. 1 Universidad de León

    Universidad de León

    León, España


XLIV Jornadas de Automática: libro de actas: Universidad de Zaragoza, Escuela de Ingeniería y Arquitectura, 6, 7 y 8 de septiembre de 2023, Zaragoza
  1. Ramón Costa Castelló (coord.)
  2. Manuel Gil Ortega (coord.)
  3. Óscar Reinoso García (coord.)
  4. Luis Enrique Montano Gella (coord.)
  5. Carlos Vilas Fernández (coord.)
  6. Elisabet Estévez Estévez (coord.)
  7. Eduardo Rocón de Lima (coord.)
  8. David Muñoz de la Peña Sequedo (coord.)
  9. José Manuel Andújar Márquez (coord.)
  10. Luis Payá Castelló (coord.)
  11. Alejandro Mosteo Chagoyen (coord.)
  12. Raúl Marín Prades (coord.)
  13. Vanesa Loureiro-Vázquez (coord.)
  14. Pedro Jesús Cabrera Santana (coord.)

Publisher: Servizo de Publicacións ; Universidade da Coruña

ISBN: 9788497498609

Year of publication: 2023

Pages: 849-854

Congress: Jornadas de Automática (44. 2023. Zaragoza)

Type: Conference paper


Porosity inspection of manufactured parts has traditionally been performed using microscopy manipulated by a human technician. However, the person involved needs experience in this task, and the number of parts that can be inspected per unit of time is limited. The presence of porosity in the material is critical, as it can negatively affect the mechanical properties and the quality of the part. In this paper, we propose to automate the classification of the porosity defects that appear inside the parts manufactured by casting. First, we acquire images from aluminum parts manufactured by two casting methods: a traditional one using sand molding and a more modern one with the Binder Jetting (BJ) additive manufacturing technique. Then, we crop regions with and without pores we later describe using SIFT descriptors encoded into BoVW features to feed and train two SVM classifiers: one for predicting if the image contains a pore or not, and the other for also indicating if the pore detected is due to the effect of gases or by shrinkage during solidification.