Séminaire LGF – Victor Gonzalez-Castro – 27 novembre 2025

Titre

Inspection des surfaces et porosité des matériaux métalliques avec vision artificielle et apprentissage automatique

Résumé

In this talk, I will present two recent works carried out at the University of León on applying computer vision and machine learning to surface inspection in metal additive manufacturing. The first one focuses on non-contact estimation of arithmetical mean roughness (Ra) in stainless-steel parts. We developed an image-based methodology that combines classical texture descriptors (Haralick, LBP and Dense SIFT) with regression models to predict surface roughness from simple photographs.

The second work addresses defect detection in aluminium alloy castings produced using 3D-printed molds via binder jetting. We introduce the Porosity Detection Corpus, a dataset of metallographic images annotated with shrinkage and gas pores, and evaluate a YOLO-based object detection system to automatically locate and classify these defects.