The student Eneko Intxausti Arbaiza obtained an EXCELLENT CUM LAUDE grade with mention International Doctorate
The student Eneko Intxausti Arbaiza obtained an EXCELLENT CUM LAUDE grade with mention International Doctorate
The student Eneko Intxausti Arbaiza obtained an EXCELLENT CUM LAUDE grade with mention International Doctorate
- Thesis title: An Uncertainty-Aware Methodology for Reliable and Adaptive X-ray Defect Detection in Casting Manufacturing
Court:
- Presidency: Rubén Usamentiaga Fernández (Universidad de Oviedo)
- Vocal: Danijel Skocaj (University of Ljubljana)
- Secretary: Ane Alberdi Aramendi (Mondragon Unibertsitatea)
Abstract:
Quality control in the manufacturing of casting parts faces significant challenges due to the increasing complexity of industrial processes. Although deep learning techniques have shown success in computer vision, their application to industrial defect detection presents specific challenges, especially in detecting subtle defects in X-ray images and the scarcity of labeled data. This thesis develops a novel methodology for optical quality control in casting through the integration of advanced computer vision techniques with probabilistic modeling. The approach combines an enhanced detection architecture with contrastive learning, a self-supervised learning strategy that leverages unlabeled images, and a probabilistic framework based on Monte Carlo dropout to quantify uncertainty in predictions. Validation in an automotive manufacturing environment demonstrates 95% classification accuracy while automatically identifying uncertain predictions that require expert review, additionally reducing manual labeling requirements by 70%. This research demonstrates that the gap between state-of-the-art computer vision techniques and industrial quality control requirements can be effectively bridged, enabling automated and reliable defect detection in production lines.