Combination of acoustic-ultrasonic monitoring and machine learning for layer-by-layer defect detection in PBF-LB/M
Résumé
One of the major challenges in Powder Bed Fusion using a Laser Beam for Metals (PBF-LB/M) additive manufacturing is the presence of sub-surface defects and porosities, which significantly reduce the fatigue performance of produced parts. Although various in-situ monitoring strategies have been proposed, the accurate and early detection of porosity defects during actual manufacturing remains a critical challenge. Existing monitoring methods often suffer from high costs, limited precision, or poor industrial applicability. This study proposes an integrated, cost-effective approach for real-time in-situ quality control of PBF-LB/M processes to detect defects in each layer. The analysis investigates how acoustic-ultrasonic signal signatures reflect process variations associated with the formation of porosity defects in overhang zones. Furthermore, by comparing several machine learning approaches, this study provides a comprehensive evaluation of the relative performance of different algorithms. The comparative analysis demonstrates that the artificial neural network achieves the best performance, with a success rate of 92% in both accuracy and F1-score, while also highlighting the method's scalability and practical applicability for PBF-LB/M process monitoring. The findings confirm the effectiveness of the layer-by-layer monitoring framework based on acoustic-ultrasonic sensing and 1 machine learning, offering a practical and scalable solution for improving defect detection and overall part quality in industrial PBF-LB/M processes.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |