| About this Abstract |
| Meeting |
2025 TMS Annual Meeting & Exhibition
|
| Symposium
|
Additive Manufacturing Modeling, Simulation and Machine Learning
|
| Presentation Title |
Machine Vision-enabled Defect Characterization in Additively Manufactured Steels |
| Author(s) |
Can Uysalel, Jackelin Cotrina, Maziar Ghazinejad |
| On-Site Speaker (Planned) |
Can Uysalel |
| Abstract Scope |
In this study, we utilized Convolutional Neural Networks (CNNs), a subset of Machine Learning (ML), to classify defects in 3D-printed steel samples, aiming to advance a more efficient vision-based metrology technique for metal additive manufacturing. We concentrated on using Selective Laser Melting (SLM) as our primary 3D-printing method. Leveraging a library of SEM images for 3D-printed steel samples enabled us to train our CNN algorithm to recognize and classify the primary types of metal printing defects based on their geometry and composition. The main defects targeted include microcracks, lack of fusion, balling effect, gas porosity, and impurities. After training our ML model with the SEM dataset, we employed the CNN algorithm to characterize newly SLM-printed stainless steels. Confusion matrices were applied to dissect the predictions of our trained model and assess its accuracy. Our results indicate a promising potential of physics-informed machine vision in automated metrology of additively-manufactured metals. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Characterization, Additive Manufacturing, Machine Learning |