| About this Abstract |
| Meeting |
2021 TMS Annual Meeting & Exhibition
|
| Symposium
|
Advanced Real Time Imaging
|
| Presentation Title |
In-Situ Machine Learning Enabled Spatter Detection in Laser Powder Bed Fusion Additive Manufacturing |
| Author(s) |
Brandon Abranovic, Jack Lee Beuth, Rishikesh Magar, Lalit Ghule, Amir Barati Farimani |
| On-Site Speaker (Planned) |
Brandon Abranovic |
| Abstract Scope |
This work focuses on the analysis and classification of fusion images from laser powder bed fusion (LPBF) additive manufacturing processes for the detection of laser spatter on the powder bed. In-situ laser spatter identification is a key component in assuring build quality since it can be correlated to severe failures later in the building process, making it of vital interest to the additive manufacturing community. The work relied upon the thousands of fusion images normally collected in the operation of the EOS M290 LPBF system. Various convolutional artificial neural network architectures including AlexNet, VGG16, and UNnet were tested for both identification of spatter behavior on the powder bed as well as localization of spatter flaws in the powder bed. Preliminary results have shown substantial promise for these approaches in the identification of spatter yielding identification accuracy of roughly 90% and patch-wise segmentation accuracy of roughly 87%. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Additive Manufacturing, Machine Learning, Process Technology |