| About this Abstract |
| Meeting |
2024 TMS Annual Meeting & Exhibition
|
| Symposium
|
Additive Manufacturing Modeling, Simulation and Machine Learning
|
| Presentation Title |
Deep Neural Network for Image Segmentation and Feature Quantification during Laser Powder Bed Fusion Additive Manufacturing |
| Author(s) |
Wei Li, Rubén Lambert-Garcia, Anna Getley, Kwan Kim, Shishira Bhagavath, Peter D Lee, Chu Lun Alex Leung |
| On-Site Speaker (Planned) |
Chu Lun Alex Leung |
| Abstract Scope |
High-speed synchrotron X-ray imaging has been employed to detect the dynamic behaviour of molten pools in additive manufacturing (AM). Due to the ultra-high spatial-temporal resolution of X-ray beams, a large volume of imaging data is generated, making it time-consuming to perform manual processing and analysis. Therefore, we proposed an efficient and robust deep neural network for automatic image segmentation and feature analysis. A novel lightweight convolution block and customised attention mechanism were used to achieve remarkable computation efficiency without sacrificing the model’s accuracy and robustness. We built a large-scale benchmark database for network training and testing, consisting of over 10,000 pixel-labelled X-ray images collected from different synchrotron experiments. Experimental results demonstrate that our model can perform semantic segmentation and feature quantification with exceptional accuracy and speed. Its application can also be extended to other AM experiments using the transfer learning technique. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Additive Manufacturing, Machine Learning, |