About this Abstract |
| Meeting |
MS&T26: Materials Science & Technology
|
| Symposium
|
Additive Manufacturing: Equipment, Instrumentation and In-Situ Process Monitoring
|
| Presentation Title |
Beyond Deep Learning: A Bayesian-Optimized Computer Vision Framework for Rapid Spatter Detection and Tracking in Laser Powder Bed Fusion |
| Author(s) |
Monish Piraimudi, Nicholas O’Brien, Jashanpreet Saini, Jack Beuth |
| On-Site Speaker (Planned) |
Monish Piraimudi |
| Abstract Scope |
Spatter in laser powder bed fusion (L-PBF) contributes to defect formation. Prior works have shown optical sensors can image spatter across the build area, but they often use slower, deep learning approaches, which are not practical for high-speed monitoring. Traditional computer vision filters, however, are faster but are less versatile and require careful fine-tuning and postprocessing. This research has developed a lightweight computer vision framework for accurate and rapid spatter detection and tracking in a video dataset. The framework optimizes an ensemble of blob detectors using Bayesian optimization. Combined with a Kalman filter for temperature history-based size estimation, the framework is applied to infrared video of spatter generated by an Inconel 718 part in keyholing and nominal printing conditions. The detections are validated with spatter particles captured on sticky plates during the build. The framework architecture, information derived from detections, and implications for in-situ L-PBF process monitoring will be discussed. |
