About this Abstract |
Meeting |
2025 TMS Annual Meeting & Exhibition
|
Symposium
|
Additive Manufacturing Modeling, Simulation and Machine Learning
|
Presentation Title |
Real-Time Detection of Keyhole Pore Generation in Laser Powder Bed Fusion Via a Multi-Sensor System and Physics-Informed Machine Learning |
Author(s) |
Jiayun Shao, Zhongshu Ren, Tao Sun |
On-Site Speaker (Planned) |
Jiayun Shao |
Abstract Scope |
Keyhole porosity has been a major obstacle that hinders the widespread adoption of laser powder bed fusion. In this study, data fusion with features extracted using SqueezeNet neural network (NN) architecture and classification using different machine learning algorithms were adapted to locally detect the generation of keyhole pores with a suite of photodiode, 100kHz microphone, 20kHz microphone and near-infrared (NIR) cameras. Operando synchrotron x-ray imaging was employed to obtain high-fidelity ground truth data. The influence of data variability, different combination of sensors, frequency bandwidth, etc. was investigated. We found that single NIR camera signal gave us the highest keyhole porosity prediction accuracy, followed by 100kHz microphone, 20kHz microphone and photodiode with the lowest accuracy. 94.6% accuracy with a temporal resolution of 100μs was achieved by optimizing all the aforementioned aspects. |
Proceedings Inclusion? |
Planned: |
Keywords |
Additive Manufacturing, Machine Learning, Titanium |