About this Abstract |
| Meeting |
TMS Specialty Congress 2026
|
| Symposium
|
World Congress on Reproducibility, Qualification, and Standards Development of Additive Manufacturing and Beyond (RQSD 2026)
|
| Presentation Title |
Multi-Sensor Data Registration Strategy to be Used in Machine Learning |
| Author(s) |
Callie Zawaski, Wesley Mitchell, Matthew Pantano, Griffin Jones, Ted Reutzel, Jan Petrich |
| On-Site Speaker (Planned) |
Jan Petrich |
| Abstract Scope |
In-situ process monitoring for additive manufacturing (AM) has shown significant benefits for part quality assessment and theoretically can predict the lifetime of a printed part. A core requirement for any successful monitoring strategy is spatio-temporal data registration which enables sensor fusion, allowing the data from the distinct imaging and time-series sensors to be combined and analyzed coherently. This presentation details a comprehensive data processing pipeline developed at Penn State's Applied Research Laboratory (ARL) that aims to further facilitate the development of AM data analytics tools. The pipeline is engineered specifically to handle the large data sets characteristic of typical AM builds. We discuss the assembly, calibration, and registration of taking data from multiple sensor inputs. A multi-camera system design, using 4 inputs, is used to capture the for full build plate to achieve a high-resolution image of the surface to enable the capturing of small defects. Time-series process data is collected via a pair of photodiodes to capture the weld plume, similar to how process monitoring is done in traditional welding. Ground truth for defects is established using X-ray Computed Tomography (CT) data. Critically, both the in-situ sensor data and the X-ray CT scan data are precisely transformed and referenced to a unified build plate coordinate frame. The resulting registered data become inputs for machine learning (ML) algorithms, where CT data is used to train the model to predict defects using the in-situ process monitoring sensors. This research was developed with funding from the Defense Advanced Research Projects Agency (DARPA). |
| Proceedings Inclusion? |
Undecided |