Algorithmic Solutions for Third Party Low-Cost Sensing of LPBF

Quartus’ Neural Networks for In-situ Inspection of LPBF (NNIIL) program leverages the latest machine learning and convolutional neural network (CNN) methods to perform data processing of manufacturing information captured during in-situ inspection of laser powder bed fusion (LPBF) parts. This methodology has the potential to generate more accurate flaw detection results with much less development time and effort than traditional data processing with explicit algorithms. Preliminary trials using CNNs trained on actual LPBF data and in a few days’ time have approached the accuracy achieved with more traditional algorithms developed over a much longer period. Quartus will develop these methods using actual LPBF data already collected under DLA and NASA funded programs performed by team partner Flightware, thus accelerating this work and leveraging prior Government investment. Quartus has performed prior software tasks for Flightware as a subcontractor on these programs, and is therefore very familiar with many of the data processing issues. In this program, Quartus and Flightware have agreed to switch prime and subcontractor roles, because the majority of the effort needed lies with Quartus. Flightware will support this work as a subcontractor, and has agreed to allow Quartus to access its LPBF profile data.