Multisensor Insitu Data with Machine Learning Phase II

The goal of this program is to develop a Multi-Sensor Flaw Prediction (MSFP) Software for laser powder bed fusion (LPBF) applications. The ??????software willĀtakeĀinput data from multiple in situ sensors monitoring the LPBF build process and predictĀflaw formation and flaw type. The software development will start with prototyping the core data processing module which includes a sensor fusion component and a machine learning component ultimately performing flaw predictions. This core software module will be matured after exploring several sensor fusion and machine learning architectures, and downĀselecting the best performing approach. Furthermore, the core sensor fusion and machine learning software module will be integrated into two different software products: A standalone GUI software for post processing inspection, and an AMSENSE plugin which can make near real-time flaw predictions during a build. ??????? ???????This effort will leverage the AMSENSE sensor suite developed by the Open Additive group ofĀArctos Technology Solutions, LLC. The existing sensor suite includes five in situ sensing modalities: Thermal Tomography (NIR Imaging), Spatter Tracking, Long Wave IR Imaging, Coaxial Melt Pool Thermal Imaging, and Recoat Imaging. During this program, the sensor suite will be updated to add an acoustic sensor, and to substantially increase the frame rate of the Coaxial Meltpool Camera.Ā Training and test data for the machine learning pipeline will be obtained by executing several builds throughout the program, yielding in situ data, and CT post inspection data for all builds. Test builds will be performed throughout the program, with the goal of incrementally improvingĀdata capture, sensor fusion, and machine learning performance. Early builds will consist of simple test coupons with seeded flaws and registration features for sensor alignment. Later in the program, the flaw prediction system will be trained and tested on builds of more complex parts. Ground truth flaw information will be obtained via analysis of CT scans performed on all parts produced during the program.