The Multisensor Insitu Data with Machine Learning (MIDML) program will develop a convolutional neural network (CNN) to leverage data generated from multiple AM inprocess sensors. This can provide more accurate and reliable assessments and predictions of final part quality during layer-by-layer fabrication, in real time. The CNN will be developed for maximum prediction accuracy from multiple sensor inputs using CT scans of the finished part to provide ground truth of final melt quality. Employing insitu data, corrective action can be taken much earlier in the manufacturing cycle to improve part quality, process yield and cost effectiveness of AM for critical applications. The MIDML program will initially leverage LPBF data already developed under six prior AM inprocess inspection contracts performed by Quartus and its partners for DoD and NASA. This includes inprocess and final part microCT data that has been volumetrically registered. This approach permits starting CNN development immediately on Day 1. Then, during the remainder of the Phase I BASE program, we will fabricate new test specimens while capturing three inprocess sensing modalities simultaneously: thermal tomography, laser profilometry, and melt pool monitoring. Coupons will be microCT scanned and the data used to refine and then test the effectiveness of our CNN.
Benefit: The MIDML program provides the Navy with significant
Benefits: Improves the prediction accuracy of final part quality from insitu sensor data by leveraging disparate sensor modalities and final part CT data. Effective insitu inspection allows parts made with fatal (i.e., nonrepairable) flaws to be abandoned as early as possible (at the layer level) at the lowest possible cost of the scrapped part. Effective inprocess or closed loop control may later be developed to permit repairable flaws found by insitu inspection to be repaired and salvaged mid Build. This reduces LPBF process scrap rate. The combination of insitu inspection and repair can improve confidence in AM parts for critical Navy applications such as fracture critical parts, thereby allowing the Navy to use AM more broadly.
Keywords: Machine Learning (ML), Machine Learning (ML), In-Situ Inspection, laser powder bed fusion (LPBF), Neural Network (NN), Convolutional Neural Network (CNN), Additive Manufacturing (AM)