PNNL

Abstract

Additive manufacturing (AM) has gained popularity across a variety of industry sectors for its unique capabilities, but defects affect the overall quality and integrity of the AM-produced part. As a result, there is increasing interest in in-situ monitoring for defects, specifically keyhole pores. Acoustic data collected during the AM process can provide nondestructive insights into pore formation. In this study, we use a convolutional neural network to predict pore formation from acoustic data to investigate its application as an in-situ monitoring technique. Acoustic data was collected during laser powder bed fusion (L-PBF) experiments performed at Lawrence Livermore National Laboratory, and pores were identified post-build using radiography. Due to intrinsic process controls of AM, where a pore-forming event is relatively rare, the dataset is imbalanced. The mixup data augmentation technique is applied to balance the dataset. When data augmentation is applied, we show that pores can be successfully identified from acoustic emissions with up to 99% accuracy.