An Image Segmentation Framework for In-Situ Monitoring in Laser Powder Bed Fusion Additive Manufacturing

Abstract

Laser powder bed fusion (LPBF) is an additive manufacturing technique that offers a number of competitive advantages over conventional manufacturing methods; however, challenges in assuring final part quality hinder its broader industrial adoption. As a step towards closing the loop in quality control for LPBF, this paper presents an image segmentation framework for extracting process characteristics from in-situ image data collected by a coaxial imaging system in the visible spectrum. Since manual data annotation can be time-consuming and error prone, we present an image processing pipeline for generating precise segmentation labels semi-automatically. The resulting dataset is used to train a machine learning model for segmentation of processed zone boundaries for in-situ monitoring. Experimental results validate the effectiveness of our approach for process monitoring in LPBF across a variety of challenging illumination and process conditions.