Development and Implementation of a Deep Learning Algorithm to Evaluate the Powder Distribution Process During 3D Printing Using the LPBF Method

This article presents research work on an intelligent system that was developed to monitor and continuously evaluate the quality of metal powder distribution in the laser powder bed fusion (LPBF) 3D printing process. The 3D printer that was used to carry out the work was equipped with an industrial vision system to capture images immediately after spreading powder on the work field. The powder distribution tests showed that the most common defects were identified as an insufficiently thick layer of powder applied to the working field (super elevation), unevenly distributed powder as a result of recoater vibration (so called recoater hopping), and its wear (so called recoater streaking). In the first stage of research, a set of training data (images) was collected. Then, the implementation of the machine learning process was prepared in the Roboflow environment. After that, the learning, validation, and prediction process was carried out several times using the selected machine learning model (YOLO model implemented in a Python environment) in order to select the most effective parameters. The study showed that deep machine learning can be effectively used to identify defects in powder distribution during the laser powder bed fusion (LPBF) process.