Application of machine vision for the detection of powder bed defects in additive manufacturing processes

Abstract

The quality of the powder layers in the 3D printing process is extremely important and directly corresponds to the quality of the structures made with this technology. Therefore, it is essential to control it. It can be made in-line with a vision system combined with image processing algorithms, which can significantly improve control of the process and help with the adjustment of powder spreading systems, especially in case of difficult-to-feed powders like magnetic ones - e.g., Fe-based metallic glass powder - Fe56.04Co13.45Nb5.5B25. In this work, two algorithms - machine learning - Support Vector Machines (SVM), deep learning - Convolutional Neural Networks (CNN) - were evaluated for their ability to detect and classify the enumerated anomalies based on powder layer images. The SVM algorithm makes it possible to efficiently and quickly analyze the powder-spreading process. CNN, however, appears to be a more promising choice for the developed application, as they alleviate the need for complex image operations.