Automated porosity assessment of parts produced by Laser Powder Bed Fusion using Convolutional Neural Networks

Abstract

Laser Powder Bed Fusion (LPBF) is especially interesting for applications in industries with high quality requirements. There are different expensive and time-consuming strategies for quality assurance. A cheaper and faster approach is to analyze the data acquired during fabrication. In this work Convolutional Neural Networks (CNN) are investigated as a tool for data analysis of meltpool monitoring data. The goal is to automatically distinguish between porous and non-porous part regions. Therefore, the training data is categorized based on CT-scans of the test specimens. For increased interpretability of the results, Gradient-Weighted Class Activation Maps (Grad-CAM) are used.