New dataset for smarter 3D printing released

Oak Ridge National Laboratory’s Peregrine software, used to monitor and analyze parts created through powder bed additive manufacturing, has released its most advanced dataset to date.

The dataset is titled “In situ Visible Light and Thermal Imaging Data from a Laser Powder Bed Fusion Additive Manufacturing Process Co-Registered to X-ray Computed Tomography and Fatigue Data.”

In its ongoing effort to support the nation’s additive manufacturing industry with comprehensive datasets, the Department of Energy’s Manufacturing Demonstration Facility produced this new dataset as part of a study to establish strong correlations between manufacturing anomalies, internal defects, and resulting mechanical performance.

This dataset contains state-of-the-art monitoring data for laser powder bed fusion (L-PBF), which uses a laser to melt and fuse metal powder to create the layers of a metal part. The dataset includes machine process parameters and sensor data, geometries, and detailed images of the 3D-build process captured from multiple angles and lighting types, combining high-resolution visible and near-infrared imaging along with X-ray scans of the printed parts.

“Peregrine takes images during printing, using AI to look for anomalies,” said Luke Scime, a researcher in the Manufacturing Systems Analytics Group at ORNL.

“You do that for every single layer, and you build up a three-dimensional map of all the locations that might have issues, and then you try to predict which of those might cause a problem in the final part.”

The Peregrine software’s custom algorithm uses pixel values of images to scrutinize the composition of edges, lines, corners, and textures, and sends an alert to operators about any problems during the printing process so they can make quick adjustments.

Through its Dynamic Multilabel Segmentation Convolutional Neural Network, or DMSCNN, Peregrine looks at data from multiple sensors to detect problems and send an alert. For instance, L-PBF prints experience spatter, where molten material is ejected as the laser melts the metal powder. These spattered particles can land elsewhere on the part, affecting the overall quality.

The new dataset includes all DMSCNN segmentation results and fatigue-tested specimens subjected to such spatter-induced perturbations.

This comprehensive ensemble of information supports AI model development for digital qualification of AM processes. By using the improved open-source Peregrine dataset, researchers and manufacturers can develop even smarter, adaptive quality assurance and quality control systems for their 3D-printed parts.

Other ORNL researchers who contributed to the new dataset include Zackary Snow, Chase Joslin, William Halsey, Andres Marquez Rossy, Amir Ziabari, Vincent Paquit, and Ryan Dehoff.