A Semi-Automated Image Processing Tool for Additive Manufacturing Image Segmentation

Image segmentation is a crucial technique for evaluating part fidelity and identifying print defects in Advanced Manufacturing (AM). Open-source tools such as ImageJ and Fiji, and proprietary software, such as Dragonfly are commonly used for image analysis. While these GUI-based tools allow users to leverage domain knowledge to identify features, they rely heavily on subjective decisions, require substantial expertise and experience, and are not scalable for the volume and rate of data generated. Machine learning-based approaches address the scalability issue, but require extensive manual annotations and large domain-specific datasets which are often challenging to obtain due to data sharing concerns in the materials science community. To address these challenges, we developed a semi-automated image processing tool that strikes a balance between automation and manual control. The tool includes an initial segmentation pipeline as a starting point, an interface for fine-tuning segmented results to improve accuracy and real-time feedback to visualize and refine adjustments to the segmentation output. Additionally, the tool records user modifications to facilitate reproducibility and enable future refinement of segmentation processes. This tool is demonstrated on Laser Powder Bed Fusion (L-PBF) datasets as a case example, highlighting its potential to generalize to other image modalities and datasets. Additionally, it provides statistical morphological properties of the segmented features. This semi-automated tool is a crucial step towards advancing image analysis methods for the AM community, improving both reproducibility and efficiency in image characterization.

This material is based upon research in the Materials Data Science for Stockpile Stewardship Center of Excellence (MDS3-COE), and supported by the U.S. Department of Energy's National Nuclear Security Administration under Award Number(s) DE-NA0004104 and partially under the auspices of Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344, LLNL-ABS-872322.