3D Computer Vision and Machine Learning for Build Microstructures

In this project, we apply the tools of computer vision (CV) and machine learning (ML) to discover processing-microstructure-property relationships for AM builds. To achieve this goal, we capitalize on CMU’s access to unique 3D structural data and expertise in developing CV/ML systems for the analysis of micrographs. We first collect 3D image data derived from computer simulations and 3D computed tomography experiments, and we apply CV methods to encode the visual information contained in the 3D images. We then use supervised and unsupervised ML methods to develop advanced microstructural metrics and PSP relationships. Finally, we utilize supervised ML to learn the 3D microstructure of an AM build from inexpensive 2D microscopy images. This project adopts as its tenent that Processing-Structure-Properties (PSP) relationships can be built by parameterizing the images used to quantify microstructure. The practical challenge is to be able to relate the more detailed 3D information available from x-ray micro-computed tomography (CT) to the limited 2D information available from metallographic cross-sections.