Materials Informatics for Images and Multi-dimensional Datasets: Poster Session
Sponsored by: ACerS Basic Science Division, ACerS Electronics Division
Program Organizers: Amanda Krause, Carnegie Mellon University; Kristen Brosnan, General Electric Research; Alp Sehirlioglu, Case Western Reserve University

Tuesday 10:00 AM
November 3, 2020
Room: Poster Hall
Location: MS&T Virtual


Identifying Crack Initiation Sites with CNNs: Katelyn Jones1; Elizabeth Holm1; Anthony Rollett1; 1Carnegie Mellon University
    The application of machine learning techniques in materials science has allowed for a greater understanding of microstructure and more efficient analysis of large amounts of data. Convolutional neural networks (CNNs) have been used with images to make connections between microstructure, stress state, and fatigue life. This project uses CNNs on a combination of experimental and simulated image data to identify high stress points that can initiate cracks and cause fatigue failure. We focus on aerospace materials such as alloy 718, which will be studied because of its applicability for high temperature service and cyclic loading. These results will be used to create a model to locate the crack sites before they form and predict the causes of failure and life of future parts. The application of CNNs in this instance, simulations used, and identified causes of crack initiation will be presented.

Keyhole Porosity Threshold in Laser Melting Revealed by High-Speed X-ray Imaging: Runbo Jiang1; Benjamin Gould2; Andy Ramlatchan3; Joseph Aroh1; Anthony Rollett1; 1Carnegie Mellon University; 2Argonne National Laboratory; 3NASA Langley Research Center
    It’s generally accepted that operation outside the conduction regime results in the formation of subsurface trapped gas pores (i.e. keyhole pores). However, the transition between conduction and keyhole mode inducing subsurface porosity is relatively large and not well explored. In-situ high-speed x-ray imaging has proven useful in investigating the keyhole behavior of laser powder bed fusion. The undulation of the area and the variance of front wall angle of the vapor cavity were measured in each time frame in x-ray images. It was found that the boundary in between is not discrete. A threshold of keyhole pore formation in Al6061 and IN718 was determined. Operation under keyhole condition can lead to faster print time. Therefore, if the keyhole porosity threshold facilitates the use of this mode without the formation of porosity, build rate can be significantly improved. Alternative methods are being used to expand this transition regime.