About this Abstract |
Meeting |
MS&T22: Materials Science & Technology
|
Symposium
|
Additive Manufacturing Modeling, Simulation, and Machine Learning: Microstructure, Mechanics, and Process
|
Presentation Title |
Parent Grain Reconstruction Using Orientation Imaging Microscopy and Deep Learning |
Author(s) |
Patxi Fernandez-Zelaia, Andres Marquez Rossy, Quinn Campbell, Andrzej Nycz, Chris Ledford, Michael Kirka |
On-Site Speaker (Planned) |
Patxi Fernandez-Zelaia |
Abstract Scope |
We present a deep convolutional neural network model which estimates parent grain austenitic structure from observed child phase martensite electron backscatter diffraction micrographs. The model was trained using only four micrographs by using a novel data augmentation strategy that exploits properties of the orientation representation. Despite the model being trained on a traditionally fabricated martensitic alloy, when tested on additively manufactured material, which exhibits vastly different austenitic grain structure, the model performs exceptionally well. The model efficacy depends on the microstructure length scale and receptive field of the vision model. This work demonstrates that modern machine learning vision models are well suited for analyzing the complex spatial-orientation structure found within orientation imaging micrographs. The model evaluates much faster than traditional approaches which may enable rapid characterization of additively manufactured materials. |