Electron backscatter diffraction (EBSD) is a technique that describes the orientation of crystals in a sample. A series of deep convolutional neural networks was built to determine the orientation of a polycrystalline sample based on its electron diffraction patterns. By establishing a fixed coordinate system, a mathematical model will take jpg files of a series of rotations as training data at a rate of approximately 250 seconds on each epoch. The networks determine the orientation of a sample image at substantially higher orders than traditional physics-based forward models. Statistical techniques are applied within the model to attain a loss in accuracy of only 5-8 degrees for rotations between 0 and 360 degrees. This simple and robust python software tool can be utilized in further offline analysis to index electron diffraction patterns much more efficiently while maintaining an average level of accuracy greater than or equal to 80%.
5:00 PM–7:00 PM Apr 24, 2019 (US - Arizona)
PCC North, 300 Level, Exhibit Hall C-E