About this Abstract |
| Meeting |
MS&T26: Materials Science & Technology
|
| Symposium
|
Additive Manufacturing Modeling, Simulation, and Machine Learning: Microstructure, Mechanics, and Process
|
| Presentation Title |
A Neural Network Approach for the Automated Classification of Material Textures |
| Author(s) |
Asimit Bhattarai, Marc De Graef |
| On-Site Speaker (Planned) |
Asimit Bhattarai |
| Abstract Scope |
This study investigates a neural network application for the classification of crystallographic textures by utilizing a 2D visualization representation derived from Clifford Torus mapping. Compared to traditional representations, Clifford Torus mapping provides 2D images from orientation data optimized for machine learning training while preserving the orientation distribution function (ODF) of the material textures. The synthetic orientation datasets are generated from a diverse set of FCC texture groups and crystallographic fibers to create standardized 2D image inputs. The machine learning framework utilized is a convolutional neural network (CNNs) with circular boundary condition to process the periodic nature of the orientation space and extract efficient features of the texture images without the distortion that exists in Euclidian projections. This study shows that a neural network approach can accurately automate the classification of material textures, offering an alternative to traditional manual indexing or dictionary indexing. |