|About this Abstract
||2018 TMS Annual Meeting & Exhibition
||Advanced Characterization Techniques for Quantifying and Modeling Deformation
||Statistical Model Based on Large Field-of-view Images Predicts Microstructural Damage-sites
||Benjamin Cameron, C. Cem Taşan
|On-Site Speaker (Planned)
Predicting in advance where damage events will occur in a metal microstructure could enable preventive engineering treatments. However, despite recently developed novel imaging and characterization tools that can provide vast increases in quality and quantity of microstructure data, predictive capabilities are still limited due to microstructure complexity and spatial uniqueness. To overcome this, we demonstrate an alternative machine learning approach. Specifically, we pre-process a large number of images of deformed dual phase steel, compute the n-point statistics, extract parameters using principle component analysis, and relate these parameters to any damage events that occur. Our model correctly identifies whether 93.1% of our microstructural regions have damage events. This demonstrates significant predictive ability and is a major step forward quantitatively relating the microstructure to local damage events.
||Planned: Supplemental Proceedings volume