Dresden 2020 – wissenschaftliches Programm
Die DPG-Frühjahrstagung in Dresden musste abgesagt werden! Lesen Sie mehr ...
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 11: Topical Session: Data Driven Materials Science - Machine Learning for Damage Prediction
MM 11.2: Vortrag
Montag, 16. März 2020, 16:15–16:30, BAR 205
Large-area, high-resolution characterisation and classification of damage mechanisms in dual-phase steel using deep learning — •Setareh Medghalchi1, Carl F. Kusche1, Tom Reclik1, Martina Freund1, Ulrich Kerzel2, Talal Al-Samman1, and Sandra Korte-Kerzel1 — 1Institut für Metallkunde und Metallphysik, RWTH, Aachen,Germany — 2IUBH University of Applied Sciences,Bad Honnef,Germany
Dual-phase steels are popular in the automotive industry as they allow lightweight design with high stiffness and good ductility. However, their damage behavior is not yet fully understood and their heterogeneity at different length scales impedes a full characterization based on small excerpts of the microstructure. Understanding their damage behavior therefore requires detailed investigations of many damage sites at high-resolution over large areas. To this end, we have collected a large amount of data by means of panoramic imaging in a scanning- electron-microscope before and after deformation following different strain paths. Machine-learning allows us to tackle the challenges of automated analysis of the microstructure of dual phase steel samples. A deep-learning based algorithm has been developed to classify the detected damage sites in the microstructure. Furthermore, we have now enhanced the accuracy and robustness of our method by data-augmentation, making it applicable on samples which were subjected to different deformation conditions, e.g. uniaxial or biaxial tensile testing. This reduces the need for manual interventions, aiding the high-statistics-microstructural-analysis under variable conditions.