SKM 2023 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
MM: Fachverband Metall- und Materialphysik
MM 33: Topical Session: Fundamentals of Fracture – Amorphous Metals
MM 33.8: Talk
Wednesday, March 29, 2023, 17:45–18:00, SCH A 215
Using deep neural networks to bridge the gap between statistical fractographic analysis and fracture toughness prediction for polymers — •Guillaume de Luca1,2, Mohammed Idri2, and Laurent Ponson1,2 — 1Institut Jean le Rond d’Alembert, Sorbonne Universite, CNRS, 4 place Jussieu 75006 Paris, France — 2Tortoise, 231 rue Saint-Honoré, 75001 Paris, France
We propose to deploy DNNs to bridge the gap between statistical fractography and the toughness KIc prediction for polymers as well as shed light on the role played by the different structures on their fracture surfaces. We generate the fracture surfaces in laboratory through tensile fracture tests while using DIC (Digital Image Correlation) to locally measure and compute the sought mechanical properties along the crack. An interferometric profilometer extracts the topography from the fracture surfaces, and the resulting height fields are post-treated with tools from statistical fractography to be used as input, while the experimental data are used as labels for the regression problem.
By doing so, we can estimate from a scanned fracture surface a material toughness value KIc(x) along the crack propagation direction. Furthermore, the advancements in explainable neural networks allow us to go one step further by making assumptions about what roughness elements present on the fracture surface influence the most the results coming out of the pipeline.