h1

%0 Thesis
%A Medghalchi, Setareh
%T Investigation of damage sites in dual phase microstructures by artificial intelligence
%I Rheinisch-Westfälische Technische Hochschule Aachen
%V Dissertation
%C Aachen
%M RWTH-2024-06967
%P 1 Online-Ressource : Illustrationen
%D 2024
%Z Veröffentlicht auf dem Publikationsserver der RWTH Aachen University
%Z Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2024, Kumulative Dissertation
%X The current thesis delves into the multifaceted analysis of damage formation in advanced metallic materials with heterogeneous microstructure under various loading conditions, providing invaluable insights into damage mechanisms, stress-state dependence, strain rate effects, and three dimensional characterisation as well as phase distribution. The studies encompass dual-phase steels and a cast MgAlCa alloy, shedding light on diverse aspects of damage behaviour and its implications for material performance. Comprehensive high-resolution data acquisition from relatively large experimental sample surfaces is mandated to elicit statistically robust inferences, given the profound influence of manufacturing processes on microstructural attributes. Predominantly, the vast heterogeneity within industrially manufactured sheets remains elusive to traditional characterisation techniques, which are mostly constrained to relatively diminutive analysis scopes, thus undermining the representativeness of empirical scrutiny. Recent advancements in deep learning, particularly convolutional neural networks (CNNs), have proven effective in analysing large datasets and extracting visual features in materials science. Six research publications are comprised in this thesis, in which the image data of deformation induced damage sites, collected by scanning electron microscope (SEM) across relatively large areas, have been characterised and quantified from different perspectives. Publication 1, explores the application of deep learning in damage analysis for dual-phase steel (DP800 specifically), extending the methodology from uniaxial to biaxial straining conditions. Leveraging data augmentation, the research enhances network performance and investigates changes in damage sites’ behaviour, particularly the martensite crack angle distribution, in response to different stress states. Facilitated by the enhanced method in publication 1, the damage behaviour of DP800 steel is characterised under single and combined strain paths, revealing a strong connection between loading direction and damage formation in publication 2. The study employs the enhanced deep learning based method, for damage quantification, finite element modelling for stress state determination, and nanoindentation for strain hardening assessment. Publication 3 focuses on a cast MgAlCa alloy and explores the rate dependence of damage formation at elevated temperatures. Utilising convolutional neural networks, the research identifies changing dominant damage mechanisms with varying strain rate. It demonstrates that the strain rate influences the evolution of microcracks and interface decohesion, shedding light on the influence of plastic co-deformation of the α-Mg matrix and Laves phase as well as thermally activated processes at the interface. Publication 4 addresses the three dimensional characterisation of the damage sites in dual phase steel, presenting a comprehensive methodology that combines metallographic serial sectioning, and deep learning assisted automatic image analysis. This approach allows for the analysis of thousands of individual damage sites in three dimensions, offering a unique perspective on active damage mechanisms during deformation. In publication 5, phase segmentation of dual phase steel has been accomplished through the implementation of two deep learning algorithms developed in this work. They have facilitated the calculation of precise martensite phase fraction with statistical relevance, and the extraction of statistics pertaining to the geometrical properties of damage sites and the surrounding microstructure such as phase fraction. In publication 6, a deep learning algorithm was trained using a large set of segmented microstructural images, generated in publication 5, and corresponding finite element simulation results, demonstrate promising potential for predicting mechanical stress and strain fields. This has the potential to bridge computational simulations and experimental materials science, expediting stress and strain analysis for robust material design strategies. Collectively, these studies provide a holistic understanding of damage formation in two groups of advanced metallic materials, considering factors such as stress state, strain path, 3D configuration and local microstructural properties. Such insights are invaluable for optimising material performance in real-world applications, such as automotive engineering and beyond.
%F PUB:(DE-HGF)11
%9 Dissertation / PhD Thesis
%R 10.18154/RWTH-2024-06967
%U https://publications.rwth-aachen.de/record/989755