Machine Learning and Data Mining for Enhanced Efficiency of Dislocation Simulations and Microstructure-Property Relations A. Demirci1,∗, S. Sandfeld1,2 1 Institute for Advanced Simulations – Materials Data Science and Informatics (IAS-9), Forschungszentrum Jülich GmbH, 52425 Jülich, Germany 2 Chair of Materials Data Science and Materials Informatics, Faculty 5 – Georesources and Materials Engineering, RWTH Aachen University ∗ a.demirci@fz-juelich.de Keywords: dislocation, discrete dislocation dynamics, plasticity, machine learning Discrete Dislocation Dynamics (DDD) simulations are highly precise methods used to determine the plastic deformation of materials at the micro level. These simulations directly quantify dislocation movements and interactions, which are the fundamental physical processes involved in the plastic deformation of crystalline materials. The complex and dataintensive nature of DDD simulations requires multiple steps to post-process and analyze data to understand the material behavior. Data mining and machine learning methods can significantly aid in the analysis of the extensive data generated by these simulations. For instance, these techniques can provide valuable insights into dislocation mechanisms, such as the cross-slip mechanism, which are not readily observable through experiments. Furthermore, surrogate models and hybrid simulations can remedy the computational burden of DDD simulations by identifying and utilizing the relationships between microstructure and properties in solving forward problems. Our study primarily focuses on identifying these relationships within microstructures through data mining and machine learning techniques and explores the key descriptors that outline the behaviors and evolved microstructures in dislocation simulations. We also present our initial promising attempts at using physics-based Deep Learning methods as a replacement for certain aspects of DDD simulations. 110
RkJQdWJsaXNoZXIy MjM0NDE=