Aluminum (Al) plays a crucial role in the circular economy, contributing to a sustainable future due to its ability to be recycled indefinitely often without losing its properties at a reduced energy consumption by up to 95\% compared to primary Al production. In the conventional reference recycling technology, i.e., remelting, one of the most difficult types of Al scraps to be recycled are Al machining chips because a large fraction of their surface area is covered with high-melting-point oxides and the accumulation of deleterious impurities. Brittle oxide inclusions and deleterious impurities can act as stress concentrators and may promote defect formation. Solid-state materials processing through friction extrusion (FE) emerges as a promising recycling technology due to utilization of frictional energy to generate heat, break down its oxides and consolidate the Al chips. Additionally, FE can suppress the formation of other deleterious inclusions and intermetallics as they commonly appear during solidification in remelting. FE can also be used to upcycle Al chips, for instance with additional reinforcement particles, to a metal matrix composite as proposed in this project. Establishing structure-property relationships using micromechanical models allows for a quantitative understanding of the effects of oxides on the mechanical properties of the recycled and upcycled Al chips. These models rely on 3D microstructural images that are expensive and time consuming to obtain experimentally. Generative artificial intelligence and stochastic geometry provide versatile tools to overcome these limitations, where more readily available 2D image data of a relatively small number of specimens can be used to calibrate parametric models that allow for the generation of virtual, but realistic 3D microstructures. In the present project, we will leverage experimentally measured data of recycled Al, parametric stochastic 3D modeling and micromechanical modeling to quantify process-structure-property relationships for the FE process of various feedstocks. In particular, the project aims to establish an experimental basis for FE of Al chips with varying oxide content to directly enhance the recyclability and sustainability of Al chips, while also utilizing Al oxide as a reinforcing agent to produce value-added materials such as metal matrix composites. We pursue a quantitative characterization and prediction of 3D microstructures via data-driven stochastic microstructure modeling approaches in combination with generative adversarial networks to predict 3D microstructure based on 2D image data. Furthermore, after establishing quantitative process-structure-property relationships by combining experiments with data-driven stochastic and micromechanical modeling, an inverse design strategy will be pursued to identify process parameters that lead to desirable microstructures, and to determine microstructures with tailored mechanical properties.