High-throughput computational screening of Al-based alloys from scrap metal mixtures

Abstract

In order to meet current societal challenges such as climate neutrality by 2050, circular economy, bottlenecks in raw material supply chains, increasing raw material and energy costs, material discovery must become faster and more flexible. New modern technological developments such as rapid computational screening, fast 3D/4D characterization or machine learning aided data analysis enable to respond quickly to socio-political changes. A new Al-based alloy suitable for additive manufacturing (AM) developed from scrap metal mixtures by using a simulation based high throughput alloy screening approach, complemented by experimental data and validation. As the trend moves towards combustion-less engines, Al-Si piston alloys will have no further use in the near future, it is imperative to consider how to exploit soon available large quantity of high-quality scrap. Scrap from 2xxx Al alloys used in aerospace and 6xxx Al alloys available in a wide range of sectors were selected as additional alloys for mixing. The developed alloy is intended for aerospace applications and has to meet requirements such as low sensitivity to hot-cracking, appropriate strength and elongation and corrosion resistance. For the high-throughput computational alloy screening, up to 10k alloy designs were initially generated based on mixing ratios between the three available scrap alloys. The CALPHAD method was used to run equilibrium and non-equilibrium calculations to calculate relevant thermo-physical and mechanical variables, such as content of volatile elements such as Mg and Zn, phases forming and their fractions, solidification intervals, thermo-physical parameters, yield strength and hot cracking sensitivity, for each design and a sensitivity analysis was carried out to find correlations within this complex material system. A random forest surrogate model was used to predict properties based on simulations, experiments and literature. Finally, the acquired comprehensive database was combined with multi-objective evolutionary algorithms to filter designs with appropriate target properties. Uncertainty calculations account for the expected compositional variation within the Al scrap due to elemental additions and to only obtain robust designs that are tolerant to such variations. Complementary experimental validation such as SEM/EDX, laser flash analysis (LFA), DSC, synchrotron-radiation-based techniques and laser track experiments were conducted to validate the simulation models and identify 1 - 2 scrap mixtures for actual powder production and LPBF processing.