Facilitating Recycling of 6xxx Series Aluminum Alloys by Machine Learning-Based Optimization

Aug 5, 2026

Upload by Rowan

Aluminum alloy grades have expanded over the past century due to aluminum’s strength-to-weight advantages, yet the large variety of grades complicates recycling of aluminum scrap. Aimed at reducing the number of alloying grades while maintaining performance, the study builds an optimization loop combining machine learning with materials-science knowledge. Using 292 datasets across 42 6xxx-series grades under T5, T6, and T7 tempers, clustering and re-clustering identify optimal alloys for sub-clusters, ultimately reducing 42 grades to 10 optimized alternatives.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages12

Reading Time30 min

AccessFree to download

Abstract

Graphical Abstract

Keywords

Introduction

Journal of Sustainable Metallurgy [https://doi.org/10.1007/s40831-025-01](https://doi.org/10.1007/s40831-025-01)112-4 Facilitating Recycling of 6xxx Series Aluminum Alloys by Machine Learning‑Based Optimization TanuTiwari1 · Chamini Mendis1 · Dmitry Eskin1 Received: 13 November 2024 / Accepted: 23 April 2025 © The Author(s) 2025 Abstract Aluminum alloys throughout the last century have experienced extensive development, owing to their unique strength-toweight ratio. This led to generating multiple alloy grades. However, large number of grades present challenges when it comes to the recycling of aluminum scrap, which is the current and future trend in aluminum alloy production and application. Therefore, there is an urgent need to decrease the number of alloying grades while preserving their performance. In this study, we designed an optimization loop based on Machine Learning (ML) and material science knowledge for the 292 sets of data collected on 42 grades of 6xxx series aluminum alloys, focusing on their mechanical, service, and technological properties under T5, T6, and T7 tempering conditions. K-means clustering and principal component analysis algorithms were applied to form various clusters of alloys and are further re-clustered into fine sub-clusters. An optimal alloy (OA) for each sub-cluster was identified based on optimization criteria. After successive iteration, we were able to reduce 42 grades of the 6xxx series into a set of 10 OA’s each performing optimally. This method not only support the capability of machine learning in selecting OA’s but also introduce a future direction for recycling practices in the aluminum industry. Graphical Abstract Keywords 6xxx Series aluminum alloys · Machine learning · Alloy optimization · Recyclability The contributing editor for this article was Zhi Sun. Extended author information available on the last page of the article Introduction Aluminum is the third most common element and the most abundant metal (8%) in the earth's crust. The versatility of aluminum makes it the most widely used structural metal after steel due to its high strength-to-weight ratio, making it easy to design and construct lightweight and sturdystructures [1] . The mechanical properties of pure aluminum are significantly enhanced by the addition of up to 7% of major alloying elements, such as manganese, copper, silicon, zinc, and magnesium. Furthermore, minor alloying elements (less than 0.5%) are added to further improve its properties. Due to the exceptional combination of properties, aluminum demand is projected to double by the year 2050, leading to the continuous development of new aluminum alloys [2]. At the same time the fraction of recycled (scrap) alloys should increase to at least 50%[3–5] . Traditionally, aluminum alloy design relies on trial and error, driven by the domain knowledge of materials researchers and the current requirements of manufacturers, as well as proprietary considerations. This is done by varying alloying element concentrations and processing conditions to improve mechanical properties. However, this approach is extremely time-and cost-intensive and does not cover the vast design space of potential alloys. Materials scientists, in discovering new alloys, often rely on the thermodynamic information presented by phase diagrams. However, the relationship between changes in single input variables and the target property often cannot be interpreted by a human. Recently,'ab initio'methods have been used to discover alloys, involving structural calculations from scratch. However, this approach cannot be generalized for all alloy design issues due to the limitations of the method, a number of assumptions, nonlinearity, and the high dimensionality of alloy property variations with composition [6] . In addition, ab initio approaches do not fully harness the information of alloys that are already known. The enormous complexity due to the interplay of struc

FAQs

Why is recycling of 6xxx-series aluminum scrap challenging?

The recycling challenge arises because many alloy grades exist, making scrap sorting and matching to performance requirements difficult.

How does the study use machine learning in alloy optimization?

It creates an ML-based optimization loop with materials-science knowledge to evaluate mechanical, service, and technological properties and to identify optimal alloys for clustered sub-groups.

What was the final reduction achieved for 6xxx-series alloy grades?

After iterative optimization and re-clustering, the approach reduced 42 original 6xxx-series grades to a set of 10 optimized alloys.

Aug 5, 2026

Upload by Rowan