Aluminum alloys, particularly the 7xxx series (Al-Zn-Mg-Cu), are widely used in aerospace and are showing increasing potential in rail transportation due to their excellent properties and manufacturability. However, the emergence of competitive materials like magnesium and titanium alloys necessitates further performance improvements in 7xxx alloys to maintain their competitiveness. A critical aspect is mechanical strength, with commercial 7xxx alloys typically exhibiting UTS below 700 MPa. While advanced techniques like severe plastic deformation (SPD) and rapid solidification/powder metallurgy (RS/PM) can achieve UTS exceeding 750 MPa, these methods are often limited by scalability, cost, and complexity, hindering widespread industrial application. Therefore, optimizing alloy composition presents a practical approach to developing high-strength 7xxx alloys suitable for industrial production. 7xxx alloys typically contain Zn, Mg, and Cu as primary alloying elements, along with trace elements such as Cr, Mn, Zr, Ti, Sc, etc. Research efforts have focused on tailoring the content of these elements, with high Zn and Mg contents often linked to ultra-high strength but also increased susceptibility to localized corrosion, hot tearing, and macro-segregation. Zr is crucial as a grain refiner and anti-recrystallization agent, while the combined addition of Zr and Ti enhances strengthening. Rare-earth elements like Sc offer potent grain refinement and recrystallization inhibition but are expensive. The wide composition range and sensitivity of 7xxx alloys suggest many undiscovered compositions with superior properties. However, traditional trial-and-error methods are inefficient given the complexity of the alloy system and its numerous processing steps. This study utilizes machine learning to overcome these challenges and efficiently discover novel 7xxx alloys with enhanced UTS.

The literature extensively documents efforts to improve the mechanical properties of 7xxx aluminum alloys. Studies have explored the effects of varying the concentrations of Zn, Mg, and Cu, and the impact of trace elements such as Zr, Ti, and rare-earth elements like Sc. Advanced processing techniques like severe plastic deformation (SPD), rapid solidification and powder metallurgy (RS/PM), and spray forming have been employed to achieve ultra-high strength. However, these methods often face limitations in terms of scalability, cost, and complexity. Recently, machine learning has emerged as a powerful tool for materials discovery and optimization, offering the potential to accelerate the development of novel alloys with desired properties. This research builds upon these previous studies, leveraging machine learning to efficiently explore the vast compositional space of 7xxx alloys and identify promising candidates for high-strength applications.

This research employed a modified Kriging model-based efficient global optimization (EGO) algorithm for composition optimization of 7xxx alloys. The process started with the creation of a training dataset comprising selected Al-Zn-Mg-Cu-(Ti)-(Y)-(Ce) alloys with known UTS values. The machine-learning model was validated using leave-one-out cross-validation. An iterative adaptive design loop was implemented, involving model construction, prediction using the "expected improvement" function to suggest the next experiments, and model refinement with newly generated data points. Zr, excluded during initial composition optimization due to its casting challenges, was added (0.2 wt.%) to the best-performing alloy after the iterative process. The optimized alloy was then processed using traditional techniques (casting, homogenization, extrusion) and subjected to systematic characterization. The search space encompassed AlxZnyMgzCuuTivYwCe, with constraints on the weight percentages of each element. The initial dataset included 38 quaternary, 25 six-component, and 10 seven-component alloys. All alloys underwent identical heat treatments before tensile testing to assess their age-hardening capacity. The machine-learning model’s performance was evaluated using diagnostic plots and standardized cross-validated residuals. The iterative process aimed to reach a targeted UTS of 550 MPa. After reaching the target UTS, the optimized alloy (Al-6.49Zn-2.52Mg-1.92Cu-0.25Zr-0.07Ti-0.29Y in wt.%) underwent detailed microstructural analysis, including XRD, optical microscopy (OM), scanning electron microscopy (SEM), and transmission electron microscopy (TEM), along with tensile testing under various heat treatment conditions.

The machine-learning-guided iterative process successfully identified an optimized Al-Zn-Mg-Cu-Zr-Ti-Y alloy with exceptional mechanical properties. The alloy, after a double-stage solution treatment and retrogression-reaging (RRA) treatment, exhibited an UTS of 896 MPa and 4.7% elongation. A further optimized heat treatment (double-stage solution treatment followed by standard T6 treatment) resulted in a remarkable UTS of 952 MPa and 6.3% elongation. This is considered a record for traditionally processed 7xxx alloys based on ingot metallurgy. Microstructural analysis revealed the presence of MgZn2, Al9Cu4Y, and Al20Ti2Y phases. The microstructure was characterized by a bimodal grain size distribution, with both coarse and fine grains, contributing to improved ductility. Three types of second phases were identified: small Al9Cu4Y particles, coarse Al20Ti2Y particles, and coarse primary Al3Zr and Al3(Zr,Ti) particles. TEM analysis revealed fine, dense matrix precipitates, discontinuous grain boundary precipitates of η-MgZn2, high-density dislocations, and ultrafine subgrains. Significantly, the Al9Cu4Y phase exhibited a novel quasi-continuous nanoscale network structure along sub-grain boundaries, a feature not previously reported in deformed microstructures. This structure, in addition to the high density of dislocations and fine subgrains, likely contributes substantially to the high strength. Comparison with previously reported ultra-high-strength 7xxx alloys showed that the optimized alloy in this study achieved a superior combination of UTS and elongation, using only traditional processing methods, even with a lower Zn content. The study highlights the significant ageing strengthening capacity of alloy 4-3, the precursor to the optimized alloy, setting the stage for its exceptionally high strength.

The findings demonstrate the effectiveness of machine learning in accelerating the discovery of high-performance aluminum alloys. The optimized alloy surpasses the strength of many commercial 7xxx alloys and those produced by more complex methods like RS/PM, highlighting the potential for machine learning-guided alloy design to produce superior materials with cost-effective processing. The exceptional strength of the optimized alloy is attributed to a combination of factors, including its effective age-hardening capacity, the presence of fine dispersoids, high dislocation density, the bimodal grain structure, and, importantly, the novel Al3Cu4Y nanoscale network structure along sub-grain boundaries. While the addition of Zr aimed to refine the microstructure and inhibit recrystallization, the formation of coarse Zr-containing particles was observed, reducing the anticipated impact of Zr on strengthening. However, the study suggests these coarse particles have a limited detrimental effect on the ultimate tensile strength, as the overall strength is remarkably high. The discovery of the Al3Cu4Y network structure represents a significant finding with potential implications for future alloy design. This new morphology may offer opportunities to further enhance the mechanical properties of 7xxx alloys through targeted compositional and processing adjustments.

This research successfully employed machine learning to accelerate the discovery of high-strength 7xxx aluminum alloys. An optimized alloy was developed exhibiting a record UTS of 952 MPa and 6.3% elongation using traditional processing techniques. The discovery of the Al3Cu4Y nanoscale network structure along sub-grain boundaries presents a novel contribution to the field. Further research could explore multi-objective optimization to improve other properties beyond strength and investigate the effects of Y compared to Sc. Detailed microstructural studies are needed to fully understand the formation and strengthening mechanisms of the Al3Cu4Y network structure. The study confirms the potential of machine learning as a powerful tool for accelerated materials discovery, offering significant advantages in both efficiency and cost-effectiveness.

While this study demonstrates the potential of machine learning in optimizing 7xxx alloy composition, some limitations should be acknowledged. The initial dataset, although relatively large, might not fully represent the vast compositional space. The focus on age-hardening capacity during composition optimization neglected the effects of homogenization treatment and plastic deformation, which are crucial aspects of industrial processing. The formation of coarse Zr-containing particles was an unintended outcome, and further optimization to mitigate this is recommended. Future studies employing more sophisticated sampling techniques and a comprehensive consideration of all processing steps are encouraged to fully realize the potential of machine learning in developing high-performance aluminum alloys.