Design of melt purification agent for recycled aluminum based on machine learning

Chengbo Li; Gangzhi Yu; Honglin Zhou; Yan Yan; Zhaowei Wang; Tonghan Yang; Chengyi Huang; Yuliao Meng; Xiaoyang Lu; Kaijie Jiang; Shuanglan Xie; Cailiu Yin; Chengbo Li; Gangzhi Yu; Honglin Zhou; Yan Yan; Zhaowei Wang; Tonghan Yang; Chengyi Huang; Yuliao Meng; Xiaoyang Lu; Kaijie Jiang; Shuanglan Xie; Cailiu Yin

doi:10.3934/matersci.2026026

AIMS Materials Science

2026, Volume 13, Issue 3: 538-559. doi: 10.3934/matersci.2026026

Previous Article Next Article

Research article Topical Sections

Design of melt purification agent for recycled aluminum based on machine learning

1.
School of Materials and Environment, Guangxi Key Laboratory of Advanced Structural Materials and Carbon Neutralization, Guangxi Minzu University, Nanning 530105, China
2.
Guangzhou Xiaopeng Motors Technology Co, Ltd. Guangzhou 510640, China
3.
ALG Aluminum lnc. Guangxi Key Laboratory of Materials and Processes of Aluminum Alloys, Nanning 530031, China

Received: 10 February 2026 Revised: 26 May 2026 Accepted: 01 June 2026 Published: 16 June 2026

In this study, we proposed a data-driven approach integrating machine learning and experimental validation for the intelligent formula design of melt purification agents for recycled aluminum. Five machine learning algorithms, including random forest regression (RFR), gradient boosting regression (GBR), extreme GBR, polynomial kernel support vector regression, and radial basis function kernel support vector regression, were comprehensively evaluated in terms of root mean square error, mean absolute error, and coefficient of determination. RFR was selected as the optimal model. Combined with Pearson correlation matrix and SHapley Additive exPlanations (SHAP) interpretability analysis, the influence mechanism of each component on the hot-cracking susceptibility and ultimate tensile strength of recycled aluminum alloy was clarified. Multi-objective optimization was carried out to minimize hot-cracking susceptibility and maximize tensile strength, and a novel melt purification agent primarily based on CeF₃ (NF-GX) was developed. The predicted values of hot-cracking susceptibility and tensile strength of NF-GX are 189.97 and 207.80 MPa, respectively, while the corresponding experimental measured values are 214.00 ± 10.39 and 191.33 ± 6.11 MPa, with relative errors of 11.23% and 8.61%. Comparative tests with three commercial purification agents show that NF-GX reduces the metal burn-off rate from 29.19 ± 1.36% to 21.30 ± 0.56%, and refines the average grain size from 191.64 ± 1.98 to 106.36 ± 0.92 μm. Microstructural characterization confirms that the synergistic effect of multi-component fluorides in NF-GX can effectively remove oxide inclusions, break continuous grain boundary liquid films, and induce heterogeneous nucleation, thereby achieving microstructure densification and grain refinement. This study demonstrates that machine learning-driven formula design can break through the limitations of the traditional empirical trial-and-error method, providing a new idea and method for the customized development of high-performance melt purification agents for recycled aluminum.
- machine learning–driven design,
- melt purification agent,
- recycled aluminum alloy,
- grain refinement,
- property improvement
Citation: Chengbo Li, Gangzhi Yu, Honglin Zhou, Yan Yan, Zhaowei Wang, Tonghan Yang, Chengyi Huang, Yuliao Meng, Xiaoyang Lu, Kaijie Jiang, Shuanglan Xie, Cailiu Yin. Design of melt purification agent for recycled aluminum based on machine learning[J]. AIMS Materials Science, 2026, 13(3): 538-559. doi: 10.3934/matersci.2026026

Related Papers:

Abstract

In this study, we proposed a data-driven approach integrating machine learning and experimental validation for the intelligent formula design of melt purification agents for recycled aluminum. Five machine learning algorithms, including random forest regression (RFR), gradient boosting regression (GBR), extreme GBR, polynomial kernel support vector regression, and radial basis function kernel support vector regression, were comprehensively evaluated in terms of root mean square error, mean absolute error, and coefficient of determination. RFR was selected as the optimal model. Combined with Pearson correlation matrix and SHapley Additive exPlanations (SHAP) interpretability analysis, the influence mechanism of each component on the hot-cracking susceptibility and ultimate tensile strength of recycled aluminum alloy was clarified. Multi-objective optimization was carried out to minimize hot-cracking susceptibility and maximize tensile strength, and a novel melt purification agent primarily based on CeF₃ (NF-GX) was developed. The predicted values of hot-cracking susceptibility and tensile strength of NF-GX are 189.97 and 207.80 MPa, respectively, while the corresponding experimental measured values are 214.00 ± 10.39 and 191.33 ± 6.11 MPa, with relative errors of 11.23% and 8.61%. Comparative tests with three commercial purification agents show that NF-GX reduces the metal burn-off rate from 29.19 ± 1.36% to 21.30 ± 0.56%, and refines the average grain size from 191.64 ± 1.98 to 106.36 ± 0.92 μm. Microstructural characterization confirms that the synergistic effect of multi-component fluorides in NF-GX can effectively remove oxide inclusions, break continuous grain boundary liquid films, and induce heterogeneous nucleation, thereby achieving microstructure densification and grain refinement. This study demonstrates that machine learning-driven formula design can break through the limitations of the traditional empirical trial-and-error method, providing a new idea and method for the customized development of high-performance melt purification agents for recycled aluminum.

References

[1]	Yang Y, Zhang H, Wu L, et al. (2024) Supply potential, carbon emission reduction, energy conservation, and sustainable pathways for aluminum recycling in China. Sustain Prod Consump 50: 239–252. https://doi.org/10.1016/j.spc.2024.07.034 doi: 10.1016/j.spc.2024.07.034
[2]	Raabe D, Ponge D, Uggowitzer PJ, et al. (2022) Making sustainable aluminum by recycling scrap: The science of "dirty" alloys. Prog Mater Sci 128: 100947. https://doi.org/10.1016/j.pmatsci.2022.100947 doi: 10.1016/j.pmatsci.2022.100947
[3]	Gaustad G, Olivetti E, Kirchain R (2012) Improving aluminum recycling: A survey of sorting and impurity removal technologies. Resour Conserv Recycl 58: 79–87. https://doi.org/10.1016/j.resconrec.2011.10.010 doi: 10.1016/j.resconrec.2011.10.010
[4]	Cai B, Kao A, Lee P, et al. (2019) Growth of β intermetallic in an Al-Cu-Si alloy during directional solidification via machine learned 4D quantification. Scr Mater 165: 29–33. https://doi.org/10.1016/j.scriptamat.2019.02.007 doi: 10.1016/j.scriptamat.2019.02.007
[5]	Du S, Zhang S, Wang J, et al. (2024) Energy flow of aerospace aluminum scraps cycle and advanced integration principles for upcycling technologies: A review. J Clean Prod 448: 141176. https://doi.org/10.1016/j.jclepro.2024.141176 doi: 10.1016/j.jclepro.2024.141176
[6]	Su Z, Jin C, Zeng Z, et al. (2024) Simultaneous improvement of mechanical and castability properties of Al-Cu-Mn based alloys by Ca/Ni micro-alloying. J Alloys Compd 996: 174824. https://doi.org/10.1016/j.jallcom.2024.174824 doi: 10.1016/j.jallcom.2024.174824
[7]	Li W, Qian F, Li J, et al. (2023) Design strategy for eliminating cracking and improving mechanical properties of Al-Mg-Si alloys fabricated by laser melting deposition. Addit Manuf 68: 103513. https://doi.org/10.1016/j.addma.2023.103513 doi: 10.1016/j.addma.2023.103513
[8]	Xu Y, Zhang Z, Gao Z, et al. (2021) Effect of Sc on the hot cracking properties of 7xxx aluminum alloy and the microstructure of squeeze castings. Materials 14: 6881. https://doi.org/10.3390/ma14226881 doi: 10.3390/ma14226881
[9]	Yang B, Friedrich S, Friedrich B (2025) Advances in non-metallic inclusion removal from aluminum melts towards cleaner and higher-performance materials. J Mater Sci 60: 12291–12314. https://doi.org/10.1007/s10853-025-11195-9 doi: 10.1007/s10853-025-11195-9
[10]	Yang B, Gao M, Liu Y, et al. (2023) Formation mechanism of refined Al6 (Mn, Fe) phase particles during continuous rheo-extrusion and its contribution to tensile properties in Al–Mg–Mn–Fe alloys. Mater Sci Eng A 872: 144952. https://doi.org/10.1016/j.msea.2023.144952 doi: 10.1016/j.msea.2023.144952
[11]	Chen X, Saada MB, Lavisse B, et al. (2025) Recent advances in the remelting process for recycling aluminium alloy chips: A critical review. Int J Mater Form 18: 42. https://doi.org/10.1007/s12289-025-01904-9 doi: 10.1007/s12289-025-01904-9
[12]	Mair P, Kaserer L, Braun J, et al. (2022) Dependence of mechanical properties and microstructure on solidification onset temperature for Al2024–CaB6 alloys processed using laser powder bed fusion. Mater Sci Eng A 833: 142552. https://doi.org/10.1016/j.msea.2021.142552 doi: 10.1016/j.msea.2021.142552
[13]	Kim MJ, Yun JP, Yang JBR, et al. (2020) Prediction of the temperature of liquid aluminum and the dissolved hydrogen content in liquid aluminum with a machine learning approach. Metals 10: 330. https://doi.org/10.3390/met10030330 doi: 10.3390/met10030330
[14]	Datta A, Rana AK, Ghadai RK (2025) Machine learning-driven nonlinear analysis of inclusion effects in aluminium alloys. Sci Rep 15: 35866. https://doi.org/10.1038/s41598-025-19756-3 doi: 10.1038/s41598-025-19756-3
[15]	Zhang G, Lu W, Wu X, et al. (2023) A new strategy on designing fluxes for aluminum alloy melt refinement. Materials 16: 2322. https://doi.org/10.3390/ma16062322 doi: 10.3390/ma16062322
[16]	Chen X, Zhong J, Lin H, et al. (2024) High precision modeling with stacked neural network to predict the mechanical property of aluminum alloy. Mater Lett 375: 137187. https://doi.org/10.1016/j.matlet.2024.137187 doi: 10.1016/j.matlet.2024.137187
[17]	Cao X, Zhang Y, Li J, et al. (2020) Composition design of 7XXX aluminum alloys optimizing stress corrosion cracking resistance using machine learning. Mater Res Express 7: 046506. https://doi.org/10.1088/2053-1591/ab8492 doi: 10.1088/2053-1591/ab8492
[18]	Tang J, Aizezi N, Liu J, et al. (2025) A new method for alloy classification based on multimodal fusion of LIBS & exp-LIPA and machine learning. Opt Laser Technol 192: 114043. https://doi.org/10.1016/j.optlastec.2025.114043 doi: 10.1016/j.optlastec.2025.114043
[19]	Wang Z, Aizezi N, Ye Y, et al. (2025) Enhanced classification of aluminum alloys via time-frequency dual-domain acoustic feature fusion with laser-induced breakdown spectroscopy. Talanta 300: 129208. https://doi.org/10.1016/j.talanta.2025.129208 doi: 10.1016/j.talanta.2025.129208
[20]	Jiang L, Fu H, Zhang Z, et al. (2024) Synchronously enhancing the strength, toughness, and stress corrosion resistance of high-end aluminum alloys via interpretable machine learning. Acta Mater 270: 119873. https//doi. org/10.1016/j. actamat. 2024.119873 doi: 10.1016/j.actamat.2024.119873
[21]	Kwak S, Kim J, Ding H, et al. (2024) Prediction improvement of compressive strength and strain of directionally solidified TiAl alloy based on training data size adjustment. J Mater Res Technol 30: 5017–5027. https://doi.org/10.1016/j.jmrt.2024.04.165 doi: 10.1016/j.jmrt.2024.04.165
[22]	Fatriansyah JF, Satrio MRR, Federico A, et al. (2024) Machine learning-based forward and inverse designs for prediction and optimization of fracture toughness of aluminum alloy. Results Eng 23: 102717. https://doi.org/10.1016/j.rineng.2024.102717 doi: 10.1016/j.rineng.2024.102717
[23]	Juan Y, Niu G, Yang Y, et al. (2024) Accelerated design of Al-Zn-Mg-Cu alloys via machine learning. Trans Nonferrous Met Soc China 34: 709–723. https://doi.org/10.1016/s1003-6326(23)66429-5 doi: 10.1016/s1003-6326(23)66429-5
[24]	Abioye SO, Babatunde YO, Abikoye OA, et al. (2025) Optimized machine learning algorithms with SHAP analysis for predicting compressive strength in high-performance concrete. AI Civ Eng 4: 16. https://doi.org/10.1007/s43503-025-00061-x doi: 10.1007/s43503-025-00061-x
[25]	Xiong X, Zhang N, Yang J, et al. (2024) Machine learning-assisted prediction of corrosion behavior of 7XXX aluminum alloys. Metals 14: 401. https://doi.org/10.3390/met14040401 doi: 10.3390/met14040401
[26]	Fu K, Zhu D, Zhang Y, et al. (2023) Predictive modeling of tensile strength in aluminum alloys via machine learning. Materials 16: 7236. https://doi.org/10.3390/ma16227236 doi: 10.3390/ma16227236
[27]	Kim BJ, Kim SH, Kayani SH, et al. (2023) Melt thermal-rate treatment for uniform solute distribution and improved mechanical properties of an Al-Zn-Mg-Cu alloy prepared by direct-chill casting. J Alloys Compd 967: 171745. https://doi.org/10.1016/j.jallcom.2023.171745 doi: 10.1016/j.jallcom.2023.171745
[28]	Duan Y, Xu G, Peng X, et al. (2015) Effect of Sc and Zr additions on grain stability and superplasticity of the simple thermal–mechanical processed Al-Zn-Mg alloy sheet. Mater Sci Eng A 648: 80–91. https://doi.org/10.1016/j.msea.2015.09.049 doi: 10.1016/j.msea.2015.09.049
[29]	Drezet JM, Mireux B, Kurtuldu G, et al. (2015) Measurement of mechanical coherency temperature and solid volume fraction in Al-Zn alloys using in situ X-ray diffraction during casting. Metall Mater Trans A 46: 4183–4190. https://doi.org/10.1007/s11661-015-3041-y doi: 10.1007/s11661-015-3041-y
[30]	Christy JV, Arunachalam R, Mourad A-HI, et al. (2020) Processing, properties, and microstructure of recycled aluminum alloy composites produced through an optimized stir and squeeze casting processes. J Manuf Process 59: 287–301. https://doi.org/10.1016/j.jmapro.2020.09.067 doi: 10.1016/j.jmapro.2020.09.067

Reader Comments

Your name:*

Email:*
© 2026 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)