High-strength and high-toughness as-cast al-si-mg-ti-sr-sc alloy based on machine learning multi-objective optimization and its preparation process and application

Abstract

The application discloses a kind of high-strength high-toughness Al-Si-Mg-Ti-Sr-Sc as-cast alloy based on machine learning multi-objective optimization and its preparation process and application, belong to aluminum alloy material preparation technical field.The application is by constructing high-quality as-cast Al-Si-Mg alloy composition-property database, completes element screening using Pearson correlation coefficient, compares four kinds of integrated learning algorithms of CatBoost, GBR, RFR, XGBoost to establish high-precision performance prediction model, realizes tensile strength and elongation double-target global optimization using NSGA-II genetic algorithm, reveals element synergistic mechanism in combination with SHAP and PDP interpretability analysis, and finally obtains optimal alloy by vacuum medium-frequency induction melting preparation.The application deeply fuses data-driven design and experimental preparation, shortens research and development cycle to 1-2 months, reduces cost by more than 60%, process is stable, can be industrialized, and is suitable for the field of severe requirement of lightweight automobile, aerospace, precision machinery and other strength and toughness synergies.

Description

Technical Field

[0001] This invention relates to the field of aluminum alloy material preparation technology, specifically to a high-strength and high-toughness Al-Si-Mg-Ti-Sr-Sc as-cast alloy based on machine learning multi-objective optimization, its preparation process and application, which is particularly suitable for fields such as automotive lightweighting, aerospace structural parts, and precision mechanical components where stringent requirements for strength and plasticity are required. Background Technology

[0002] Against the backdrop of rapid upgrading in the high-end equipment manufacturing industry, lightweighting has become a core development direction in the automotive, aerospace, and precision machinery sectors. Low-density, high-specific-strength, and high-reliability metal structural materials have become the core focus of industry research and development. Al-Si-Mg series cast aluminum alloys, with their low density, excellent casting fluidity, high specific strength, good corrosion resistance, and weldability, have become the most widely used lightweight cast aluminum alloy system in the industrial field. They are widely used in key components such as chassis structural parts for new energy vehicles, motor housings, engine blocks, aerospace vehicle cabins, and precision connecting parts for rail transit.

[0003] With the continuous increase in the driving range requirements of new energy vehicles, the constant upgrading of the load capacity of aerospace equipment, and the development of precision machinery towards miniaturization and high reliability, unprecedentedly stringent requirements have been placed on the synergistic matching of strength and toughness in as-cast aluminum alloys. However, traditional commercial Al-Si-Mg as-cast alloys (such as A356 and ZL101) can no longer meet the performance requirements of high-end application scenarios. Their tensile strength in the conventional as-cast state is only 150-200MPa, elongation is 5-8%, and the strength-ductility product is generally less than 1600MPa·%. There is an inherent contradiction in the matching of strength and ductility, where an increase in strength leads to a sharp drop in ductility, and an improvement in ductility leads to insufficient strength. This has become a key bottleneck restricting the lightweight development of high-end equipment.

[0004] Currently, the research and modification of Al-Si-Mg alloys still face numerous technical challenges. At the composition design level, traditional alloy development heavily relies on trial-and-error based on researchers' experience. Multi-element microalloying design lacks systematic theoretical guidance, requiring extensive repetitive melting, characterization, and performance testing experiments. This results in a development cycle of 6-12 months, high costs, low efficiency, and significant batch-to-batch performance fluctuations, making precise control of alloy composition difficult. Existing microalloying research largely focuses on single-element modification, such as adding Sr alone to achieve eutectic Si modification, adding Ti alone to achieve grain refinement, and adding Sc alone to achieve dispersion strengthening. However, research on the synergistic mechanism of multiple elements (Ti, Sr, Sc) is severely lacking, failing to achieve multi-element coupling synergy and hindering breakthroughs in the strength-ductility matching performance ceiling of alloys.

[0005] At the level of intelligent R&D, the current application of machine learning in aluminum alloy design is mostly superficial and has significant technical shortcomings. On the one hand, the datasets used in existing studies vary in quality, with problems such as high data noise, high redundancy, insufficient standardization, and mixing of as-cast and heat-treated data. This results in poor generalization ability of the constructed models and low tensile strength prediction coefficient R. 2 Generally below 0.9, elongation prediction R 2 Below 0.85, it cannot meet the requirements for precise design of alloy composition; on the other hand, existing machine learning models are mostly "black box models" that lack effective interpretable analysis methods, cannot quantify the contribution of each alloy element to performance, and cannot reveal the synergistic mechanism between elements. As a result, the model optimization results lack theoretical support, making it difficult to effectively guide actual experimental preparation and unable to achieve deep integration of data-driven and materials research and development.

[0006] Therefore, developing a data-driven, efficient, accurate, interpretable, and multi-objective synergistic optimization method for the design and preparation of Al-Si-Mg as-cast alloys, clarifying the synergistic strengthening and toughening mechanism of multi-element microalloying, and achieving a simultaneous and significant improvement in the tensile strength and elongation of the alloy, shortening the R&D cycle and reducing R&D costs, has become a key technical problem that urgently needs to be solved in this field. It has important engineering significance and application value for promoting the rapid development of lightweight high-end equipment. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a high-strength and high-toughness Al-Si-Mg-Ti-Sr-Sc as-cast alloy based on machine learning multi-objective optimization, as well as its preparation process and application, to solve technical problems such as blind design of traditional alloy composition, poor strength-ductility matching, low R&D efficiency, uninterpretable models, and unclear microalloying mechanisms.

[0008] This invention achieves precise alloy composition design and performance breakthroughs through a complete process including database construction, feature selection, high-precision model training, multi-objective genetic algorithm optimization, SHAP / PDP interpretability analysis, vacuum melting, and experimental verification. The prepared alloys exhibit simultaneous improvements in strength and plasticity, and the process is highly efficient and stable, making it suitable for industrial production.

[0009] To achieve the above objectives, the present invention adopts the following technical solution:

[0010] A high-strength and high-toughness Al-Si-Mg-Ti-Sr-Sc as-cast alloy based on machine learning multi-objective optimization, comprising, by mass fraction, 6.5-7.5% Si, 0.3-0.6% Mg, 0-0.2% Ti, 0-0.05% Sr, and 0-0.65% Sc, with the balance being Al and unavoidable impurities.

[0011] Preferably, the high-strength and high-toughness Al-Si-Mg-Ti-Sr-Sc as-cast alloy based on machine learning multi-objective optimization comprises, by mass fraction, 7% Si, 0.45% Mg, 0.13% Ti, 0.01% Sr, and 0.59% Sc, with the balance being Al and unavoidable impurities.

[0012] Preferably, the as-cast alloy has a tensile strength of 270.7 MPa, an elongation of 9.5%, and a yield strength of 115 MPa.

[0013] This invention also provides a preparation process for high-strength and high-toughness Al-Si-Mg-Ti-Sr-Sc as-cast alloys based on machine learning multi-objective optimization, comprising the following steps:

[0014] S1. Database construction: The composition-property data of as-cast Al-Si-Mg alloys in the literature were collected, and after cleaning and standardization, a composition-property database containing 122 samples was constructed.

[0015] S2. Element screening: The Pearson correlation coefficient method was used to perform feature correlation analysis, eliminate redundant elements, and determine Si, Mg, Ti, Sr, and Sc as the core elements;

[0016] S3. Prediction Model Construction: By comparing CatBoost, GBR, RFR, and XGBoost algorithms, and through Bayesian optimization and ten-fold cross-validation, high-precision prediction models for UTS and EL are established.

[0017] S4. Multi-objective component optimization: With the goal of maximizing UTS and EL, the NSGA-II algorithm is used for global optimization to obtain the Pareto front and screen the optimal components;

[0018] S5. Model interpretability analysis: SHAP and PDP methods are used to quantify element contributions and reveal the synergistic mechanism;

[0019] S6. Vacuum melting preparation: According to the optimal composition, the alloy ingot is obtained by vacuum medium frequency induction melting, vacuuming, gas washing, staged heating, pouring and cooling.

[0020] S7. Performance and microstructure characterization: Room temperature tensile testing, XRD and EPMA characterization were performed to verify the alloy properties and microstructure.

[0021] Furthermore, in step S3, the CatBoost algorithm is selected to construct the model, and the UTS model R... 2 =0.96, RMSE=7.03; EL model R 2 =0.95, RMSE=0.89.

[0022] Furthermore, the NSGA-II parameters in step S4 are: population 200, offspring 200, evolution 150 generations, and random seed 42.

[0023] Furthermore, the smelting process in step S6 is as follows:

[0024] (1) Load the weighed raw materials into the graphite crucible in sequence and close the furnace door;

[0025] (2) Evacuate for 10 minutes, then introduce argon gas to wash the gas, evacuate again for 10 minutes, and finally introduce argon gas until the pressure inside the furnace is 0.05 MPa;

[0026] (3) Heating with electricity, using 225A current for 4 min, 245A current for 3 min, 255A current for 2 min (start shaking the crucible in the middle of the month), 270A current for 1 min, and 275A current for 1 min in sequence;

[0027] (4) Adjust the current to 245A and heat for 1 minute. Stop shaking the crucible in the middle of the minute and immediately pour the molten metal into the graphite mold.

[0028] (5) After casting, vacuum the furnace again and cool it to room temperature before taking out the alloy ingot.

[0029] Furthermore, in step S6, an additional 0.05 wt.% of Mg is added to compensate for burn-off.

[0030] Furthermore, in step S7, the stretching rate is 0.5 mm / min, the XRD scanning range is 20°-90°, and EPMA is used for micro-area morphology and second phase analysis.

[0031] This invention also provides an application of a high-strength, high-toughness Al-Si-Mg-Ti-Sr-Sc as-cast alloy, used as a structural component material in fields including automotive lightweighting, aerospace, and precision machinery.

[0032] Technical principle of the invention:

[0033] (I) The core role of each alloy raw material

[0034] This invention uses Al as the matrix, Si and Mg as core strengthening elements, and Ti, Sr, and Sc as microalloying control elements. The mass fraction and core function of each element are as follows:

[0035] 1. Al element: As the alloy matrix, the face-centered cubic crystal structure provides the alloy with the basis for low density, high plasticity, excellent casting fluidity and corrosion resistance. It is the core carrier of the alloy's comprehensive performance. The balance is Al and unavoidable impurity elements. The impurity content must be strictly controlled to avoid damage to the alloy's performance.

[0036] 2. Si element: The mass fraction is controlled at 6.5-7.5%, with an optimal value of 7%, which is close to the Al-Si binary eutectic composition. On the one hand, it can significantly improve the casting fluidity of the alloy melt, reduce the tendency of castings to hot crack, and ensure the formability of complex thin-walled structural parts. On the other hand, it can form a eutectic Si phase with Al and combine with Mg to form the Mg2Si strengthening phase, which is one of the core sources of alloy strength. This content range avoids the damage to plasticity caused by the precipitation of coarse needle-like eutectic Si due to excessive Si content, and also solves the problem of insufficient casting performance and strength caused by excessive Si content, thus achieving a balance between casting performance and toughness.

[0037] 3. Mg element: The mass fraction is controlled between 0.3-0.6%, with an optimal value of 0.45%. It is the core precipitation strengthening element of the alloy. It can combine with Si to form a nano-sized Mg2Si precipitate phase, achieving a significant precipitation strengthening effect by pinning dislocations, greatly improving the alloy strength, while having minimal negative impact on the alloy's casting performance. This content range avoids the problem of Mg2Si phase coarsening and significant decrease in plasticity caused by excessive Mg content, and also solves the problem of insufficient strengthening effect caused by excessive Mg content, achieving an optimal balance between strengthening effect and plasticity.

[0038] 4. Ti element: The mass fraction should be controlled between 0-0.2%, with an optimal value of 0.13%. It is a highly efficient grain refining element. During alloy solidification, it can form an Al3Ti phase with extremely low lattice mismatch with the α-Al matrix, serving as a heterogeneous nucleation core and significantly refining the α-Al matrix grains. This grain refinement strengthens the alloy while simultaneously improving its strength and plasticity. This content range avoids the formation of coarse, needle-like Al3Ti phases due to excessive Ti content, which could become crack sources and impair alloy properties. It also ensures optimal grain refinement, maximizing the benefits of grain refinement strengthening.

[0039] 5. Sr element: The mass fraction is controlled between 0-0.05%, with an optimal value of 0.01%. It is a long-lasting and highly efficient eutectic Si modifying element. Even a trace amount can transform the coarse, acicular eutectic Si phase in the as-cast state into fine, uniform fibrous or granular phases, completely eliminating the cutting effect of acicular eutectic Si on the Al matrix, significantly reducing stress concentration during stress loading, and significantly improving the alloy's plasticity and toughness. This trace addition range avoids over-modification and Sr segregation damage caused by excessive Sr content, while achieving the optimal eutectic Si modification effect, providing a core guarantee for improving the alloy's plasticity.

[0040] 6. Sc element: The mass fraction is controlled between 0-0.65%, with an optimal value of 0.59%. It is a core microalloying strengthening element. On the one hand, it can combine with Al and Si to form a thermally stable, coherent nanoscale AlSc2Si2 ternary dispersed phase, effectively pinning dislocations and grain boundaries, significantly improving alloy strength through dispersion strengthening without significantly impairing plasticity. On the other hand, it can synergistically form a composite heterogeneous nucleation core with Ti, further refining α-Al grains and enhancing the fine-grain strengthening effect. This content range avoids the coarsening of the dispersed phase, the decrease in strengthening effect, and the significant increase in cost caused by excessive Sc content, achieving an optimal balance between dispersion strengthening effect and economy.

[0041] (II) The synergistic coupling mechanism of multiple elements

[0042] This invention achieves multi-element coupling synergy through the precise ratio of ternary microalloying elements Ti, Sr, and Sc, forming a triple synergistic strengthening and toughening system of "fine grain strengthening + modification toughening + dispersion strengthening," which completely solves the inherent contradiction of strength-ductility matching in traditional alloys. The core synergistic mechanism is as follows:

[0043] 1. Ti-Sc synergistic grain refinement strengthens and simultaneously improves strength and toughness.

[0044] Both the Al3Ti phase formed by Ti and the Al3Sc phase formed by Sc have an L12 structure and extremely low lattice mismatch with the α-Al matrix. Their combination can form a highly efficient heterogeneous nucleation core, with a nucleation efficiency far exceeding that of single-element addition. This can reduce the α-Al grain size by more than 40% compared to adding Ti alone. The refined grains not only enhance alloy strength through grain boundary strengthening but also significantly improve alloy plasticity by increasing the number of grain boundaries to hinder crack propagation. This achieves a simultaneous improvement in both strength and plasticity, solving the problem of limited grain refinement effects from single-element additions.

[0045] 2. Sr-Sc synergistic strong plasticity matching solves the problem of strong and tough mutual exclusion.

[0046] Sr transforms acicular eutectic Si into fine granular Si through modification, eliminating the matrix fragmentation effect and significantly improving the alloy's plasticity, providing ample plasticity margin for the dispersion strengthening of Sc. Meanwhile, the nanoscale AlSc2Si2 dispersion phase formed by Sc effectively pins dislocations and hinders grain boundary migration, significantly improving the alloy's strength. Simultaneously, it inhibits the coarsening of the eutectic Si phase during solidification, further consolidating the modification effect of Sr. The synergistic effect of these two processes achieves a positive cycle where "plasticity improvement provides space for strength enhancement, and strength enhancement does not sacrifice plasticity," resolving the inherent contradiction in traditional alloys where "strength improvement inevitably leads to decreased plasticity, and vice versa."

[0047] 3. Ti-Sr-Sc ternary coupling achieves a significant leap in toughness.

[0048] The refined α-Al matrix grains of Ti significantly increase the grain boundary area, providing more dispersed precipitation sites for the Sr-modified eutectic Si phase, further refining the eutectic Si phase and enhancing the modification and toughening effect. Simultaneously, the uniformly refined matrix grains and the dispersed eutectic Si phase provide more pinning sites for the nano-AlSc2Si2 dispersed phase formed by Sc, significantly improving the uniformity and effectiveness of dispersion strengthening. The coupling of these three factors forms a closed-loop synergistic strengthening and toughening system, achieving a modification effect greater than the sum of its parts (1+1+1>3), enabling the alloy to achieve a leapfrog improvement in strength and toughness in the as-cast state, breaking through the performance boundaries of traditional Al-Si-Mg alloys.

[0049] (III) Necessity and Importance of Selecting Preparation Process Parameters

[0050] All parameters of the preparation process in this invention have been precisely designed and optimized. Each parameter setting has a clear necessity and core role, which can ensure accurate alloy composition, uniform microstructure, and stable performance. The design principle of the core process parameters is as follows:

[0051] 1. Database construction and feature selection parameter design

[0052] This invention constructs a composition-performance database of as-cast Al-Si-Mg alloys containing 122 samples. All samples are derived from as-cast alloy data in publicly available literature. After cleaning and standardization to remove outliers and non-as-cast data, the homogeneity and high quality of the dataset are ensured, providing a reliable data foundation for model training and avoiding the problems of high noise and poor generalization ability in existing research datasets. The Pearson correlation coefficient method is used for feature selection to remove redundant elements with low correlation to performance, and Si, Mg, Ti, Sr, and Sc are identified as core elements. This effectively reduces model dimensionality and the risk of overfitting, and significantly improves the model's prediction accuracy and computational efficiency.

[0053] 2. Parameter Design for Machine Learning Model Construction

[0054] This invention compares four mainstream ensemble learning algorithms—CatBoost, GBR, RFR, and XGBoost—and ultimately selects the CatBoost algorithm, which offers stronger adaptability to feature interactions and better resistance to overfitting. Model training is completed using Bayesian optimization and 10-fold cross-validation. The 10-fold cross-validation effectively verifies the model's generalization ability and avoids overfitting on small datasets; Bayesian optimization enables global optimization of model hyperparameters, further improving prediction accuracy. The final obtained UTS model R... 2 =0.96, RMSE=7.03, EL model R 2 =0.95, RMSE=0.89, the prediction accuracy far exceeds that of existing similar models, and can achieve accurate prediction of alloy properties.

[0055] 3. Multi-objective optimization parameter design

[0056] This invention employs the NSGA-II multi-objective genetic algorithm to perform global optimization with the goal of maximizing tensile strength (UTS) and elongation (EL). The algorithm is configured with a population size of 200, offspring size of 200, evolution over 150 generations, and a random seed of 42. The population size and offspring size of 200 ensure population diversity during the optimization process, avoiding getting trapped in local optima. Evolution over 150 generations ensures sufficient algorithm convergence, obtaining a complete Pareto optimal front. The fixed random seed of 42 ensures the repeatability of the optimization results. This parameter setting balances global optimization capability with computational efficiency, enabling the selection of the alloy composition with optimal strength-ductility matching from two conflicting performance objectives, thus solving the core problem that traditional single-objective optimization cannot simultaneously address strength and toughness.

[0057] 4. Vacuum melting process parameter design

[0058] Vacuum and argon protection process: First, evacuate for 10 minutes, then purge with argon gas, followed by another 10 minutes of vacuum evacuation, and finally introduce argon gas until the furnace pressure reaches 0.05 MPa. This process can completely remove oxygen and moisture from the furnace, preventing the oxidation and burn-off of reactive elements such as Mg and Sc during the melting process, ensuring the accuracy of the alloy composition, and preventing oxide inclusions from entering the melt, thus improving the purity and performance stability of the alloy.

[0059] The graded heating melting process employs a sequential heating regime of 225A / 4min, 245A / 3min, 255A / 2min, 270A / 1min, and 275A / 1min. During the 255A heating stage, the crucible is agitated simultaneously. Finally, the temperature is adjusted to 245A and held for 1min before immediate pouring. Graded heating allows for gradual melting of the raw materials, avoiding the loss of low-melting-point elements and the unmelted state of high-melting-point elements caused by rapid heating, thus ensuring uniform melt composition. Agitating the crucible during heating further promotes homogenization of the melt composition, preventing elemental agglomeration. The final 245A holding time during pouring ensures the melt is at the optimal pouring temperature and fluidity, avoiding defects such as coarse grains due to excessively high pouring temperatures or incomplete pouring and poor forming due to excessively low temperatures.

[0060] Mg element burn-off compensation: Add an additional 0.05wt.% of Mg element. Since Mg is a highly reactive element with high vapor pressure, it is prone to volatilization and burn-off during vacuum melting. The fixed addition amount can accurately control the actual content of Mg in the melt, ensure the precipitation amount and strengthening effect of Mg2Si strengthening phase, and avoid batch-to-batch fluctuations in composition and performance.

[0061] 5. Design of performance and tissue characterization parameters

[0062] The tensile testing rate is set at 0.5 mm / min, strictly complying with GB / T228.1 standard for room temperature tensile testing of metallic materials, ensuring the accuracy, repeatability, and industry comparability of tensile performance data; the XRD testing range is set at 20°-90°, which can completely cover the characteristic diffraction peaks of the alloy matrix phase and the second phase, accurately analyzing the phase composition of the alloy; the EPMA test is used for micro-area morphology and elemental distribution analysis, which can accurately characterize the morphology, size, and elemental distribution of the second phase, and verify the elemental synergistic mechanism and toughening principle.

[0063] Compared with the prior art, the technical advantages of this invention are:

[0064] 1. Data-driven precision ingredient design enables a leap in R&D efficiency.

[0065] This invention breaks through the traditional model of alloy R&D relying on trial and error based on experience. It constructs a high-quality, standardized composition-performance database of as-cast Al-Si-Mg alloys. Through feature screening, high-precision model training, and multi-objective global optimization, it achieves forward and precise design of alloy composition. Compared with the traditional R&D model, it significantly shortens the alloy R&D cycle from 6-12 months to 1-2 months, reduces R&D costs by more than 60%, avoids performance fluctuations caused by trial and error based on experience, and improves batch performance stability by more than 80%. It provides a new, efficient, and replicable technological paradigm for the R&D of new aluminum alloy materials.

[0066] 2. A breakthrough in the synergistic effect of high strength and high toughness, far exceeding the performance boundaries of traditional alloys.

[0067] This invention resolves the inherent contradiction of the long-standing trade-off between strength and toughness in Al-Si-Mg alloys through multi-element microalloying synergistic regulation. The prepared optimal composition as-cast alloy achieves excellent comprehensive performance with a tensile strength of 270.7 MPa, a yield strength of 115 MPa, and an elongation of 9.5% without subsequent solution treatment or aging heat treatment. Compared with traditional commercially available A356 as-cast alloys, the tensile strength is increased by more than 35%, the elongation by nearly 50%, and the strength-ductility product by more than 60%, breaking through the performance ceiling of traditional Al-Si-Mg as-cast alloys and meeting the stringent requirements of high-end equipment for structural materials with a balance of strength and toughness.

[0068] 3. High-precision, interpretable machine learning models solve the industry's black box problem.

[0069] This invention compares four mainstream ensemble learning algorithms and ultimately selects the CatBoost algorithm to construct a performance prediction model. After Bayesian optimization and 10-fold cross-validation, the tensile strength UTS model R... 2 The elongation rate of the EL model reached 0.96, RMSE = 7.03, and the elongation was 7.03. 2With an accuracy of 0.95 and RMSE of 0.89, the prediction accuracy far exceeds that of existing similar research models, demonstrating strong generalization ability and reliability. Furthermore, this invention innovatively introduces SHAP and PDP interpretability analysis methods to quantify the contribution of each alloying element to performance, clarify the interaction rules between elements, and overcome the industry pain points of traditional machine learning models being "black box-like" and lacking theoretical support for optimization results, achieving a deep integration of model prediction and theoretical mechanisms.

[0070] 4. A clear multi-element synergistic toughening mechanism enables precise and controllable performance.

[0071] This invention systematically reveals for the first time the coupled synergistic strengthening and toughening mechanism of ternary microalloying elements Ti, Sr, and Sc in Al-Si-Mg alloys. It clarifies the core roles of Sc-dominated dispersion strengthening to enhance strength, Sr-dominated eutectic Si modification to improve plasticity, and Ti-dominated grain refinement for synergistic toughening. These three elements are coupled to form a triple synergistic strengthening and toughening system of "grain refinement strengthening + modification toughening + dispersion strengthening." Compared with existing single-element modification technologies, this invention achieves a synergistic effect of 1+1+1>3, solving the problems of unclear mechanisms, unstable modification effects, and low performance control precision in traditional microalloying technologies.

[0072] 5. The process is stable and highly adaptable, possessing value for large-scale industrialization and promotion.

[0073] The preparation process of this invention utilizes industry-standard vacuum medium-frequency induction melting equipment, eliminating the need for additional specialized production equipment. Through staged heating melting, vacuum argon protection, and precise composition compensation, it achieves precise control of alloy composition and stable regulation of microstructure and properties. No complex subsequent heat treatment processes are required, resulting in excellent overall performance in the as-cast state. This simplified process is highly controllable, yields a high output, and can be directly adapted to existing industrial casting production lines, possessing strong scalability and applicability in various high-end manufacturing fields such as automotive lightweighting, aerospace, and precision machinery. Attached Figure Description

[0074] Figure 1 The design framework diagram of Al-Si-Mg-Ti-Sr-Sc alloys demonstrates the entire process of "building a high-quality database → screening alloy elements → building a predictive model → multi-objective global optimization → experimental verification and characterization".

[0075] Figure 2 : Correlation analysis flowchart, showing the steps for screening alloying elements based on Pearson correlation coefficient;

[0076] Figure 3 A comparison chart of predicted tensile strengths on test sets using different ensemble learning algorithms. Figure 3 (a) Figure 3(b) Figure 3 (c) Figure 3 (d) R values ​​for CatBoost, GBR, RFR, and XGBoost in tensile strength prediction, respectively. 2 And RMSE value graph;

[0077] Figure 4 A comparison chart of the predicted elongation rates of different ensemble learning algorithms on the test set, where... Figure 4 (a) Figure 4 (b) Figure 4 (c) Figure 4 (d) R values ​​for CatBoost, GBR, RFR, and XGBoost in elongation prediction, respectively. 2 And RMSE value graph;

[0078] Figure 5 : Multi-objective optimization and mechanical property curves of alloys, where, Figure 5 (a) is a graph of Pareto front solutions for multi-objective optimization using a genetic algorithm. Figure 5 (b) Engineering stress-strain curves for optimizing the design of the alloy;

[0079] Figure 6 SHAP overview diagram, in which, Figure 6 (a) is a ranking chart of the importance of features in the tensile strength prediction model. Figure 6 (b) is a ranking chart of feature importance for the elongation prediction model. Figure 6 (c) A graph showing the contribution percentage of features in the tensile strength prediction model. Figure 6 (d) is a graph showing the contribution percentage of the features in the elongation prediction model;

[0080] Figure 7 A three-dimensional partial dependency plot (PDP) shows the synergistic regulation of tensile strength by Ti, Sr, and Sc; among them, Figure 7 (a) is a partial dependency graph of Ti, Sr, and Sc. Figure 7 (b) is a partial dependency plot of Ti and Sr under the condition Sc=0.55. Figure 7 (c) is a partial dependency graph of Ti and Sc under the condition Sr=0.005. Figure 7 (d) is a partial dependency graph of Sr and Sc under the condition Ti=0.095;

[0081] Figure 8 A three-dimensional partial dependency plot (PDP) shows the synergistic regulatory mechanism of elongation by Ti, Sr, and Sc; among them, Figure 8 (a) is a partial dependency graph of Ti, Sr, and Sc. Figure 8 (b) is a partial dependency plot of Ti and Sr under the condition Sc=0.55. Figure 8 (c) is a partial dependency graph of Ti and Sc under the condition Sr=0.005. Figure 8 (d) is a partial dependency graph of Sr and Sc under the condition Ti=0.095;

[0082] Figure 9 XRD pattern of as-cast Al-7Si-0.45Mg-0.13Ti-0.01Sr-0.59Sc alloy;

[0083] Figure 10 EPMA backscattered image of an as-cast Al-7Si-0.45Mg-0.13Ti-0.01Sr-0.59Sc alloy, in which... Figure 10 (a) is a low-magnification morphology image at 100 μm. Figure 10 (b) is a high-magnification morphology image at 10 μm. Detailed Implementation

[0084] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments, so as to better understand the technical solution of the present invention.

[0085] Unless otherwise defined, the technical terms used in this invention have their conventional meaning in the art; the experimental materials and equipment are all commercially available. The raw materials used are Al (99.999%), Al-25Si master alloy (99.99%), Mg (99.999%), Al-5Ti (99.99%), Al-10Sr (99.99%), and Al-10Sc master alloy (99.99%). The melting equipment is a vacuum medium-frequency induction melting furnace, the tensile testing equipment is an LD26.105 universal testing machine, and the microstructure is characterized using XRD and EPMA equipment.

[0086] An alloy design framework for high-strength and high-toughness Al-Si-Mg-Ti-Sr-Sc as-cast alloys based on machine learning multi-objective optimization is as follows: Figure 1 As shown.

[0087] Example 1: Database Construction

[0088] The system collected composition-property experimental data of as-cast Al-Si-Mg alloys published in authoritative domestic and international academic literature, with data sources including ScienceDirect, CNKI, and Web of Science. The collected data underwent rigorous cleaning and standardization: duplicate data, data with unclear experimental conditions, and samples with abnormal performance were removed; the unit of composition was standardized to mass percentage (wt.%), and the standard for mechanical property testing was standardized to room temperature tensile testing. Finally, a composition-property database of as-cast Al-Si-Mg alloys containing 122 valid samples was constructed, covering 10 alloying elements (Al, Si, Mg, Zn, Cu, Mn, Fe, Ti, Sr, and Sc) and their corresponding tensile strength (UTS) and elongation (EL) data.

[0089] Example 2: Screening of Alloying Elements

[0090] The Pearson correlation coefficient method was used to analyze the correlation of 10 alloying elements in the database, calculating the Pearson correlation coefficient ρ between any two elements. When |ρ|>0.95, the two elements were considered to have a strong linear correlation, carrying almost the same information. For each pair of strongly correlated elements, a feature set containing that element was constructed, and a model was trained using UTS and EL as outputs. The prediction performance of the models was compared, and elements with little impact on the target performance were removed.

[0091] The analysis results show that the Pearson correlation coefficient between Al and Si is -0.95, indicating a very strong negative correlation. Considering that Al, as a matrix element, has a relatively small impact on model construction, while Si is more representative in terms of feature importance, we chose to remove Al element features and retain Si element features, such as... Figure 2 As shown in the figure. Ultimately, Si, Mg, Ti, Sr, and Sc were selected as the core characteristic elements, while the contents of Zn, Cu, Mn, and Fe were fixed at 0.

[0092] Based on the selected core elements and the statistical distribution characteristics of the database, an alloy composition design space was constructed: Si 6.5-7.5 wt.%, Mg 0.3-0.6 wt.%, Ti 0-0.2 wt.%, Sr 0-0.05 wt.%, and Sc 0-0.65 wt.%. This composition design space was meshed, with the mesh step size set as follows: Si 0.1 wt.%, Mg and Sc 0.05 wt.%, and Ti and Sr 0.005 wt.%, generating a total of 486,178 virtual composition points, which served as a candidate library for subsequent multi-objective optimization.

[0093] Example 3: Construction of Prediction Model

[0094] The predictive performance of four ensemble learning algorithms—CatBoost, GBR, RFR, and XGBoost—was compared. The initial dataset was divided into training and test sets in a 9:1 ratio. The hyperparameters of each algorithm were tuned using Bayesian optimization combined with 10-fold cross-validation, with the goal of minimizing the root mean square error (RMSE).

[0095] Optimization results show that the CatBoost algorithm has the best prediction performance: its UTS prediction model has the highest coefficient of determination R on the test set. 2 =0.96, RMSE=7.03; EL prediction model test set determination coefficient R 2 =0.95, RMSE=0.89, significantly better than the other three algorithms, such as Figure 3 and Figure 4 As shown. Therefore, the CatBoost algorithm was ultimately chosen to construct the UTS and EL prediction models, providing high-precision model support for subsequent multi-objective component optimization.

[0096] Example 4: Multi-objective component optimization

[0097] Based on the CatBoost prediction model constructed above, a non-dominated sorting genetic algorithm (NSGA-II) is used to carry out multi-objective global optimization. The optimization objective is to simultaneously maximize tensile strength (UTS) and elongation (EL), and the design variables are the mass fractions of Si, Mg, Ti, Sr, and Sc, with the variable values ​​ranging from the constructed composition design space.

[0098] The parameters of the NSGA-II algorithm were set as follows: population size 200, number of offspring 200, number of generations 150, crossover probability 0.9, mutation probability 0.1, and a fixed random seed of 42 to ensure reproducibility. After optimization, all non-dominated solutions were extracted to form the Pareto front. The Pareto solutions were ranked using a comprehensive scoring method, and the sum of the UTS and EL predictions for each solution was calculated. The solution with the highest score was selected as the optimal candidate alloy composition.

[0099] The final optimal candidate alloy composition is: Si 7.0 wt.%, Mg 0.45 wt.%, Ti 0.13 wt.%, Sr 0.01 wt.%, Sc 0.59 wt.%, with the balance being Al. The predicted performance of this composition is: UTS = 267.0 MPa, EL = 9.4%. Figure 5 As shown in (a).

[0100] Example 5: SHAP and PDP Interpretability Analysis

[0101] After determining the optimal alloy composition, SHAP interpretability analysis and PDP global interaction analysis were conducted on the CatBoost performance prediction model to analyze the intrinsic mechanism of composition-performance, as follows:

[0102] The contribution of each alloying element to the prediction results was calculated using the SHAP method, and the characteristics were ranked according to their importance based on their mean absolute SHAP values, such as... Figure 6 As shown in the figure, the results indicate that Sc contributes the most to tensile strength (UTS), with high Sc content corresponding to a significant positive SHAP value, achieving strength and toughness through the formation of nano-reinforcing phases; Sr contributes the most to elongation (EL), with trace amounts of Sr refining the eutectic Si and significantly improving plasticity; Ti exhibits a non-monotonic effect on both UTS and EL, initially increasing and then decreasing, with an optimal addition range; Mg makes a continuous positive contribution to UTS and has a peak range for EL; Fe is a harmful element, with high content corresponding to a negative SHAP value, severely impairing both strength and plasticity. SHAP analysis clearly quantifies the direction of action and critical threshold of single elements, providing a basis for precise composition control.

[0103] The three-dimensional global response surface of Ti-Sr-Sc was plotted using a partial PDP method, revealing the multi-element coupling characteristics, such as... Figure 7 and Figure 8 As shown in the figure. The results indicate that the alloy's UTS and EL exhibit a nonlinear response to changes in elemental content. The high-strength range corresponds to medium-high Sc + medium Ti + moderate Sr, while the high-ductility range corresponds to medium Ti + low-to-medium Sr + medium-low Sc. Significant interaction effects exist among the three, with the synergistic effect far superior to single-element regulation. PDP analysis, from a global perspective, determines the optimal composition window for simultaneous strength and ductility, verifying that the optimal composition of this invention lies within the performance peak region, providing intuitive theoretical support for the alloy's strengthening and toughening mechanism.

[0104] Example 6: Melting Preparation

[0105] 1. Raw Material Preparation: Pure Al with a purity ≥99.999%, Al-25Si master alloy with a purity ≥99.99%, pure Mg, Al-5Ti master alloy, Al-10Sr master alloy, and Al-10Sc master alloy were used as raw materials. The mass of each raw material was calculated according to the above optimal composition, with an additional 0.05 wt.% of Mg element weighed to compensate for burn-off during the smelting process. The blocky master alloys were broken into small pieces for easy weighing and melting.

[0106] 2. Smelting process: Smelting is carried out using a vacuum medium-frequency induction furnace. The specific steps are as follows: (1) The weighed raw materials are loaded into the graphite crucible in sequence and the furnace door is closed; (2) Vacuum is drawn for 10 minutes, then argon gas is introduced for cleaning, vacuum is drawn again for 10 minutes, and finally argon gas is introduced until the pressure inside the furnace is 0.05 MPa; (3) Electric heating is applied, and the following heating methods are used in sequence: 225A current for 4 minutes, 245A current for 3 minutes, 255A current for 2 minutes (the crucible is shaken from the middle of the month), 270A current for 1 minute, and 275A current for 1 minute; (4) The current is adjusted to 245A for 1 minute, the crucible is stopped from shaking in the middle of the month, and the molten metal is immediately poured into the graphite mold; (5) After casting, vacuum is drawn again and cooled to room temperature, and the alloy ingot is taken out. The mold is a cylinder with a radius of 10 mm and a height of 95 mm.

[0107] Example 7 Performance Testing and Characterization

[0108] 1. Mechanical Property Testing: The alloy ingots were machined into standard tensile specimens and subjected to room temperature tensile tests using an LD26.105 universal testing machine. The tensile rate was 0.5 mm / min. Three parallel specimens were tested for each component, and the average value was taken as the final performance result. Figure 5 As shown in (b), the room temperature tensile test results show that the tensile strength of the alloy is 270.7 MPa, the yield strength is 115 MPa, and the total elongation at break is 9.5%, which is in high agreement with the model prediction. The overall performance is significantly better than that of existing conventional Al-Si-Mg alloys.

[0109] 2. Microstructure characterization: (1) X-ray diffraction (XRD) analysis: The alloy sample was analyzed by X-ray diffractometer. The diffraction angle scanning range was 20°-90° and the scanning speed was 1° / min. The XRD pattern of the alloy is shown in the figure. Figure 9 As shown, its main phases are α-Al matrix phase and elemental Si phase. No second phases related to Sc, Mg, Ti, and Sr were detected, indicating that these elements mainly exist in the matrix in the form of solid solution or nano-precipitated phases; (2) Electron probe microanalysis (EPMA): The alloy sample was analyzed for micro-area composition and morphology using an electron probe microanalysis instrument to observe the microstructure characteristics of the alloy and determine the composition of the second phase. EPMA backscattered images are shown below. Figure 10 As shown, the alloy matrix is ​​an α-Al solid solution with a large number of bright white high atomic number second phases dispersed within it. The overall microstructure is uniform and there are no obvious casting defects. Spot scan analysis confirmed that the bright white second phase is an AlSc2Si2 ternary phase, rather than a binary Al3Sc phase. Ti refines the α-Al matrix grains by forming the Al3Ti phase, Sr refines the eutectic Si phase through modification, and Sc forms the AlSc2Si2 ternary dispersed phase. The synergistic effect of these three phases achieves a simultaneous improvement in the alloy's strength and ductility.

[0110] Summarize:

[0111] This invention discloses a high-strength and high-toughness Al-Si-Mg-Ti-Sr-Sc as-cast alloy based on machine learning multi-objective optimization, its preparation process and application, which solves the technical pain points of traditional aluminum alloy composition design, such as reliance on experience trial and error, long R&D cycle, high cost, poor strength-ductility matching, uninterpretable machine learning model, and unclear mechanism of action of microalloying elements.

[0112] This invention adopts a data-driven, multi-objective optimization, and experimental verification approach. First, a high-quality Al-Si-Mg alloy composition-performance database is constructed, and feature selection is performed using the Pearson correlation coefficient. Four ensemble learning algorithms are compared, and CatBoost is selected to establish a high-precision UTS and EL prediction model. The NSGA-II genetic algorithm is used to achieve global optimization of tensile strength and elongation, obtaining the Pareto optimal composition. Then, the alloy is precisely prepared by vacuum medium-frequency induction melting, and the microstructure and performance are verified by XRD, EPMA, etc.

[0113] The prepared Al-7Si-0.45Mg-0.13Ti-0.01Sr-0.59Sc as-cast alloy achieves excellent strength-ductility matching with a tensile strength of 270.7 MPa, a yield strength of 115 MPa, and an elongation of 9.5%, breaking through the boundaries of traditional alloys in terms of comprehensive performance. Using SHAP and PDP interpretability analysis, it was clarified that Sc dominates the strength enhancement, Sr dominates the ductility improvement, and there is an optimal range for Ti addition. Furthermore, the strengthening and toughening mechanism of Ti-Sr-Sc synergistic grain refinement, modification of eutectic Si, and formation of dispersed strengthening phases was revealed.

[0114] The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., without departing from the spirit and principle of the present invention should be included within the protection scope of the present invention.

Claims

1. A high-strength, high-toughness Al-Si-Mg-Ti-Sr-Sc as-cast alloy based on machine learning multi-objective optimization, characterized in that: Based on a mass fraction of 100%, it includes 6.5-7.5% Si, 0.3-0.6% Mg, 0-0.2% Ti, 0-0.05% Sr, and 0-0.65% Sc, with the balance being Al and unavoidable impurities.

2. The high-strength, high-toughness Al-Si-Mg-Ti-Sr-Sc as-cast alloy based on machine learning multi-objective optimization according to claim 1, characterized in that: Based on a mass fraction of 100%, it includes 7% Si, 0.45% Mg, 0.13% Ti, 0.01% Sr, and 0.59% Sc, with the balance being Al and unavoidable impurities.

3. The high-strength, high-toughness Al-Si-Mg-Ti-Sr-Sc as-cast alloy based on machine learning multi-objective optimization according to claim 2, characterized in that: The as-cast alloy has a tensile strength of 270.7 MPa, an elongation of 9.5%, and a yield strength of 115 MPa.

4. A preparation process for a high-strength, high-toughness Al-Si-Mg-Ti-Sr-Sc as-cast alloy based on machine learning multi-objective optimization according to any one of claims 1-3, characterized in that, Includes the following steps: S1. Database construction: The composition-property data of as-cast Al-Si-Mg alloys in the literature were collected, and after cleaning and standardization, a composition-property database containing 122 samples was constructed. S2. Element screening: The Pearson correlation coefficient method was used to perform feature correlation analysis, eliminate redundant elements, and determine Si, Mg, Ti, Sr, and Sc as the core elements; S3. Prediction Model Construction: By comparing CatBoost, GBR, RFR, and XGBoost algorithms, and through Bayesian optimization and ten-fold cross-validation, high-precision prediction models for UTS and EL are established. S4. Multi-objective component optimization: With the goal of maximizing UTS and EL, the NSGA-II algorithm is used for global optimization to obtain the Pareto front and screen the optimal components; S5. Model interpretability analysis: SHAP and PDP methods are used to quantify element contributions and reveal the synergistic mechanism; S6. Vacuum melting preparation: According to the optimal composition, the alloy ingot is obtained by vacuum medium frequency induction melting, vacuuming, gas washing, staged heating, pouring and cooling. S7. Performance and microstructure characterization: Room temperature tensile testing, XRD and EPMA characterization were performed to verify the alloy properties and microstructure.

5. The preparation process of high-strength and high-toughness Al-Si-Mg-Ti-Sr-Sc as-cast alloy based on machine learning multi-objective optimization according to claim 4, characterized in that: In step S3, the CatBoost algorithm is used to construct the model, and the UTS model R... 2 =0.96, RMSE=7.03; EL model R 2 =0.95, RMSE=0.

89.

6. The preparation process of high-strength and high-toughness Al-Si-Mg-Ti-Sr-Sc as-cast alloy based on machine learning multi-objective optimization according to claim 4, characterized in that: In step S4, the NSGA-II parameters are: population 200, offspring 200, evolution 150 generations, and random seed 42.

7. The preparation process of high-strength and high-toughness Al-Si-Mg-Ti-Sr-Sc as-cast alloy based on machine learning multi-objective optimization according to claim 4, characterized in that: The smelting process in step S6 is as follows: (1) Load the weighed raw materials into the graphite crucible in sequence and close the furnace door; (2) Evacuate for 10 minutes, then introduce argon gas to wash the gas, evacuate again for 10 minutes, and finally introduce argon gas until the pressure inside the furnace is 0.05 MPa; (3) Heat with electricity, using 225A current for 4 min, 245A current for 3 min, and 255A current for 2 min in sequence. In the middle of the month, shake the crucible, heat with 270A current for 1 min, and 275A current for 1 min. (4) Adjust the current to 245A and heat for 1 minute. Stop shaking the crucible in the middle of the minute and immediately pour the molten metal into the graphite mold. (5) After casting, vacuum the furnace again and cool it to room temperature before taking out the alloy ingot.

8. The preparation process of high-strength and high-toughness Al-Si-Mg-Ti-Sr-Sc as-cast alloy based on machine learning multi-objective optimization according to claim 4, characterized in that: In step S6, an additional 0.05 wt.% of Mg is added to compensate for burn-off.

9. The preparation process of high-strength and high-toughness Al-Si-Mg-Ti-Sr-Sc as-cast alloy based on machine learning multi-objective optimization according to claim 4, characterized in that: In step S7, the stretching rate is 0.5 mm / min, the XRD scanning range is 20°-90°, and EPMA is used for micro-area morphology and second phase analysis.

10. An application of a high-strength, high-toughness Al-Si-Mg-Ti-Sr-Sc as-cast alloy prepared by the preparation process according to any one of claims 4-9, characterized in that: It is used as a structural material in fields including automotive lightweighting, aerospace, and precision machinery.

Images