Method for designing and optimizing corrosion-resistant magnesium alloy based on machine learning
By combining machine learning with Latin hypercube and orthogonal experimental design, a Kriging model was established to solve the problem of mismatch between the corrosion resistance and strength of magnesium alloys. This enabled efficient and precise design and optimization of magnesium alloy composition, resulting in high-performance magnesium alloys.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHENGDU AIRCRAFT INDUSTRY GROUP
- Filing Date
- 2026-02-10
- Publication Date
- 2026-06-02
AI Technical Summary
Existing technologies do not match the corrosion resistance and room temperature tensile strength of magnesium alloys, which limits their application in large structural components. Furthermore, traditional trial-and-error methods are time-consuming and costly, and lack models that quantitatively describe the relationship between alloy composition and performance.
Machine learning methods were employed, and experimental sample points were designed using Latin hypercube and orthogonal experimental sampling methods. A Kriging model was established, and the optimal magnesium alloy composition was iteratively sought through cross-validation and global optimization algorithms to achieve efficient and accurate design and optimization.
Simultaneous optimization of corrosion resistance and strength of magnesium alloys was achieved, resulting in a high-strength corrosion-resistant magnesium alloy with neutral salt spray corrosion resistance ≤0.3mg/cm2/day and tensile strength of 400MPa. This reduces the number of experiments and costs, and the model can be transferred to other alloy systems.
Smart Images
Figure CN122133466A_ABST
Abstract
Description
Technical Field
[0001] This invention discloses a machine learning-based method for designing and optimizing corrosion-resistant magnesium alloys, relating to the field of magnesium alloy material composition design technology. Background Technology
[0002] Magnesium alloys have shown great promise in the aerospace field, but their poor corrosion resistance has consistently hampered their further development. Researchers both domestically and internationally primarily rely on experimental trial-and-error methods for material development, resulting in low efficiency, long development cycles, and high costs. In recent years, with the continuous development of high-throughput computing and machine learning technologies, machine learning algorithms have been used for the design and calculation of high-performance magnesium alloys. However, a model that can quantitatively describe the relationship between alloy composition and performance is still lacking. Dong Zhihua of Chongqing University used machine learning algorithms to analyze the implicit structure-property relationships between magnesium alloy composition, processing technology, and mechanical properties, thereby achieving efficient design of new high-performance magnesium alloys based on mechanical performance requirements. However, the experimental data mainly came from online sources, and a dedicated database was not established. This invention provides a multi-component design and optimization method based on machine learning and oriented towards performance requirements, enabling rapid, efficient, and accurate design of magnesium alloy compositions to meet service performance needs. Summary of the Invention
[0003] With the rapid increase in the demand for lightweighting in aerospace technology, magnesium alloys have advantages due to their low density, good specific strength and stiffness, and good electromagnetic shielding performance. However, the mismatch between the corrosion resistance and room temperature tensile strength of magnesium alloys severely restricts their application in large structural components. One method for fine-tuning the overall performance control of magnesium alloys is to address this issue. Currently, traditional trial-and-error methods suffer from long cycles and high costs. This invention employs machine learning methods, based on data-driven theory, abandoning the pursuit of complex internal mechanism analysis and starting from analyzing data correlations to achieve efficient and precise design and optimization of corrosion-resistant magnesium alloys.
[0004] To achieve the above-mentioned objectives, the technical solution of the present invention is as follows: A machine learning-based method for the design and optimization of corrosion-resistant magnesium alloys includes the following steps: Step 1: Design the initial sample points for the experiment using a combination of Latin hypercube and orthogonal experimental sampling methods; Step 2: Establish an experimental dataset corresponding to the composition parameters and corrosion resistance performance using experimental data; Step 3: Establish an optimization model; Step 4: Validation of experimental data for sample points, mainly using the "cross-validation" method; Step 5: Using a global optimization algorithm, combined with the Kriging model and the expected improvement function, with the goal of efficient global optimization, we iteratively search for the optimal solution and use an improved efficient global optimization algorithm to search for the component parameters of the next iteration. Step 6: Return the input values and objective function values from the previous iteration to the initial dataset, rebuild and optimize the Kriging model, obtain the optimal solution of the current Kriging model and compare it with the required target performance. If the target is achieved, stop; otherwise, repeat the iteration multiple times.
[0005] Preferably, before step one, the method further includes determining the entire composition space based on the influence of alloying elements on the corrosion resistance of magnesium alloys, while also considering strength, and controlling the composition accuracy to 0.05% mass fraction, thus establishing an optimization space of 625,251,312 sample points.
[0006] Preferably, the entire composition space includes: Al: 1%-5%; Gd: 8%-10%; Y: 2%-4%; Zn: 1%-3%; Zr: 0.2%-0.5%; Ge: 0.05%-0.2%; Sn: 0.05%-0.2%; with the balance being Mg.
[0007] Preferably, step one includes the following steps: Step S11: First, determine the value range of each component, discretize each component into several levels, and use Latin hypercube sampling to generate initial sample points. The number of samples N must meet the minimum sample requirement of orthogonal experimental design. Step S12: Select an orthogonal array based on the number and level of component parameters, and map the parameter values of the sample points generated by Latin hypercube sampling to the levels of the orthogonal array; adjust the parameter combinations through the orthogonal array to ensure that each level of each parameter appears evenly in the sample and is evenly distributed in combination with the levels of other parameters. Step S13: Generate final sample points. Based on the Latin hypercube sampling samples, adjust the parameter combinations through an orthogonal array to ensure that each level of each parameter appears the same number of times and that the combinations between parameters are balanced, and determine the final initial sample points for the experiment.
[0008] Preferably, in step two, a Kriging surrogate model of input feature value x and output objective function performance f(x) is constructed using experimental data, where: x is a half-component parameter; f(x) is corrosion resistance performance.
[0009] Preferably, step three includes: establishing an optimization objective function, selecting optimization parameters, and determining constraints.
[0010] Preferably, after the initial input design is completed, the corresponding input feature value x and the objective function performance f(x) are obtained with shapeability as an indirect objective.
[0011] Preferably, in step four, the formula for the standard cross-validation residual is as follows: ,in This represents the experimental values of the sample points. This represents the model's predicted values. is the standard deviation of the prediction points.
[0012] Preferably, the initially established Kriging model is evaluated, and if the standard cross-validation residual value ( residual If the sample data is within the range of [-3, +3], the feasibility of the Kriging model is determined. At the same time, "outliers" in the sample data are analyzed based on the cross-validation results, and double-repeat the experiment. If the retest results are consistent, the outliers are removed; if the retest results are inconsistent, the retest data replaces the original data. Then, the Kriging model is initially optimized based on the corrected sample data.
[0013] Preferably, the "outlier" is: unconventional data where the standard cross-validation residual value deviates from the range [-3, +3].
[0014] The beneficial effects of this invention are: I. The machine learning-based design and optimization method for corrosion-resistant magnesium alloys provided in this invention achieves a neutral salt spray corrosion resistance of ≤0.3 mg / cm³. 2 / day, still has a high-strength corrosion-resistant magnesium alloy with a tensile strength of 400MPa.
[0015] Second, the machine learning-based design and optimization method for corrosion-resistant magnesium alloys provided by this invention applies machine learning methods to corrosion-resistant magnesium alloys and uses performance-driven design to forward-design the composition of corrosion-resistant magnesium alloys, saving time.
[0016] Third, the machine learning-based design and optimization method for corrosion-resistant magnesium alloys provided by this invention reduces the number of experiments through intelligent sampling, saving time and materials and reducing experimental costs.
[0017] IV. The machine learning-based design and optimization method for corrosion-resistant magnesium alloys provided by this invention simultaneously optimizes corrosion resistance, strength, etc., to achieve multi-objective collaborative optimization. At the same time, the model can be transferred to the design of other alloy systems (such as aluminum / titanium alloys). Attached Figure Description
[0018] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation
[0019] The present invention will be further described in detail below with reference to embodiments, but the implementation of the present invention is not limited thereto.
[0020] Example 1 like Figure 1As shown, the machine learning-based design and optimization method for corrosion-resistant magnesium alloys includes the following steps: Step 1: Design the initial sample points for the experiment using a combination of Latin hypercube and orthogonal experimental sampling methods; Step 2: Establish an experimental dataset corresponding to the composition parameters and corrosion resistance performance using experimental data; Step 3: Establish an optimization model; Step 4: Validation of experimental data for sample points, mainly using the "cross-validation" method; Step 5: Using a global optimization algorithm, combined with the Kriging model and the expected improvement function, with the goal of efficient global optimization, we iteratively search for the optimal solution and use an improved efficient global optimization algorithm to search for the component parameters of the next iteration. Step 6: Return the input values and objective function values from the previous iteration to the initial dataset, rebuild and optimize the Kriging model, obtain the optimal solution of the current Kriging model and compare it with the required target performance. If the target is achieved, stop; otherwise, repeat the iteration multiple times.
[0021] Example 2 like Figure 1 As shown, the machine learning-based design and optimization method for corrosion-resistant magnesium alloys includes the following steps: Step 1: Design the initial sample points for the experiment using a combination of Latin hypercube and orthogonal experimental sampling methods; Step 2: Establish an experimental dataset corresponding to the composition parameters and corrosion resistance performance using experimental data; Step 3: Establish an optimization model; Step 4: Validation of experimental data for sample points, mainly using the "cross-validation" method; Step 5: Using a global optimization algorithm, combined with the Kriging model and the expected improvement function, with the goal of efficient global optimization, we iteratively search for the optimal solution and use an improved efficient global optimization algorithm to search for the component parameters of the next iteration. Step 6: Return the input values and objective function values from the previous iteration to the initial dataset, rebuild and optimize the Kriging model, obtain the optimal solution of the current Kriging model and compare it with the required target performance. If the target is achieved, stop; otherwise, repeat the iteration multiple times.
[0022] Before step one, the process also includes determining the composition space based on the influence of alloying elements on the corrosion resistance of magnesium alloys, while also considering strength. The composition space includes Al, Gd, Y, Zn, Zr, Ge, Sn, and Mg. The composition accuracy is controlled at 0.05% mass fraction, and an optimization space of 625,251,312 sample points is established.
[0023] The entire composition space includes: Al: 1%-5%; Gd: 8%-10%; Y: 2%-4%; Zn: 1%-3%; Zr: 0.2%-0.5%; Ge: 0.05%-0.2%; Sn: 0.05%-0.2%; with the balance being Mg.
[0024] Step one includes the following steps: Step S11: First, determine the value range of each component, discretize each component into several levels, and use Latin hypercube sampling to generate initial sample points. The number of samples N must meet the minimum sample requirement of orthogonal experimental design. Step S12: Select an orthogonal array based on the number and level of component parameters, and map the parameter values of the sample points generated by Latin hypercube sampling to the levels of the orthogonal array; adjust the parameter combinations through the orthogonal array to ensure that each level of each parameter appears evenly in the sample and is evenly distributed in combination with the levels of other parameters. Step S13: Generate final sample points. Based on the Latin hypercube sampling samples, adjust the parameter combinations through an orthogonal array to ensure that each level of each parameter appears the same number of times and that the combinations between parameters are balanced, and determine the final initial sample points for the experiment.
[0025] In step two, a Kriging surrogate model is constructed using experimental data, which consists of input feature value x and output objective function performance f(x), where x is a half-component parameter and f(x) is the corrosion resistance performance.
[0026] Step three includes: establishing an optimization objective function, selecting optimization parameters, and determining constraints (including composition range constraints, process constraints on castability, and performance constraints on corrosion resistance and strength).
[0027] After the initial input design is completed, the corresponding input feature value x and the performance of the objective function f(x) are obtained with shapeability as an indirect objective.
[0028] In step four, the formula for the standard cross-validation residuals is as follows: ,in This represents the experimental values of the sample points. This represents the model's predicted values. is the standard deviation of the prediction points.
[0029] Among them, the initially established Kriging model is evaluated, and if the standard cross-validation residual value ( residualIf the sample data is within the range of [-3, +3], the feasibility of the Kriging model is determined. At the same time, "outliers" in the sample data are analyzed based on the cross-validation results, and double-repeat the experiment. If the retest results are consistent, the outliers are removed; if the retest results are inconsistent, the retest data replaces the original data. Then, the Kriging model is initially optimized based on the corrected sample data.
[0030] The “outliers” are: unconventional data where the standard cross-validation residual values deviate from the range [-3, +3].
[0031] Example 3 Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, this invention provides a machine learning-based design and optimization technique for corrosion-resistant magnesium alloys, comprising the following steps: Step A: Based on the influence of alloying elements on the corrosion resistance of magnesium alloys, and taking into account strength, the entire composition space was determined to include 8 elements: Al (1%-5%), Gd (8%-10%), Y (2%-4%), Zn (1%-3%), Zr (0.2%-0.5%), Ge (0.05%-0.2%), Sn (0.05%-0.2%), and Mg (balance). The composition accuracy was controlled at 0.05% mass fraction, and an optimization space of 625,251,312 sample points was established.
[0032] Step B: One hundred sets of magnesium alloy compositions were designed using a combination of Latin hypercube and orthogonal experimental sampling methods. Melting experiments were conducted, and the neutral salt spray corrosion resistance was tested after casting. An experimental dataset corresponding to the composition parameters and corrosion resistance was established using the experimental data. A Kriging surrogate model was constructed using the experimental data, with input feature values x (semi-composition parameters) and output objective function performance f(x) (corrosion resistance).
[0033] Step C: Validate the model using experimental data from sample points, primarily employing cross-validation to cross-validate the initially established Kriging model and determine its feasibility. Simultaneously, based on the cross-validation results, promptly remove "bad" data points and perform preliminary model optimization.
[0034] Step D: Use the Global Optimization Algorithm (EGO) to search for the optimal composition parameters of the corrosion-resistant magnesium alloy, and then perform melting and casting to test the neutral salt spray corrosion performance. Obtain experimental values of corrosion resistance under the preliminary optimized composition parameters and compare them with the predicted values of the Kriging model. If the deviation between the experimental value and the predicted value is too large (>10%), the point needs to be added back to the training set, the Kriging model needs to be updated, and iterative optimization needs to be continued. Otherwise, proceed to the next step.
[0035] Step E: Return the input values and objective function values from the previous iteration to the initial dataset, reconstruct and optimize the Kriging model, obtain the optimal solution of the current Kriging model, and compare it with the required target performance. After 6 rounds of experiments, a neutral salt spray corrosion performance of ≤0.3mg / cm² was obtained. 2 / day, still has a high-strength corrosion-resistant magnesium alloy with a tensile strength of 400MPa.
[0036] It is understood that the present invention has been described through some embodiments, and those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.
Claims
1. A machine learning-based method for the design and optimization of corrosion-resistant magnesium alloys, characterized in that, Includes the following steps: Step 1: Design the initial sample points for the experiment using a combination of Latin hypercube and orthogonal experimental sampling methods; Step 2: Establish an experimental dataset corresponding to the composition parameters and corrosion resistance performance using experimental data; Step 3: Establish an optimization model; Step 4: Validation of experimental data for sample points, mainly using the "cross-validation" method; Step 5: Using a global optimization algorithm, combined with the Kriging model and the expected improvement function, with the goal of efficient global optimization, we iteratively search for the optimal solution and use an improved efficient global optimization algorithm to search for the component parameters of the next iteration. Step 6: Return the input values and objective function values from the previous iteration to the initial dataset, rebuild and optimize the Kriging model, obtain the optimal solution of the current Kriging model and compare it with the required target performance. If the target is achieved, stop; otherwise, repeat the iteration multiple times.
2. The machine learning-based design and optimization method for corrosion-resistant magnesium alloys according to claim 1, characterized in that: Before step one, the process also includes determining the entire composition space based on the influence of alloying elements on the corrosion resistance of magnesium alloys, while also considering strength. This space includes Al, Gd, Y, Zn, Zr, Ge, Sn, and Mg, with the composition accuracy controlled at 0.05% mass fraction. An optimization space of 625,251,312 sample points was established.
3. The machine learning-based design and optimization method for corrosion-resistant magnesium alloys according to claim 2, characterized in that: The entire composition space includes: Al: 1%-5%; Gd: 8%-10%; Y: 2%-4%; Zn: 1%-3%; Zr: 0.2%-0.5%; Ge: 0.05%-0.2%; Sn: 0.05%-0.2%; balance Mg.
4. The machine learning-based design and optimization method for corrosion-resistant magnesium alloys according to claim 3, characterized in that: Step one includes the following steps: Step S11: First, determine the value range of each component, discretize each component into several levels, and use Latin hypercube sampling to generate initial sample points. The number of samples N must meet the minimum sample requirement of orthogonal experimental design. Step S12: Select an orthogonal array based on the number and level of component parameters, and map the parameter values of the sample points generated by Latin hypercube sampling to the levels of the orthogonal array; adjust the parameter combinations through the orthogonal array to ensure that each level of each parameter appears evenly in the sample and is evenly distributed in combination with the levels of other parameters. Step S13: Generate final sample points. Based on the Latin hypercube sampling samples, adjust the parameter combinations through an orthogonal array to ensure that each level of each parameter appears the same number of times and that the combinations between parameters are balanced, and determine the final initial sample points for the experiment.
5. The machine learning-based design and optimization method for corrosion-resistant magnesium alloys according to claim 4, characterized in that: In step two, a Kriging surrogate model of input feature value x and output objective function performance f(x) is constructed using experimental data, where x is a half-component parameter and f(x) is the corrosion resistance performance.
6. The machine learning-based design and optimization method for corrosion-resistant magnesium alloys according to claim 5, characterized in that: Step three includes: establishing the objective function, selecting optimization parameters, and determining constraints.
7. The machine learning-based design and optimization method for corrosion-resistant magnesium alloys according to claim 6, characterized in that: After the initial input design is completed, the corresponding input feature value x and the performance of the objective function f(x) are obtained with shapeability as an indirect objective.
8. The machine learning-based design and optimization method for corrosion-resistant magnesium alloys according to claim 7, characterized in that: In step four, the formula for the standard cross-validation residuals is as follows: ,in This represents the experimental values of the sample points. This represents the model's predicted values. is the standard deviation of the prediction points.
9. The machine learning-based design and optimization method for corrosion-resistant magnesium alloys according to claim 8, characterized in that: The initially established Kriging model was evaluated. If the standard cross-validation residual value ( residual If the data is in the range [-3, +3], then the feasibility of the Kriging model is determined. At the same time, based on the cross-validation results, "outliers" in the sample data are analyzed and double-repeated. If the revalidation results are consistent, the outliers are removed. If the retest results are inconsistent, the retest data will replace the original data; then the Kriging model will be initially optimized based on the corrected sample data.
10. The machine learning-based design and optimization method for corrosion-resistant magnesium alloys according to claim 9, characterized in that: The "outliers" are: unconventional data where the standard cross-validation residual values deviate from the range [-3, +3].