A closed-loop reverse design method for corrosion resistance of magnesium alloy based on electrochemical mechanism-data fusion driving

Abstract

The application discloses a kind of based on electrochemical mechanism-data fusion driven magnesium alloy corrosion resistance closed loop reverse design method, belong to magnesium alloy material intelligent design and corrosion protection technical field.The method includes: collecting magnesium alloy corrosion data and constructing standardization database;Based on the mixed potential theory calculation second phase and α-Mg matrix electrochemical potential difference mechanism parameters such as physical prior characteristics;Build "composition / process→microstructure→electrochemical parameters→corrosion rate" electrochemical mechanism-data fusion chain explainable prediction model;Using SHAP method, the feature importance of corrosion rate prediction result is decomposed, and the key influence factor is identified;With target corrosion rate as constraint, candidate alloy formula that meets corrosion resistance requirement is searched in composition space using multi-objective optimization algorithm;Through accelerated corrosion test verification and feedback update model, form mechanism modeling-data prediction-reverse design-experiment verification-feedback update closed loop iteration.The application enhances the physical explainability and small sample prediction accuracy of model through the deep fusion of electrochemical mechanism and data driving, realizes the closed loop reverse design from target corrosion rate to alloy composition, can greatly shorten the development cycle of corrosion resistant magnesium alloy and reduce experimental cost.

Description

Technical Field

[0001] This invention belongs to the field of intelligent design and corrosion protection technology for magnesium alloy materials. Specifically, it relates to a closed-loop reverse design method for the corrosion resistance performance of magnesium alloys driven by electrochemical mechanism-data fusion. In particular, it is a method that introduces electrochemical mechanism parameters such as the potential difference between the second phase and the matrix as physical priors, constructs a chain-like interpretable predictive model of "composition / process → microstructure → electrochemical parameters → corrosion rate", and realizes the reverse design of corrosion-resistant alloy composition and experimental feedback closed-loop iteration with the target corrosion rate as a constraint. Background Technology

[0002] Lightweight structural upgrades are a core development trend in high-end equipment manufacturing. Magnesium alloys, with their unique lightweight structural characteristics, have become key candidate materials to replace traditional steel and aluminum materials and achieve weight reduction and efficiency improvement in components. However, the magnesium matrix itself has extremely high electrochemical activity, and the passivation film lacks density and stability. In complex service environments such as chlorine-containing media and alternating wet and dry conditions, it is prone to triggering continuous electrochemical corrosion, leading to structural component failure and reduced service reliability. This greatly limits the large-scale application of magnesium alloys in automotive structural components, aerospace load-bearing components, rail equipment, and precision electronic devices. How to accurately predict the corrosion behavior of magnesium alloys and efficiently iterate and optimize corrosion-resistant alloy systems is a key challenge that urgently needs to be overcome in the research and development of lightweight alloy corrosion protection and intelligent materials.

[0003] The macroscopic corrosion behavior of magnesium alloys depends on the electrochemical coupling between the microstructure and the corrosive medium. The potential difference between the second phase and the α-Mg matrix is ​​the core inducing factor for microgalvanic corrosion. Simultaneously, it is nonlinearly regulated by multiple factors such as alloy composition, heat treatment process, and corrosion conditions, exhibiting typical multi-scale, strongly coupled characteristics. Traditional R&D relies on single-variable debugging and experimental trial and error, making it difficult to establish a quantitative correlation between composition, process, microstructure, electrochemical response, and macroscopic corrosion resistance. Conventional corrosion evaluation methods suffer from long physical testing cycles, high costs, and low screening efficiency. Empirical and semi-empirical models cannot characterize the multi-factor coupled corrosion laws, lacking sufficient predictive accuracy and versatility, thus failing to meet the needs of rapid iterative development of high-performance corrosion-resistant magnesium alloys.

[0004] While machine learning methods have effectively improved the efficiency of corrosion performance prediction, existing purely data-driven models have significant technical limitations. These models only construct a black-box mapping based on composition and process data, failing to incorporate core electrochemical mechanism features such as potential difference and exchange current density. They lack physical priors, resulting in poor predictive interpretability, weak generalization ability with small samples, and a lack of intermediate modeling steps for microstructure and electrochemical parameters, making it impossible to achieve full-chain traceability of corrosion mechanisms. Furthermore, current technologies only support forward performance prediction, lacking the ability to reverse-engineer composition and processes based on target corrosion resistance indicators. The models also cannot rely on experimental data for closed-loop iterative updates, leading to a disconnect between mechanistic knowledge and data modeling. Currently, there is no integrated R&D solution in the industry that deeply integrates mechanisms and data, possesses multi-scale chain prediction capabilities, reverse optimization of corrosion-resistant components, and experimental closed-loop iteration, representing a key technological gap in this field. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a closed-loop reverse design method for the corrosion resistance of magnesium alloys based on electrochemical mechanism-data fusion. This method uses electrochemical mechanism parameters as physical priors throughout the modeling process, constructs a multi-scale chain-like interpretable predictive model, uses SHAP analysis to identify key influencing factors, achieves reverse design of corrosion-resistant alloy composition with target corrosion rate as constraint, and forms a closed-loop iteration through accelerated corrosion experiments. This shortens the R&D cycle of corrosion-resistant magnesium alloys and reduces experimental costs.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A closed-loop reverse design method for the corrosion resistance of magnesium alloys based on electrochemical mechanism-data fusion is described in the following flowchart: Figure 1 As shown, it includes the following steps: S1. Multi-source corrosion data acquisition and standardized database construction Corrosion data of magnesium alloys were collected from literature, patent publications, and laboratory data. The data included at least: alloy composition (content of major and trace elements), heat treatment state (solution temperature, solution time, aging temperature, and aging time), corrosive medium (NaCl concentration, pH value, and temperature), corrosion rate (mm / year), corrosion potential (V vs. SCE), and corrosion current density (μA / cm²). The collected data were cleaned, units were standardized, missing values ​​were handled, and outliers were removed to construct a structured corrosion database.

[0007] S2. Extraction of second-phase information and calculation of electrochemical mechanism parameters For each sample in the database, based on the alloy composition and heat treatment state, the type and volume fraction of the second phase were determined using thermodynamic calculation software or literature data. The electrochemical potential difference ΔE between each second phase and the α-Mg matrix, as well as electrochemical mechanism parameters such as the area ratio of the second phase to the matrix and the exchange current density, were calculated based on the mixing potential theory. These electrochemical parameters were then combined with the original composition and process characteristics as physical a priori features to form an enhanced feature vector. The method for determining the potential difference between the second phase and the matrix and its characterization process are as follows: Figure 2 As shown.

[0008] S3. Construct a chain-interpretable predictive model that integrates electrochemical mechanisms and data. A chain-like modeling strategy is adopted to construct a multi-scale prediction model of "composition / process → microstructure → electrochemical parameters → corrosion rate". The first-level model takes composition and process as input and the second phase type and volume fraction as output. The second-level model takes the second phase information as input and electrochemical parameters such as potential difference and corrosion current density as output. The third-level model takes the electrochemical parameters and corrosion medium conditions as input and corrosion rate as output. Each level of the model can be implemented using algorithms such as random forest, XGBoost or deep neural network.

[0009] S4. Feature Importance Analysis and Mechanism Attribution Based on SHAP The SHAP method is used to decompose the feature importance of the trained corrosion rate prediction model, quantitatively evaluate the contribution and direction of each alloying element, process parameter and electrochemical mechanism parameter to the corrosion rate, identify the key factors affecting corrosion resistance, and achieve physical interpretability of the prediction results.

[0010] S5. Closed-loop reverse design of corrosion-resistant components constrained by target corrosion rate. Using the target corrosion rate as a constraint and material cost or mechanical properties as optimization objectives, a multi-objective optimization algorithm is employed to search for candidate alloy formulations that meet corrosion resistance requirements in the alloy composition space. The optimization variables are the content ranges of each alloy element, and the constraints include an upper limit for corrosion rate, a lower limit for mechanical properties, and composition range constraints. The Pareto front obtained through inverse design is shown below. Figure 3 As shown.

[0011] S6. Accelerated corrosion experiment verification and closed-loop iterative update Short-cycle accelerated corrosion experiments were designed to verify the alloy formulation recommended by reverse design. The corrosion rate and electrochemical parameters measured in the experiment were fed back to the database to incrementally update or retrain the prediction model, forming a closed-loop iteration of mechanism modeling, data prediction, reverse design, experimental verification, and feedback update.

[0012] Preferably, the corrosive medium in step S1 is a 3.5 wt.% NaCl solution with a pH of 7-12 and a temperature of 25-60°C; the corrosion rate is determined by weight loss or electrochemical methods.

[0013] Preferably, the electrochemical potential difference ΔE in step S2 is calculated based on the mixed potential theory and satisfies equation (1): Δ E = E second - E matrix (1) in, E second This represents the self-corrosion potential of the second phase. E matrix ΔE represents the self-corrosion potential of the α-Mg matrix; when ΔE>0, the second phase acts as the cathode, accelerating matrix corrosion; when ΔE<0, the second phase acts as the anode, preferentially dissolving.

[0014] Preferably, in step S2, the second phase type includes at least Mg. 17 Al 12 , Al8Mn5, Mg2Ca, MgZn2, Mg2Si, Mg 12 One or more of Zn2.

[0015] Preferably, the third-level model of the multi-scale chain prediction model in step S3 adopts the XGBoost algorithm, with input features including electrochemical parameters, corrosion medium conditions and alloy composition features, and output as corrosion rate.

[0016] Preferably, in step S4, the SHAP method calculates the Shapley value of each feature on the corrosion rate prediction result, and displays the influence direction and magnitude ranking of all features through a summary diagram; the key factors include at least the potential difference between the second phase and the matrix, Al content, Zn content, and aging temperature.

[0017] Preferably, the reverse design in step S5 adopts the NSGA-II multi-objective optimization algorithm, with the optimization objectives being to minimize the corrosion rate and maximize the tensile strength, and the constraints being the range of alloy element content and the upper limit of total cost.

[0018] Preferably, the accelerated corrosion experiment in step S6 adopts a salt spray test or a potentiodynamic polarization test, and the cycle of a single experiment is 24~72 h.

[0019] Compared with existing technologies, this invention achieves a deep integration of electrochemical mechanism knowledge and data-driven models by using electrochemical mechanism parameters such as the potential difference between the second phase and the matrix and the exchange current density as physical priors throughout the entire modeling process. This enhances the model's ability to characterize the microgalvanic corrosion mechanism of magnesium alloys and improves the prediction accuracy and physical interpretability under small sample conditions. By constructing a multi-scale chain-like interpretable prediction model of "composition / process → microstructure → electrochemical parameters → corrosion rate", the entire chain of corrosion mechanism is traceable, overcoming the shortcomings of existing black-box models that cannot explain the physical causes of prediction results. Through SHA The P-method decomposes the corrosion rate prediction results into feature importance, which can quantitatively identify key elements and process parameters affecting corrosion resistance, providing clear physical guidance for alloy composition optimization. Through a reverse design method constrained by the target corrosion rate, it directly outputs candidate alloy formulations that meet corrosion resistance requirements, realizing a leap from performance evaluation to on-demand design, significantly shortening the R&D cycle compared to traditional trial-and-error methods. By accelerating corrosion experiments and model iteration and updating mechanisms, a closed loop of "mechanism modeling - data prediction - reverse design - experimental verification - feedback update" is formed, enabling the model to continuously improve prediction accuracy and applicability as experimental data accumulates. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the overall process of the method of the present invention.

[0021] Figure 2 This is a schematic diagram illustrating the calculation and characterization of electrochemical mechanism parameters.

[0022] Figure 3 This is a bi-target Pareto front diagram obtained by reverse engineering in Example 1.

[0023] Figure 4 The image shows the three-target Pareto front obtained by reverse engineering in Example 2. Detailed Implementation

[0024] The present invention will be further described below with reference to the accompanying drawings and embodiments, but the present invention is not limited to the following embodiments.

[0025] Example 1

[0026] This embodiment uses Mg-Al-Zn series (AZ series) magnesium alloys as the object to verify the effectiveness of the method of the present invention in terms of corrosion performance prediction accuracy, physical interpretability and closed-loop reverse design capability.

[0027] (1) Multi-source corrosion data acquisition and standardized database construction: A total of 200 sets of corrosion data for AZ-based magnesium alloys were collected from publicly published literature and historical laboratory data. Compositional characteristics included: Al content 0.5–11.5 wt.%, Zn content 0.1–4.5 wt.%, and Mn content 0.1–1.2 wt.%; heat treatment conditions included: solution temperature 350–430 ℃, solution time 2–24 h, aging temperature 150–250 ℃, and aging time 4–48 h; the corrosive medium was 3.5 wt.% NaCl solution with a pH of 7–12 and a temperature of 25–50 ℃; performance indicators included: corrosion rate (0.05–5.0 mm / year), corrosion potential (-1.8–-1.2 V vs. SCE), and corrosion current density (1–500 μA / cm²). After data cleaning, 186 sets of valid data were retained and randomly divided into a training set of 149 sets and a test set of 37 sets at an 8:2 ratio.

[0028] (2) Extraction of second-phase information and calculation of electrochemical mechanism parameters: For each sample in the database, based on alloy composition and heat treatment state, the type of the second phase (mainly Mg) was determined using JMatPro thermodynamic calculation software. 17 Al 12 And Al8Mn5) and their volume fractions. The electrochemical potential difference ΔE between each second phase and the α-Mg matrix was calculated based on the mixed potential theory: Mg 17 Al 12 The potential difference between the second phase and the substrate is approximately +0.30 V (the second phase is the cathode), and the potential difference between Al8Mn5 and the substrate is approximately +0.55 V (the second phase is the cathode). The area ratio and exchange current density of the second phase are also calculated. These electrochemical parameters are used as prior physical features and concatenated with the original composition and process features to form an enhanced eigenvector. The method for determining the potential difference between the second phase and the substrate and its characterization process are as follows: Figure 2 As shown.

[0029] (3) Construct a chain-interpretable predictive model that integrates electrochemical mechanisms and data: The chain prediction model in this embodiment includes a three-level prediction path. The first-level model uses the random forest algorithm, taking composition and process parameters as inputs and the second phase type and volume fraction as outputs. The second-level model uses the XGBoost algorithm, taking second phase information as inputs and the potential difference and corrosion current density as outputs. The third-level model uses the XGBoost algorithm, taking electrochemical parameters, corrosion medium conditions, and alloy composition characteristics as inputs and the corrosion rate as output. The three levels of models are connected in series to form a complete chain prediction path.

[0030] (4) Feature importance analysis and mechanism attribution based on SHAP: The SHAP method was used to decompose the feature importance of the third-stage corrosion rate prediction model, and the contribution and direction of each feature to the corrosion rate were quantitatively evaluated. The results showed that the top five features with the greatest impact on the corrosion rate of AZ-based magnesium alloys were, in descending order: the potential difference ΔE between the second phase and the matrix, Al content, aging temperature, Zn content, and pH value of the corrosive medium.

[0031] (5) Closed-loop reverse design of corrosion-resistant components constrained by target corrosion rate: With a target corrosion rate <0.5 mm / year as the constraint, and minimizing the corrosion rate and maximizing the tensile strength as the dual optimization objectives, the NSGA-II algorithm was used to search for the Pareto optimal solution set in the composition space. The optimization variables were Al content (3–11 wt.%), Zn content (0.5–4 wt.%), and aging temperature (160–240 ℃), with constraints of corrosion rate ≤0.5 mm / year and tensile strength ≥250 MPa. The population size was 100, the number of generations was 200, the crossover probability was 0.9, and the mutation probability was 0.1. The Pareto front obtained through reverse design is shown below. Figure 3 As shown.

[0032] (6) Accelerated corrosion experiment verification and closed-loop iterative update: Three representative alloy formulations were selected from the Pareto optimal solution set, smelted and heat-treated according to the recommended composition, and their corrosion performance was verified by salt spray accelerated testing (according to GB / T 10125 standard, 5 wt.% NaCl solution, 35 ℃, 48 h). The experimentally measured corrosion rate and electrochemical parameters were fed back to the database to incrementally update the prediction model, forming a closed-loop iteration.

[0033] The experimental results are shown in Table 1. On the standard test set, the corrosion rate prediction R using the chain prediction model (in this invention) incorporating electrochemical mechanism parameters... 2 The mean value was 0.921, which is significantly better than the 0.856 of the pure data-driven model without the introduction of electrochemical parameters. In the accelerated corrosion verification experiment, the average relative error between the measured corrosion rate and the predicted value of the three recommended alloys was 8.7%, which verified the predictive reliability and closed-loop reverse design effectiveness of the method of the present invention.

[0034] Table 1. Comparison of corrosion rate prediction performance of different models in Example 1

[0035] Example 2

[0036] Based on Example 1, this embodiment expands the optimization objective of the closed-loop reverse design from a dual objective of "corrosion rate + tensile strength" to a three objective of "corrosion rate + tensile strength + elongation" to further verify the ability of the method of the present invention in multi-objective collaborative optimization and closed-loop iteration.

[0037] (1) Multi-source corrosion data acquisition and standardized database construction: This embodiment uses the AZ-based magnesium alloy corrosion database constructed in Example 1, and supplements it with elongation data. A total of 200 sets of original data were collected, and after data cleaning, 186 sets of valid data were retained, including 162 sets of samples with complete elongation data. These were randomly divided into a training set of 130 sets and a test set of 32 sets at an 8:2 ratio. The corrosion medium conditions, composition range, and heat treatment state range were the same as in Example 1.

[0038] (2) Extraction of second-phase information and calculation of electrochemical mechanism parameters: For each sample in the database, based on alloy composition and heat treatment state, the type of the second phase (mainly Mg) was determined using JMatPro thermodynamic calculation software. 17 Al 12 And Al8Mn5) and their volume fractions. The electrochemical potential difference ΔE between each second phase and the α-Mg matrix was calculated based on the mixed potential theory: Mg 17 Al 12 The potential difference between Al8Mn5 and the matrix is ​​approximately +0.30 V, and the potential difference between Al8Mn5 and the matrix is ​​approximately +0.55 V. The area ratio and exchange current density of the second phase are also calculated. These electrochemical parameters are used as prior physical features and concatenated with the original composition, process characteristics, and elongation data to form an enhanced feature vector.

[0039] (3) Construct a chain-interpretable predictive model that integrates electrochemical mechanisms and data: The chain-like prediction model in this embodiment comprises a three-level prediction path. The first-level model uses the random forest algorithm, taking composition and process parameters as inputs and outputting the second-phase type and volume fraction. The second-level model uses the XGBoost algorithm, taking second-phase information as inputs and outputting potential difference and corrosion current density. The third-level model uses the XGBoost algorithm, taking electrochemical parameters, corrosion medium conditions, and alloy composition characteristics as inputs and outputting corrosion rate, tensile strength, and elongation. The three levels of models are connected in series to form a complete chain-like prediction path. The third-level model employs a multi-output regression strategy to simultaneously predict three performance indicators.

[0040] (4) Feature importance analysis and mechanism attribution based on SHAP: The SHAP method was used to decompose the feature importance of the third-level model, and attribution analysis was performed on the three outputs: corrosion rate, tensile strength, and elongation. The results showed that the top five features with the greatest impact on corrosion rate were, in order: the potential difference between the second phase and the matrix ΔE, Al content, aging temperature, Zn content, and pH value of the corrosive medium; the top five features with the greatest impact on tensile strength were, in order: Al content, aging temperature, Zn content, second phase volume fraction, and grain size; and the top five features with the greatest impact on elongation were, in order: Al content, aging temperature, second phase volume fraction, Zn content, and solution temperature. These attribution results are consistent with physical metallurgical principles, validating the interpretability of the model.

[0041] (5) Three-target closed-loop reverse design of corrosion-resistant components constrained by target corrosion rate: With a target corrosion rate <0.5 mm / year as a constraint, and minimizing the corrosion rate, maximizing tensile strength, and maximizing elongation as the three optimization objectives, the NSGA-III algorithm was used to search for the Pareto optimal solution set in the composition space. The optimization variables were Al content (3–11 wt.%), Zn content (0.5–4 wt.%), and aging temperature (160–240 ℃), with constraints of corrosion rate ≤0.5 mm / year, tensile strength ≥250 MPa, and elongation ≥8%. The population size was 150, with 300 generations, a crossover probability of 0.9, and a mutation probability of 0.1. A reference point guidance strategy was used to maintain population diversity. The Pareto front obtained through inverse design is shown below. Figure 3 As shown.

[0042] The optimization results show that the obtained Pareto front clearly demonstrates the trade-off between corrosion rate, tensile strength, and elongation: the high corrosion resistance solution is mainly concentrated in the region with medium to low Al content (3~6 wt.%) and relatively high aging temperature (200~230℃), where the second phase Mg... 17 Al 12 The volume fraction is low and discontinuous, resulting in a weak driving force for microgalvanic corrosion; high-intensity settling is mainly concentrated in the high Al content (8~11 wt.%) and low-temperature aging (160~180 ℃) region, where Mg... 17 Al 12 The large amount of cathodic phase precipitation produces a significant precipitation strengthening effect, but the continuous network distribution of the cathodic phase accelerates the corrosion of the matrix; the equilibrium disaggregate achieves a good balance in the range of corrosion rate 0.3~0.5 mm / year, tensile strength 280~320 MPa, and elongation 8%~12%.

[0043] (6) Accelerated corrosion experiment verification and closed-loop iterative update: Three representative alloy formulations were selected from the Pareto optimal solution set: a high corrosion resistance solution (Al 4.2 wt.%, Zn 1.8 wt.%, aging temperature 215 ℃), an equilibrium solution (Al 6.5 wt.%, Zn 2.5 wt.%, aging temperature 195 ℃), and a high strength solution (Al 9.3 wt.%, Zn 3.2 wt.%, aging temperature 170 ℃). The alloys were smelted and heat-treated according to the recommended compositions, and their corrosion performance was verified using an accelerated salt spray test (according to GB / T 10125 standard, 5 wt.% NaCl solution, 35 ℃, 48 h). Simultaneously, room temperature tensile properties were tested according to GB / T 228.1-2021. The experimentally measured corrosion rate, tensile strength, and elongation were fed back into the database to incrementally update the prediction model, forming a closed-loop iteration.

[0044] The experimental results are shown in Table 2. The average relative error between the measured and predicted values ​​of corrosion rate for the three recommended alloys was 9.2%, the average relative error between the measured and predicted values ​​of tensile strength was 5.8%, and the average relative error between the measured and predicted values ​​of elongation was 7.4%. The three-objective Pareto front provides a quantitative decision-making basis for selecting the optimal composition-process combination for different service environments: the high corrosion resistance solution is suitable for high-humidity marine environments, the equilibrium solution is suitable for general industrial environments, and the high strength solution is suitable for structural components with high load-bearing capacity requirements.

[0045] Table 2 Comparison of predicted and experimental values ​​of representative Pareto solutions in Example 2

[0046] It should be noted that the above embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A closed-loop reverse design method for the corrosion resistance of magnesium alloys based on electrochemical mechanism-data fusion, characterized in that, Includes the following steps: (1) Collect magnesium alloy corrosion data from multiple sources. The data includes at least alloy composition, heat treatment state, corrosion medium conditions, corrosion rate, corrosion potential and corrosion current density. After data cleaning and standardization, construct a structured corrosion database. (2) Based on the alloy composition and heat treatment state, determine the type and volume fraction of the second phase, and calculate the electrochemical potential difference and exchange current density of each second phase with the α-Mg matrix based on the mixed potential theory. The electrochemical mechanism parameters are then combined with the original composition and process characteristics as physical a priori features to form an enhanced feature vector. (3) Construct an electrochemical mechanism-data fusion chain interpretable prediction model. The model is a multi-scale chain structure of "composition / process → microstructure → electrochemical parameters → corrosion rate". The first-level model takes composition and process as input and second-phase information as output. The second-level model takes second-phase information as input and electrochemical parameters as output. The third-level model takes electrochemical parameters and corrosion medium conditions as input and corrosion rate as output. (4) The SHAP method was used to decompose the characteristics of the corrosion rate prediction model and to quantitatively evaluate the contribution and influence direction of each alloying element, process parameter and electrochemical mechanism parameter on the corrosion rate. (5) Using the target corrosion rate as a constraint and material cost or mechanical properties as optimization objectives, a multi-objective optimization algorithm is used to search for candidate alloy formulations that meet the corrosion resistance requirements in the alloy composition space. (6) Design accelerated corrosion experiments to verify the recommended alloy formulation, feed the experimental data back to the database to iteratively update the prediction model, and form a closed-loop reverse design.

2. The method according to claim 1, characterized in that, The corrosive medium in step (1) is a 3.5 wt.% NaCl solution with a pH of 7-12 and a temperature of 25-60 °C; the corrosion rate is determined by weight loss or electrochemical methods.

3. The method according to claim 1, characterized in that, The electrochemical potential difference ΔE mentioned in step (2) is calculated based on the mixed potential theory and satisfies: D E = E second − E matrix , in, E second This represents the self-corrosion potential of the α-Mg matrix. E matrix The self-corrosion potential of the α-Mg matrix is ​​given; when ΔE > 0, the second phase acts as the cathode, accelerating matrix corrosion; when ΔE < 0, the second phase acts as the anode, preferentially dissolving; the second phase type includes at least Mg. 17 Al 12 , Al8Mn5, Mg2Ca, MgZn2, Mg2Si, Mg 12 One or more of Zn2.

4. The method according to claim 1, characterized in that, In step (2), the second phase type and volume fraction are determined by thermodynamic calculation software JMatPro or Thermo-Calc, or obtained by literature data and experimental characterization data.

5. The method according to claim 1, characterized in that, The multi-scale chain interpretable prediction model described in step (3) is implemented using random forest, XGBoost or deep neural network algorithms at each level; the input features of the third-level model include electrochemical parameters, corrosion medium conditions and alloy composition features, and the output is corrosion rate.

6. The method according to claim 1, characterized in that, The SHAP method described in step (4) calculates the Shapley value of each feature on the corrosion rate prediction result, and displays the influence direction and magnitude sorting of all features through a summary diagram.

7. The method according to claim 1, characterized in that, The multi-objective optimization algorithm mentioned in step (5) adopts the NSGA-II or NSGA-III algorithm. The optimization objectives are to minimize the corrosion rate and maximize the tensile strength. The constraints include the upper limit of corrosion rate, the lower limit of mechanical properties, and the range of alloy element content. Alternatively, the Bayesian optimization method can be used, with the target corrosion rate as the constraint and cost or strength as the optimization objective, to sequentially search for the optimal formulation in the composition space.

8. The method according to claim 1, characterized in that, The accelerated corrosion experiment described in step (6) adopts salt spray test or potentiodynamic polarization test, and the cycle of a single experiment is 24~72 h; after the verification results are fed back to the database, the prediction model is incrementally updated or retrained.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 8.

10. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor, when executing the computer program, implements the method as described in any one of claims 1 to 8.

Images