High-entropy alloy corrosion-resistant coating based on machine learning optimization and preparation method thereof

Abstract

The invention relates to the technical field of coating preparation, in particular to a high-entropy alloy corrosion-resistant coating based on machine learning optimization and a preparation method thereof.The preparation method comprises the steps that a preparation data set of the high-entropy alloy coating is constructed, and coating element components, technological parameters and corrosion environment parameters in the preparation data set serve as input characteristics; taking the thickness, the hardness and the elastic modulus of the corrosion layer as output characteristics, training an artificial neural network model, and establishing a performance prediction model; a multi-objective particle swarm optimization algorithm is adopted, and magnetron sputtering process parameters are optimized and solved by taking the minimum thickness of a corrosion layer and the maximum hardness and elasticity modulus as optimization objectives; according to the optimal process parameter combination, preparing an AlCrFeMoTi high-entropy alloy coating on a ferrite / martensitic steel matrix by adopting a magnetron sputtering five-target co-deposition technology; and carrying out corrosion resistance test on the prepared coating, and comparing a test result with a predicted value of the model to verify the accuracy of the model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of coating preparation, in particular to a high-entropy alloy corrosion-resistant coating based on machine learning optimization and a preparation method thereof. BACKGROUND

[0002] With the rapid development of advanced nuclear energy, liquid metal energy storage and other fields, the service life of core components in extreme corrosion environments has become a key problem restricting the upgrading of technology. Among them, liquid lead-bismuth alloy (such as Pb-45%Bi, mass fraction) is widely used as the coolant of the fourth generation nuclear reactor (such as lead-cooled fast reactor) and the heat transfer medium of liquid metal energy storage system due to its high thermal conductivity, low vapor pressure and excellent neutron moderation capability. However, liquid lead-bismuth has strong corrosiveness at high temperature (300-600℃), which can chemically react with structural materials (such as ferrite / martensite steel) to form a loose corrosion product layer (such as 、 , etc.), resulting in material wall thickness thinning, mechanical property degradation, and even leakage risk, seriously affecting the safety and reliability of equipment operation.

[0003] Chinese patent application publication No. CN118086887A discloses a high-hardness corrosion-resistant AlCoCrFeNiMox high-entropy alloy coating and a preparation method. The preparation method comprises: S1, placing AlCoCrFeNi high-entropy alloy powder and metal Mo powder in a ball milling tank containing stainless steel milling balls under inert atmosphere, sealing the ball milling tank and placing it in a ball mill for ball milling, sieving and vacuum drying to obtain AlCoCrFeNiMox high-entropy alloy powder with a particle size of 15-53μm, wherein 0.2≤x≤1; S2, using a steel cylindrical rod as a substrate, polishing and cleaning the surface with anhydrous ethanol; S3, using an ultra-high-speed laser cladding process, AlCoCrFeNiMox high-entropy alloy powder is cladded on the surface of the substrate to prepare a high-entropy alloy coating. The present application prepares a high-entropy alloy coating with a single BCC phase structure, and the coating has a coarse columnar crystal structure, high hardness and excellent corrosion resistance.

[0004] However, the existing technology still has the following problems: The existing technology relies on trial-and-error method to optimize parameters, which is difficult to handle the nonlinear relationship between composition, process and performance; it is not designed for liquid lead-bismuth corrosion, and the corrosion resistance of the coating is insufficient; it lacks a multi-objective optimization mechanism, making it difficult to balance performance; there is no data-model-process closed loop, and the substrate bonding is prone to problems. SUMMARY

[0005] To address these issues, this invention provides a high-entropy alloy corrosion-resistant coating and its preparation method based on machine learning optimization. This overcomes the problems of existing technologies that rely on trial-and-error to optimize parameters, making it difficult to handle the nonlinear relationship between composition, process, and performance; lacking design for liquid lead-bismuth corrosion, resulting in insufficient coating corrosion resistance; lacking a multi-objective optimization mechanism, making it difficult to balance performance; lacking a data-model-process closed loop; and being prone to problems with substrate bonding.

[0006] To achieve the above objectives, this invention provides a method for preparing a high-entropy alloy corrosion-resistant coating based on machine learning optimization. It includes: Step S1: Construct a preparation dataset for high-entropy alloy coatings. The preparation dataset includes coating elemental composition, magnetron sputtering process parameters, corrosion layer thickness in liquid lead-bismuth environment, coating hardness, and elastic modulus. Step S2: Using the coating element composition, process parameters, and corrosion environment parameters in the prepared dataset as input features, and the corrosion layer thickness, hardness, and elastic modulus as output features, train an artificial neural network model and establish a performance prediction model. Step S3: A multi-objective particle swarm optimization algorithm is used to optimize the magnetron sputtering process parameters with the optimization objectives of minimizing the corrosion layer thickness and maximizing the hardness and elastic modulus. Step S4: Based on the optimal combination of process parameters, an AlCrFeMoTi high-entropy alloy coating is prepared on a ferrite / martensitic steel substrate using magnetron sputtering five-target co-deposition technology. Step S5: Conduct corrosion resistance tests on the prepared coating, and compare the test results with the predicted values ​​of the model to verify the accuracy of the model.

[0007] Further, in step S2, establishing the performance prediction model includes: Step S21: Preprocess the prepared dataset, including data cleaning, outlier removal and feature standardization, and normalize the input features and output features respectively. Step S22: Determine the structure of the artificial neural network model, which includes an input layer, two hidden layers and an output layer, wherein the hidden layer activation function is the ReLU function and the output layer activation function is the linear activation function; Step S23: Divide the preprocessed dataset into a training set and a test set according to a preset ratio, and use the training set to train the artificial neural network model; Step S24: Use the test set to verify the performance of the trained model, calculate the root mean square error between the predicted value and the experimental value, and confirm that the model has been established when the prediction accuracy of the model on the test set is greater than the preset prediction accuracy.

[0008] Furthermore, in step S3, the optimization solution for the magnetron sputtering process parameters includes: Step S31: Initialize the parameters of the multi-objective particle swarm optimization algorithm, including particle swarm size, maximum number of iterations, inertia weight, individual learning factor and social learning factor, and optimize the search space of magnetron sputtering process parameters. Step S32: Randomly generate an initial particle swarm. The position vector of each particle represents a set of magnetron sputtering process parameters, including sputtering temperature, total sputtering power, and substrate bias voltage. Step S33: Input the position vector of each particle into the trained performance prediction model to obtain the predicted values ​​of the corresponding corrosion layer thickness, hardness and elastic modulus, and calculate the fitness value of each particle. Step S34: Based on the Pareto dominance relation of the multi-objective optimization problem, update the individual optimal position of each particle and the global optimal position set of the particle swarm, and maintain an external archive set to store the currently found non-dominated solutions. Step S35: Update the velocity and position of all particles according to the velocity update formula and position update formula of the particle swarm optimization algorithm; Step S36: Repeat steps S33 to S35 until the maximum number of iterations is reached or the convergence condition is met. Finally, select an optimal combination of magnetron sputtering process parameters from the external archive set as the optimization result.

[0009] Further, in step S35, the velocity update formula of the particle swarm optimization algorithm is: v_id(k+1)=w×v_id(k)+c1×r1×(pbest_id-x_id(k))+c2×r2×(gbest_id-x_id(k)); Where v_id(k+1) represents the updated velocity of the i-th particle in the d-th dimension; v_id(k) represents the current velocity of the i-th particle in the d-th dimension; w is the inertia weight; c1 and c2 are the individual learning factor and the social learning factor, respectively; r1 and r2 are random numbers between [0,1]; pbest_id represents the individual historical best position of the i-th particle in the d-th dimension; gbest_id represents the global best position of the particle swarm in the d-th dimension; and x_id(k) represents the current position of the i-th particle in the d-th dimension. The position update formula is: x_id(k+1)=x_id(k)+v_id(k+1); Where x_id(k+1) represents the position of the i-th particle after updating in the d-th dimension.

[0010] Further, in step S36, the convergence condition is that the number of non-dominated solutions in the external archive set no longer changes in 10 consecutive iterations, or the improvement of all objective functions is less than a preset threshold. Wherein, the improvement magnitude of all objective functions is the relative rate of change of the hypervolume, calculated by the following formula: H = |HV(k) - HV(k-1)| / HV(k-1); Where H is the relative rate of change of the hypervolume, HV(k) is the hypervolume of the current iteration, and HV(k-1) is the hypervolume of the previous iteration.

[0011] Further, in step S36, the final selection of an optimal combination of magnetron sputtering process parameters from the external archive set as the optimization result includes: A sorting method based on crowding distance is used to filter the non-dominated solutions in the final external archive set, and the solution located in the central region of the Pareto front is selected as the final combination of optimized process parameters.

[0012] Further, in step S4, the preparation of the AlCrFeMoTi high-entropy alloy coating using magnetron sputtering five-target co-deposition technology includes: Step S41: Select Al, Cr, Fe, Mo, or Ti metal targets as sputtering sources and ferritic / martensitic steel as matrix materials. Step S42 involves pre-treating the substrate material, including mechanical polishing, ultrasonic cleaning, and argon ion bombardment cleaning in a vacuum chamber. Step S43: Based on the optimized combination of process parameters, set the sputtering temperature, total sputtering power, and substrate bias voltage; Step S44: In an ultra-high vacuum environment, high-purity argon gas is introduced as the working gas, and the five-target co-deposition system is started. Magnetron sputtering deposition is performed under the set process parameters to form an AlCrFeMoTi high-entropy alloy coating on the substrate.

[0013] Further, in step S5, the corrosion resistance test of the prepared coating includes: Step S51: The cross-sectional morphology of the coating is observed using a scanning electron microscope, and the thickness of the corrosion layer formed after the experiment in a liquid lead-bismuth environment is measured. Step S52: The hardness and elastic modulus of the coating are measured using nanoindentation technology; Step S53: Compare the experimentally measured corrosion layer thickness, hardness, and elastic modulus with the predicted values ​​of the performance prediction model under the optimal process parameters, and calculate the relative error. Step S54: If the relative error between the predicted and experimental values ​​of all performance indicators is less than the preset relative error, the model is determined to be accurate; otherwise, the experimental data is added to the prepared dataset, the model is retrained and optimized, and iterative improvements are made.

[0014] Furthermore, in step S54, experimental data is added to the prepared dataset, the model is retrained and optimized, and iterative improvements are made, including: Step S541: Add the process parameters and performance data obtained in this preparation and testing as new samples to the preparation dataset to form an amplification dataset; Step S542: Based on the augmented dataset, re-execute the data preprocessing, model training and validation steps to obtain the updated performance prediction model; Step S543: Based on the updated performance prediction model, re-execute the multi-objective particle swarm optimization algorithm to solve for the new optimal combination of process parameters; Step S544: Based on the new optimal combination of process parameters, prepare the coating again and conduct corrosion resistance tests, and verify the accuracy of the model predictions; Step S545: Repeat the above iterative steps until the relative errors between the predicted and experimental values ​​of all performance indicators meet the preset accuracy requirements.

[0015] A high-entropy alloy corrosion-resistant coating applied to the high-entropy alloy corrosion-resistant coating prepared by the aforementioned machine learning-optimized high-entropy alloy corrosion-resistant coating and its preparation method, comprising: The high-entropy alloy coating is a five-element system of aluminum, chromium, iron, molybdenum, and titanium, with the following mass fractions for each element: aluminum 15%-25%, chromium 20%-30%, iron 15%-25%, molybdenum 10%-20%, and titanium 10%-20%.

[0016] Compared with existing technologies, the advantages of this invention lie in its ability to construct a complete input layer containing 19 input features and employ a double-layer hidden layer structure with 64-16 neurons. This model possesses powerful nonlinear mapping capabilities, accurately capturing the complex relationship between preparation processes, composition, and final performance. The RMSE (Root Mean Square Error) results on the actual test set all meet and exceed the preset accuracy requirements, demonstrating that the model can simultaneously predict the corrosion resistance and mechanical properties of the coating with extremely high accuracy. This provides a reliable objective function for subsequent process optimization, avoiding optimization failures due to inaccurate model predictions. A high-precision model trained with only 29 samples can achieve rapid and low-cost virtual screening of the performance of countless potential combinations of process parameters.

[0017] Furthermore, this invention employs a multi-objective particle swarm optimization algorithm based on Pareto dominance relations, capable of simultaneously handling three competing objectives: minimizing corrosion layer thickness, maximizing hardness, and maximizing elastic modulus. By maintaining an external archive set to store non-dominated solutions, the algorithm can explore the entire objective space in parallel, ultimately outputting a Pareto optimal solution set covering multiple performance trade-offs, rather than a single solution. It abandons the simple convergence judgment relying on a fixed number of iterations, introducing an intelligent convergence criterion based on the relative change rate of hypervolume, and using a crowding distance-based sorting method to select solutions located in the center region of the leading edge, automatically avoiding those that severely sacrifice other performance metrics in pursuit of a single optimal performance. The edge solutions for performance, and the solutions located in the central region, represent the best balance among multiple performance indicators. They have the best overall performance and are usually more robust in practical applications. Combining the particle swarm optimization algorithm with a trained artificial neural network model creates a powerful virtual laboratory. The performance of each particle is predicted instantaneously through the ANN model, replacing time-consuming physical experiments. This allows the algorithm to evaluate thousands of parameter combinations within hours. The global search characteristics of the particle swarm algorithm enable it to effectively explore a wide range of process parameters, greatly increasing the probability of finding the global optimum or near-global optimum process window and overcoming the shortcomings of traditional gradient descent methods that are prone to getting trapped in local optima.

[0018] Furthermore, by employing a five-target co-deposition technique and precisely allocating the target power ratio, this invention achieves precise control over the composition of AlCrFeMoTi, a multi-principal-element high-entropy alloy, ensuring that the high-entropy effect is fully utilized. At the same time, the optimized process parameters provide sufficient migration and bombardment energy for the deposited particles, thereby promoting the densification and uniform growth of the coating. Ultimately, a dense, uniform, and defect-free coating is obtained, which is the microstructural basis for its excellent corrosion resistance and mechanical properties.

[0019] Furthermore, this invention combines a multi-objective particle swarm optimization algorithm with an artificial neural network model to achieve synergistic optimization of minimizing corrosion layer thickness, maximizing hardness, and maximizing elastic modulus. The model is optimized through data iteration to strictly control the relative error of performance prediction within a preset threshold. A closed-loop iterative mechanism for experimental data, model updates, parameter optimization, coating preparation, and performance verification is established to avoid the dispersion of coating performance caused by parameter fluctuations in traditional processes. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the workflow of the machine learning-optimized method for preparing high-entropy alloy corrosion-resistant coatings according to the present invention. Figure 2This is a flowchart illustrating the process of establishing a performance prediction model in the machine learning-optimized method for preparing high-entropy alloy corrosion-resistant coatings of this invention. Figure 3 This is a flowchart illustrating the process of preparing the AlCrFeMoTi high-entropy alloy coating in the machine learning-optimized high-entropy alloy corrosion-resistant coating preparation method of this invention. Detailed Implementation

[0021] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0022] It should be noted that the data in this embodiment are all derived from a comprehensive analysis and evaluation of historical data from the six months prior to this determination and the corresponding historical determination results by the system described in this invention. Those skilled in the art will understand that the system described in this invention can determine the above-mentioned parameters for a single item by selecting the value with the highest proportion based on the data distribution as the preset standard parameter, using weighted summation to obtain the value as the preset standard parameter, substituting each historical data point into a specific formula and using the value obtained by that formula as the preset standard parameter, or other selection methods, as long as the system described in this invention can clearly define different specific situations in the single-item determination process through the obtained values.

[0023] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0024] Please see Figures 1-3 As shown, Figure 1 This is a flowchart illustrating the workflow of the machine learning-optimized method for preparing high-entropy alloy corrosion-resistant coatings according to the present invention. Figure 2 This is a flowchart illustrating the process of establishing a performance prediction model in the machine learning-optimized method for preparing high-entropy alloy corrosion-resistant coatings of this invention. Figure 3 This is a flowchart illustrating the process of preparing the AlCrFeMoTi high-entropy alloy coating in the machine learning-optimized high-entropy alloy corrosion-resistant coating preparation method of this invention.

[0025] This invention relates to a machine learning-optimized method for preparing high-entropy alloy corrosion-resistant coatings, comprising: Step S1: Construct a preparation dataset for high-entropy alloy coatings. The preparation dataset includes coating elemental composition, magnetron sputtering process parameters, corrosion layer thickness in liquid lead-bismuth environment, coating hardness, and elastic modulus. Step S2: Using the coating element composition, process parameters, and corrosion environment parameters in the prepared dataset as input features, and the corrosion layer thickness, hardness, and elastic modulus as output features, train an artificial neural network model and establish a performance prediction model. Step S3: A multi-objective particle swarm optimization algorithm is used to optimize the magnetron sputtering process parameters with the optimization objectives of minimizing the corrosion layer thickness and maximizing the hardness and elastic modulus. Step S4: Based on the optimal combination of process parameters, an AlCrFeMoTi high-entropy alloy coating is prepared on a ferrite / martensitic steel substrate using magnetron sputtering five-target co-deposition technology. Step S5: Conduct corrosion resistance tests on the prepared coating, and compare the test results with the predicted values ​​of the model to verify the accuracy of the model.

[0026] Specifically, in step S2, establishing the performance prediction model includes: Step S21: Preprocess the prepared dataset, including data cleaning, outlier removal and feature standardization, and normalize the input features and output features respectively. Step S22: Determine the structure of the artificial neural network model. The model includes an input layer, two hidden layers, and an output layer. The number of neurons in the input layer is the same as the dimension of the input features. The first hidden layer contains 64 neurons, the second hidden layer contains 16 neurons, and the output layer contains 3 neurons corresponding to the three performance indicators of corrosion layer thickness, hardness, and elastic modulus, respectively. The activation function of the hidden layer is the ReLU function, and the activation function of the output layer is the linear activation function. Step S23: Divide the preprocessed dataset into a training set and a test set according to a preset ratio, and use the training set to train the artificial neural network model; Step S24: Use the test set to verify the performance of the trained model, calculate the root mean square error between the predicted value and the experimental value, and confirm that the model has been established when the prediction accuracy of the model on the test set is greater than the preset prediction accuracy.

[0027] In this embodiment of the invention, the artificial neural network model adopts the following structure: Input layer: The number of neurons is the same as the dimension of the input features, totaling 19 neurons, corresponding to 13 elemental components (including Al, Cr, Fe, Mo, Ti, etc.), 3 process parameters (sputtering temperature, total sputtering power, substrate bias voltage), and 3 corrosion environment parameters (corrosion time, oxygen concentration, corrosion temperature); Hidden layer: The first hidden layer contains 64 neurons, and the second hidden layer contains 16 neurons, both using the ReLU activation function (f(x) = max(0,x)) to introduce nonlinearity and avoid the gradient vanishing problem; Output layer: Contains 3 neurons, which output the predicted values ​​of corrosion layer thickness, hardness, and elastic modulus, respectively, using the linear activation function (f(x) = ... To adapt to the regression task, the network parameters were initialized using the He initialization method to maintain gradient stability. The preprocessed dataset was randomly divided into a 70% training set and a 30% test set (preset ratio). The artificial neural network model was trained using the training set (29 samples). The preset prediction accuracy requirements were: root mean square error (RMSE) between predicted and experimental values ​​≤0.2 μm for corrosion layer thickness, ≤0.3 GPa for hardness, and ≤0.15 GPa for elastic modulus. Actual test results showed that the model's RMSE values ​​on the test set were 0.195 μm, 0.254 GPa, and 0.162 GPa, respectively, all meeting the preset accuracy requirements. Therefore, the performance prediction model was confirmed to be successfully established.

[0028] This invention constructs a complete input layer containing 19 input features and employs a double-layer hidden layer structure with 64-16 neurons. The model possesses powerful nonlinear mapping capabilities, accurately capturing the complex relationship between preparation processes, composition, and final performance. The RMSE (Root Mean Square Error) results on the actual test set all meet and exceed the preset accuracy requirements. This demonstrates that the model can simultaneously predict the corrosion resistance and mechanical properties of the coating with extremely high accuracy, providing a reliable objective function for subsequent process optimization and avoiding optimization failures due to inaccurate model predictions. A high-precision model trained with only 29 samples can achieve rapid and low-cost virtual screening of the performance of countless potential combinations of process parameters.

[0029] Specifically, in step S3, the optimization solution for the magnetron sputtering process parameters includes: Step S31: Initialize the parameters of the multi-objective particle swarm optimization algorithm, including particle swarm size, maximum number of iterations, inertia weight, individual learning factor and social learning factor, and optimize the search space of magnetron sputtering process parameters. Step S32: Randomly generate an initial particle swarm. The position vector of each particle represents a set of magnetron sputtering process parameters, including sputtering temperature, total sputtering power, and substrate bias voltage. Step S33: Input the position vector of each particle into the trained performance prediction model to obtain the predicted values ​​of the corresponding corrosion layer thickness, hardness and elastic modulus, and calculate the fitness value of each particle. Step S34: Based on the Pareto dominance relation of the multi-objective optimization problem, update the individual optimal position of each particle and the global optimal position set of the particle swarm, and maintain an external archive set to store the currently found non-dominated solutions. Step S35: Update the velocity and position of all particles according to the velocity update formula and position update formula of the particle swarm optimization algorithm; Step S36: Repeat steps S33 to S35 until the maximum number of iterations is reached or the convergence condition is met. Finally, select an optimal combination of magnetron sputtering process parameters from the external archive set as the optimization result.

[0030] Specifically, in step S35, the velocity update formula of the particle swarm optimization algorithm is: v_id(k+1)=w×v_id(k)+c1×r1×(pbest_id-x_id(k))+c2×r2×(gbest_id-x_id(k)); Where v_id(k+1) represents the updated velocity of the i-th particle in the d-th dimension; v_id(k) represents the current velocity of the i-th particle in the d-th dimension; w is the inertia weight; c1 and c2 are the individual learning factor and the social learning factor, respectively; r1 and r2 are random numbers between [0,1]; pbest_id represents the individual historical best position of the i-th particle in the d-th dimension; gbest_id represents the global best position of the particle swarm in the d-th dimension; and x_id(k) represents the current position of the i-th particle in the d-th dimension. The position update formula is: x_id(k+1)=x_id(k)+v_id(k+1); Where x_id(k+1) represents the position of the i-th particle after updating in the d-th dimension.

[0031] Specifically, in step S36, the convergence condition is that the number of non-dominated solutions in the external archive set no longer changes in 10 consecutive iterations, or the improvement of all objective functions is less than a preset threshold. Wherein, the improvement magnitude of all objective functions is the relative rate of change of the hypervolume, calculated by the following formula: H = |HV(k) - HV(k-1)| / HV(k-1); Where H is the relative rate of change of the hypervolume, HV(k) is the hypervolume of the current iteration, and HV(k-1) is the hypervolume of the previous iteration.

[0032] The preset threshold mentioned in this embodiment of the invention can be determined by the following method: before the formal iteration of the optimization algorithm, several pre-iterations (e.g., 20-50 times) are performed, the statistical distribution of the relative change rate H of the hypervolume during this set of pre-iterations is calculated, its standard deviation is calculated, and the preset threshold is set to 0.5 times the standard deviation. Hypervolume refers to the volume of the region enclosed by the Pareto front (i.e., the non-dominated solution in the external archive set) and a pre-selected reference point in the target space (in two dimensions, the target is area, in three dimensions it is volume, and so on). Hypervolume simultaneously measures the convergence (how close to the ideal point) and diversity (how wide the coverage) of the Pareto solution set. The larger the hypervolume, the higher the overall quality of the solution set.

[0033] Specifically, in step S36, the final selection of an optimal combination of magnetron sputtering process parameters from the external archive set as the optimization result includes: A sorting method based on crowding distance is used to filter the non-dominated solutions in the final external archive set, and the solution located in the central region of the Pareto front is selected as the final combination of optimized process parameters.

[0034] This invention employs a multi-objective particle swarm optimization algorithm based on Pareto dominance, capable of simultaneously handling three competing objectives: minimizing corrosion layer thickness, maximizing hardness, and maximizing elastic modulus. By maintaining an external archive to store non-dominated solutions, the algorithm can explore the entire objective space in parallel, ultimately outputting a Pareto optimal solution set covering various performance trade-offs, rather than a single solution. It abandons the simple convergence judgment relying on a fixed number of iterations, introducing an intelligent convergence criterion based on the relative change rate of hypervolume. A crowding distance-based sorting method is used to select solutions located in the center region of the leading edge, automatically avoiding solutions that severely sacrifice other performance metrics in pursuit of a single optimal performance. Edge solutions, located in the central region, represent the optimal balance among multiple performance metrics. They offer the best overall performance and are typically more robust in practical applications. Combining particle swarm optimization (PSO) with a trained artificial neural network model creates a powerful virtual laboratory where the performance of each particle is instantly predicted using the ANN model, replacing time-consuming physical experiments. This allows the algorithm to evaluate thousands of parameter combinations within hours. The global search characteristic of PSO enables it to effectively explore a vast process parameter space, greatly increasing the probability of finding the globally optimal or near-global optimal process window and overcoming the tendency of traditional gradient descent methods to get trapped in local optima.

[0035] Specifically, in step S4, the preparation of the AlCrFeMoTi high-entropy alloy coating using magnetron sputtering five-target co-deposition technology includes: Step S41: Select Al, Cr, Fe, Mo, or Ti metal targets as sputtering sources and ferritic / martensitic steel as matrix materials. Step S42 involves pre-treating the substrate material, including mechanical polishing, ultrasonic cleaning, and argon ion bombardment cleaning in a vacuum chamber. Step S43: Based on the optimized combination of process parameters, set the sputtering temperature, total sputtering power, and substrate bias voltage; Step S44: In an ultra-high vacuum environment, high-purity argon gas is introduced as the working gas, and the five-target co-deposition system is started. Magnetron sputtering deposition is performed under the set process parameters to form an AlCrFeMoTi high-entropy alloy coating on the substrate.

[0036] In this embodiment of the invention, Al, Cr, Fe, Mo, and Ti targets with a purity greater than 99.95% are selected, and the matrix is ​​T91 ferritic / martensitic steel. The targets are progressively sanded to 2000# and then mechanically polished to a mirror finish. Following this, ultrasonic cleaning is performed for 15 minutes each in acetone and ethanol. Finally, deep cleaning is carried out in a vacuum chamber by Ar ion bombardment for 20 minutes. Deposition: The vacuum chamber is evacuated to a minimum vacuum level. Based on the optimization results, the process parameters were set as follows: the temperature was heated to 380℃ and stabilized, the substrate bias was adjusted to -120 V, high-purity argon gas was introduced to maintain the working pressure at 0.5 Pa, and the total power of 950 W was distributed to five targets according to the preset ratio (Al: 22%, Cr: 24%, Fe: 20%, Mo: 18%, Ti: 16%). The baffle was turned on, and magnetron sputtering deposition was carried out on the T91 substrate for 120 minutes. Finally, a dense and uniform AlCrFeMoTi high-entropy alloy coating with a thickness of about 5 μm was obtained.

[0037] This invention achieves precise control over the composition of AlCrFeMoTi, a multi-principal-element high-entropy alloy, by employing a five-target co-deposition technique and accurately allocating the target power ratio. This ensures that the high-entropy effect is fully utilized. At the same time, the optimized process parameters provide sufficient migration and bombardment energy for the deposited particles, thereby promoting the densification and uniform growth of the coating. Ultimately, a dense, uniform, and defect-free coating is obtained, which is the microstructural basis for its excellent corrosion resistance and mechanical properties.

[0038] Specifically, in step S5, the corrosion resistance test of the prepared coating includes: Step S51: The cross-sectional morphology of the coating is observed using a scanning electron microscope, and the thickness of the corrosion layer formed after the experiment in a liquid lead-bismuth environment is measured. Step S52: The hardness and elastic modulus of the coating are measured using nanoindentation technology; Step S53: Compare the experimentally measured corrosion layer thickness, hardness, and elastic modulus with the predicted values ​​of the performance prediction model under the optimal process parameters, and calculate the relative error. Step S54: If the relative error between the predicted and experimental values ​​of all performance indicators is less than the preset relative error, the model is determined to be accurate; otherwise, the experimental data is added to the prepared dataset, the model is retrained and optimized, and iterative improvements are made.

[0039] In this embodiment of the invention, the preset relative error is first determined to be 5% based on the measurement system error of scanning electron microscopy and nanoindentation technology; then, combined with the performance tolerance requirements of high-entropy alloy coating engineering applications, it is set to a fixed value between 8% and 15%; preferably, the preset relative error is set to 10%. Based on the optimized process parameters (sputtering temperature 380℃, total sputtering power 950W, substrate bias voltage -120V), an AlCrFeMoTi high-entropy alloy coating (denoted as HEA-1) was prepared on a T91 steel substrate. The HEA-1 coating sample was placed in a static liquid lead-bismuth alloy at 650℃ for a 1000-hour corrosion experiment. Afterwards, the cross-sectional morphology of the coating was observed using a scanning electron microscope (SEM, such as a Zeiss Sigma 300). The elemental distribution was analyzed by line scanning using an attached energy dispersive spectroscopy (EDS) instrument to determine the interface between the corroded layer and the original coating, and the thickness of five randomly selected points was measured. The arithmetic mean was taken as the experimentally measured thickness of the corroded layer, which was 2.1 µm. Nanoindentation tests were performed on the coating surface using a nanoindenter (such as an Agilent G200). A Berkovich diamond indenter was selected, and the continuous stiffness measurement method was used to conduct tests at six different locations to obtain load-displacement curves. The experimental hardness of the coating was calculated to be 12.8 GPa and the experimental elastic modulus to be 215 GPa by analyzing the curves using the Oliver-Pharr method. The process parameters used to prepare HEA-1 (380℃, 950W, -120V) were input into a trained performance prediction model (ANN model), yielding the following predicted values: corrosion layer thickness 2.3 µm, hardness 12.5 GPa, and elastic modulus 208 GPa. The relative errors were calculated as follows: corrosion layer thickness error: |2.1 - 2.3| / 2.1 × 100% ≈ 9.5%; hardness error: |12.8 - 12.5| / 12.8 × 100% ≈ 2.3%; elastic modulus error: |215 - 208| / 215 × 100% ≈ 3.3%. If the preset relative error is 10%, the model is considered accurate because the prediction errors of all performance indicators (corrosion layer thickness 9.5%, hardness 2.3%, elastic modulus 3.3%) are less than the preset 10%. If the prediction errors of all performance indicators exceed the preset relative error of 10%, the model is considered to need to be updated and the iterative improvement process is initiated.

[0040] Specifically, in step S54, experimental data is added to the prepared dataset, the model is retrained and optimized, and iterative improvements are made, including: Step S541: Add the process parameters and performance data obtained in this preparation and testing as new samples to the preparation dataset to form an amplification dataset; Step S542: Based on the augmented dataset, re-execute the data preprocessing, model training and validation steps to obtain the updated performance prediction model; Step S543: Based on the updated performance prediction model, re-execute the multi-objective particle swarm optimization algorithm to solve for the new optimal combination of process parameters; Step S544: Based on the new optimal combination of process parameters, prepare the coating again and conduct corrosion resistance tests, and verify the accuracy of the model predictions; Step S545: Repeat the above iterative steps until the relative errors between the predicted and experimental values ​​of all performance indicators meet the preset accuracy requirements.

[0041] In this embodiment of the invention, the complete data of the HEA-2 coating (including its process parameters and measured corrosion layer thickness, hardness, and elastic modulus) is added as a new data sample to the original dataset of 82 sets, forming an expanded dataset containing 83 sets of data. Step S21 is re-executed on these 83 sets of data for data cleaning and standardization. Steps S22-S24 are re-executed using the same network structure (8-64-16-3), and the expanded dataset is re-divided into a training set (58 sets) and a test set (25 sets) in a new 7:3 ratio to retrain the ANN model. After training, the new model (ANN-v2) is validated on the new test set, and its RMSE and R² both meet the requirements. Based on the newly trained performance prediction model ANN-v2, the multi-objective particle swarm optimization algorithm in step S3 is re-executed. Since the model has been updated, its performance on the process... - The mapping of performance relationships is more accurate, so the optimization algorithm will explore a new parameter space and output a new set of optimal process parameter combinations (e.g., sputtering temperature 400℃, total sputtering power 1000W, substrate bias voltage -110V). Based on the new optimal process parameters, step S4 is repeated to prepare a new coating (denoted as HEA-3). Steps S51-S53 are repeated on the HEA-3 coating to perform corrosion and mechanical property tests, and the results are compared with the predicted values ​​of the ANN-v2 model. If the error between the test results of HEA-3 and the predicted values ​​of ANN-v2 is less than 10%, the iteration is successful and the process terminates. If the index error still exceeds the standard, steps S541 to S544 are repeated to add the data of HEA-3 back to the dataset, train the ANN-v3 model, and optimize and verify it again until the prediction accuracy of the model meets the requirements.

[0042] This invention combines a multi-objective particle swarm optimization algorithm with an artificial neural network model to achieve synergistic optimization of minimizing corrosion layer thickness, maximizing hardness, and maximizing elastic modulus. The model is optimized through data iteration to strictly control the relative error of performance prediction within a preset threshold. A closed-loop iterative mechanism is established for experimental data, model updates, parameter optimization, coating preparation, and performance verification to avoid coating performance dispersion caused by parameter fluctuations in traditional processes.

[0043] A high-entropy alloy corrosion-resistant coating applied to the high-entropy alloy corrosion-resistant coating prepared by the aforementioned machine learning-optimized high-entropy alloy corrosion-resistant coating and its preparation method, comprising: The high-entropy alloy coating is a five-element system of aluminum, chromium, iron, molybdenum, and titanium, with the following mass fractions for each element: aluminum 15%-25%, chromium 20%-30%, iron 15%-25%, molybdenum 10%-20%, and titanium 10%-20%.

[0044] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for preparing a high-entropy alloy corrosion-resistant coating based on machine learning optimization, characterized in that, include: Step S1: Construct a preparation dataset for high-entropy alloy coatings. The preparation dataset includes coating elemental composition, magnetron sputtering process parameters, corrosion layer thickness in liquid lead-bismuth environment, coating hardness, and elastic modulus. Step S2: Using the coating element composition, process parameters, and corrosion environment parameters in the prepared dataset as input features, and the corrosion layer thickness, hardness, and elastic modulus as output features, train an artificial neural network model and establish a performance prediction model. Step S3: A multi-objective particle swarm optimization algorithm is used to optimize the magnetron sputtering process parameters with the optimization objectives of minimizing the corrosion layer thickness and maximizing the hardness and elastic modulus. Step S4: Based on the optimal combination of process parameters, an AlCrFeMoTi high-entropy alloy coating is prepared on a ferrite / martensitic steel substrate using magnetron sputtering five-target co-deposition technology. Step S5: Conduct corrosion resistance tests on the prepared coating, and compare the test results with the predicted values ​​of the model to verify the accuracy of the model.

2. The method for preparing a high-entropy alloy corrosion-resistant coating based on machine learning optimization according to claim 1, characterized in that, In step S2, establishing the performance prediction model includes: Step S21: Preprocess the prepared dataset, including data cleaning, outlier removal and feature standardization, and normalize the input features and output features respectively. Step S22: Determine the structure of the artificial neural network model, which includes an input layer, two hidden layers and an output layer, wherein the hidden layer activation function is the ReLU function and the output layer activation function is the linear activation function; Step S23: Divide the preprocessed dataset into a training set and a test set according to a preset ratio, and use the training set to train the artificial neural network model; Step S24: Use the test set to verify the performance of the trained model, calculate the root mean square error between the predicted value and the experimental value, and confirm that the model has been established when the prediction accuracy of the model on the test set is greater than the preset prediction accuracy.

3. The method for preparing a high-entropy alloy corrosion-resistant coating based on machine learning optimization according to claim 2, characterized in that, In step S3, the optimization solution for the magnetron sputtering process parameters includes: Step S31: Initialize the parameters of the multi-objective particle swarm optimization algorithm, including particle swarm size, maximum number of iterations, inertia weight, individual learning factor and social learning factor, and optimize the search space of magnetron sputtering process parameters. Step S32: Randomly generate an initial particle swarm. The position vector of each particle represents a set of magnetron sputtering process parameters, including sputtering temperature, total sputtering power, and substrate bias voltage. Step S33: Input the position vector of each particle into the trained performance prediction model to obtain the predicted values ​​of the corresponding corrosion layer thickness, hardness and elastic modulus, and calculate the fitness value of each particle. Step S34: Based on the Pareto dominance relationship of the multi-objective optimization problem, update the individual optimal position of each particle and the global optimal position set of the particle swarm, and maintain an external archive set to store the currently found non-dominated solutions. Step S35: Update the velocity and position of all particles according to the velocity update formula and position update formula of the particle swarm optimization algorithm; Step S36: Repeat steps S33 to S35 until the maximum number of iterations is reached or the convergence condition is met. Finally, select an optimal combination of magnetron sputtering process parameters from the external archive set as the optimization result.

4. The method for preparing a high-entropy alloy corrosion-resistant coating based on machine learning optimization according to claim 3, characterized in that, In step S35, the velocity update formula of the particle swarm optimization algorithm is: v_id(k+1)=w×v_id(k)+c1×r1×(pbest_id-x_id(k))+c2×r2×(gbest_id-x_id(k)); Where v_id(k+1) represents the updated velocity of the i-th particle in the d-th dimension; v_id(k) represents the current velocity of the i-th particle in the d-th dimension; w is the inertia weight; c1 and c2 are the individual learning factor and the social learning factor, respectively; r1 and r2 are random numbers between [0,1]; pbest_id represents the individual historical best position of the i-th particle in the d-th dimension; gbest_id represents the global best position of the particle swarm in the d-th dimension; and x_id(k) represents the current position of the i-th particle in the d-th dimension. The position update formula is: x_id(k+1)=x_id(k)+v_id(k+1); Where x_id(k+1) represents the position of the i-th particle after updating in the d-th dimension.

5. The method for preparing a high-entropy alloy corrosion-resistant coating based on machine learning optimization according to claim 4, characterized in that, In step S36, the convergence condition is that the number of non-dominated solutions in the external archive set no longer changes in 10 consecutive iterations, or the improvement of all objective functions is less than a preset threshold. Wherein, the improvement magnitude of all objective functions is the relative rate of change of the hypervolume, calculated by the following formula: H = |HV(k) - HV(k-1)| / HV(k-1); Where H is the relative rate of change of the hypervolume, HV(k) is the hypervolume of the current iteration, and HV(k-1) is the hypervolume of the previous iteration.

6. The method for preparing a high-entropy alloy corrosion-resistant coating based on machine learning optimization according to claim 5, characterized in that, In step S36, the final selection of an optimal combination of magnetron sputtering process parameters from the external archive set as the optimization result includes: A sorting method based on crowding distance is used to filter the non-dominated solutions in the final external archive set, and the solution located in the central region of the Pareto front is selected as the final combination of optimized process parameters.

7. The method for preparing a high-entropy alloy corrosion-resistant coating based on machine learning optimization according to claim 6, characterized in that, In step S4, the preparation of the AlCrFeMoTi high-entropy alloy coating using magnetron sputtering five-target co-deposition technology includes: Step S41: Select Al, Cr, Fe, Mo, or Ti metal targets as sputtering sources and ferritic / martensitic steel as matrix materials. Step S42 involves pre-treating the substrate material, including mechanical polishing, ultrasonic cleaning, and argon ion bombardment cleaning in a vacuum chamber. Step S43: Based on the optimized combination of process parameters, set the sputtering temperature, total sputtering power, and substrate bias voltage; Step S44: In an ultra-high vacuum environment, high-purity argon gas is introduced as the working gas, and the five-target co-deposition system is started. Magnetron sputtering deposition is performed under the set process parameters to form an AlCrFeMoTi high-entropy alloy coating on the substrate.

8. The method for preparing a high-entropy alloy corrosion-resistant coating based on machine learning optimization according to claim 7, characterized in that, In step S5, the corrosion resistance test of the prepared coating includes: Step S51: The cross-sectional morphology of the coating is observed using a scanning electron microscope, and the thickness of the corrosion layer formed after the experiment in a liquid lead-bismuth environment is measured. Step S52: The hardness and elastic modulus of the coating are measured using nanoindentation technology; Step S53: Compare the experimentally measured corrosion layer thickness, hardness, and elastic modulus with the predicted values ​​of the performance prediction model under the optimal process parameters, and calculate the relative error. Step S54: If the relative error between the predicted and experimental values ​​of all performance indicators is less than the preset relative error, the model is determined to be accurate; otherwise, the experimental data is added to the prepared dataset, the model is retrained and optimized, and iterative improvements are made.

9. The method for preparing a high-entropy alloy corrosion-resistant coating based on machine learning optimization according to claim 8, characterized in that, In step S54, experimental data is added to the prepared dataset, the model is retrained and optimized, and iterative improvements are made, including: Step S541: Add the process parameters and performance data obtained in this preparation and testing as new samples to the preparation dataset to form an amplification dataset; Step S542: Based on the augmented dataset, re-execute the data preprocessing, model training and validation steps to obtain the updated performance prediction model; Step S543: Based on the updated performance prediction model, re-execute the multi-objective particle swarm optimization algorithm to solve for the new optimal combination of process parameters; Step S544: Based on the new optimal combination of process parameters, prepare the coating again and conduct corrosion resistance tests, and verify the accuracy of the model predictions; Step S545: Repeat the above iterative steps until the relative errors between the predicted and experimental values ​​of all performance indicators meet the preset accuracy requirements.

10. A high-entropy alloy corrosion-resistant coating applied to the high-entropy alloy corrosion-resistant coating prepared by the machine learning-optimized method described in claims 1-9, characterized in that, include: The high-entropy alloy coating is a five-element system of aluminum, chromium, iron, molybdenum, and titanium, with the following mass fractions for each element: aluminum 15%-25%, chromium 20%-30%, iron 15%-25%, molybdenum 10%-20%, and titanium 10%-20%.

Images