Method for designing Al-Cu series aluminum alloy with high mechanical property based on machine learning
By designing Al-Cu aluminum alloys using machine learning and optimizing the alloy composition using PSO-BP neural networks and non-dominated sorting genetic algorithms, the problems of long cycle and high cost of traditional design methods are solved, and high-performance aluminum alloy materials with high mechanical properties are efficiently prepared for application in aerospace and transportation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-25
- Publication Date
- 2026-03-31
AI Technical Summary
Existing Al-Cu aluminum alloy design methods are time-consuming, costly, and difficult to design materials with high mechanical properties.
A machine learning-based approach was adopted. By establishing an original dataset, a PSO-BP neural network model was trained and optimized, and a non-dominated sorting genetic algorithm was combined to select aluminum alloy compositions that met the conditions. High-mechanical-performance Al-Cu aluminum alloys were then prepared using a squeeze casting method.
It significantly improves the R&D efficiency of Al-Cu aluminum alloys, enabling the faster and lower-cost discovery of materials with excellent properties, suitable for fields such as aerospace and transportation.
Smart Images
Figure CN121768549A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method for designing Al-Cu based aluminum alloys. Background Technology
[0002] After heat treatment, Al-Cu aluminum alloys undergo age hardening, significantly improving their strength and hardness. Further introduction of Mg and Zn elements allows Mg to combine with Cu to form the S-Al₂CuMg phase, which exhibits superior thermal stability compared to the θ phase. The dispersed distribution of the S phase hinders dislocation movement, effectively enhancing toughness and high-temperature stability. Adding trace elements such as Sc and Zr further suppresses recrystallization, promoting θ' phase precipitation while inhibiting its growth, thus refining the θ' precipitate and improving the mechanical properties of the Al-Cu alloy. Due to its excellent comprehensive mechanical properties, Al-Cu alloys are widely used in aerospace, transportation, and other fields. To meet the demands of integration, lightweighting, and rapid response, further improvement of its mechanical properties is crucial.
[0003] Current design methods for Al-Cu aluminum alloys largely rely on traditional trial-and-error approaches, involving continuous experimentation and error correction to discover Al-Cu alloys with superior performance. However, this method is time-consuming, costly, and struggles to produce materials with high mechanical properties. Against this backdrop, artificial intelligence (AI) technology is demonstrating immense application potential. With the development of machine learning, artificial neural networks can quickly learn the relationship between material composition and properties under the same processing conditions and can predict material properties relatively well. Summary of the Invention
[0004] The present invention aims to address the technical problems of existing methods for designing Al-Cu aluminum alloys, which are characterized by long cycles, high costs, and difficulty in designing materials with high mechanical properties. Instead, it provides a method for designing Al-Cu aluminum alloys with high mechanical properties based on machine learning.
[0005] The method for designing high-mechanical-performance Al-Cu aluminum alloys based on machine learning according to the present invention is carried out according to the following steps:
[0006] S1. Establish the original dataset for Al-Cu alloy materials:
[0007] The original dataset was compiled by collecting multiple sets of different aluminum alloy compositions and their corresponding mechanical properties from various publicly available data sources.
[0008] The aluminum alloy composition elements include Al, Cu, Mg, Zn, Mn, Ti, Cr, Si, Fe, Sc, and Zr; the mechanical properties include yield strength, tensile strength, and elongation.
[0009] S2: Training the PSO-BP neural network model based on the original dataset:
[0010] 80% of the original dataset in S1 is divided into a training set and 20% into a test set. The PSO-BP neural network model is trained using the training set, and the trained model is used to predict the performance of the test set. Specifically, during training, the composition of aluminum alloy is used as input, and regression prediction is performed through the BP neural network. The final output is the yield strength, tensile strength and elongation of the aluminum alloy material.
[0011] The PSO-BP neural network model is a backpropagation neural network model optimized by the particle swarm optimization algorithm. The particle swarm optimization algorithm is used to optimize the initial weights and thresholds of the backpropagation neural network. The backpropagation neural network includes an input layer, at least one hidden layer and an output layer, wherein the nodes of the input layer correspond to the composition of the aluminum alloy and the nodes of the output layer correspond to the mechanical performance indicators to be predicted.
[0012] S3: The model obtained in S2 is used as the fitness function and embedded into the non-dominated sorting genetic algorithm. After setting the range and step size of each element in the aluminum alloy, the optimal Pareto front is obtained. Then, in the optimal Pareto front, the aluminum alloy composition that meets the requirements of mechanical performance under actual working conditions is selected.
[0013] This invention collects data on three types of mechanical properties—yield strength, tensile strength, and elongation—based on existing literature and establishes a database. It then uses machine learning to design Al-Cu aluminum alloys with high predictive ability. Compared to the traditional "trial and error" method, this invention's method can not only find more new aluminum alloy materials with excellent properties, but also greatly reduce manpower, financial resources, and time costs.
[0014] This invention provides a dataset that meets the prediction requirements, has universality, and the trained model can accurately predict the mechanical properties of Al-Cu aluminum alloys, thereby improving the R&D efficiency of this series of aluminum alloys and having broad application prospects. Attached Figure Description
[0015] Figure 1 The machine learning-based design flowchart of this invention;
[0016] Figure 2 The scatter plot shows the model performance when predicting mechanical properties using the PSO-BP neural network model in S2 of Experiment 1.
[0017] Figure 3 The Pareto optimal frontier obtained by the NSGA-II algorithm in S3 of Experiment 1;
[0018] Figure 4 The stress-strain curves are for six different aluminum alloys prepared in S4 of Experiment 1. Detailed Implementation
[0019] Specific Implementation Method 1: This implementation method is a method for designing high-mechanical-performance Al-Cu aluminum alloys based on machine learning, specifically carried out according to the following steps:
[0020] S1. Establish the original dataset for Al-Cu alloy materials:
[0021] The original dataset was compiled by collecting multiple sets of different aluminum alloy compositions and their corresponding mechanical properties from various publicly available data sources.
[0022] The aluminum alloy composition elements include Al, Cu, Mg, Zn, Mn, Ti, Cr, Si, Fe, Sc, and Zr; the mechanical properties include yield strength, tensile strength, and elongation.
[0023] S2: Training the PSO-BP neural network model based on the original dataset:
[0024] 80% of the original dataset in S1 is divided into a training set and 20% into a test set. The PSO-BP neural network model is trained using the training set, and the trained model is used to predict the performance of the test set. Specifically, during training, the composition of aluminum alloy is used as input, and regression prediction is performed through the BP neural network. The final output is the yield strength, tensile strength and elongation of the aluminum alloy material.
[0025] The PSO-BP neural network model is a backpropagation neural network model optimized by the particle swarm optimization algorithm. The particle swarm optimization algorithm is used to optimize the initial weights and thresholds of the backpropagation neural network. The backpropagation neural network includes an input layer, at least one hidden layer and an output layer, wherein the nodes of the input layer correspond to the composition of the aluminum alloy and the nodes of the output layer correspond to the mechanical performance indicators to be predicted.
[0026] S3: The model obtained in S2 is used as the fitness function and embedded into the non-dominated sorting genetic algorithm. After setting the range and step size of each element in the aluminum alloy, the optimal Pareto front is obtained. Then, in the optimal Pareto front, the aluminum alloy composition that meets the requirements of mechanical performance under actual working conditions is selected.
[0027] Specific Implementation Method Two: This implementation method differs from Specific Implementation Method One in that the training conditions described in S2 include: when predicting yield strength, the machine learning model randomly divides the original dataset into a training set and a test set, with a ratio of 8:2; the number of hidden levels is set to 16-18; the number of iterations is 3000-6000; the training function is the logsig function; the population size in the particle swarm optimization algorithm is set to 80-150 groups; and the number of evolutions is 200-300. Everything else is the same as in Specific Implementation Method One.
[0028] Specific Implementation Method Three: This implementation method differs from Specific Implementation Method One or Two in that the training conditions described in S2 include: when predicting tensile strength, the machine learning model randomly divides the original dataset into a training set and a test set, with a ratio of 8:2; the number of hidden layers is set to 16-18; the number of iterations is 3000-6000; the training function is the logsig function; the population size in the particle swarm optimization algorithm is set to 80-150 groups; and the number of evolutions is 200-300. Everything else is the same as in Specific Implementation Method One or Two.
[0029] Specific Implementation Method Four: This implementation method differs from Specific Implementation Methods One to Three in that the training conditions described in S2 include: when predicting the elongation rate, the machine learning model randomly divides the data into a training set and a test set, with a ratio of 8:2; the number of hidden layers is set to 12-14; the number of iterations is 2000-5000; the training function is the tansig function; the population size in the particle swarm optimization algorithm is set to 80-150 groups; and the number of evolutions is 200-300. Everything else is the same as in Specific Implementation Methods One to Three.
[0030] Specific Implementation Method Five: This implementation method differs from Specific Implementation Method Four in that: a scatter plot of experimental measurement and model prediction is plotted based on the original dataset in S1 and the model prediction obtained from S2. The scatter plot uses the measured mechanical properties of aluminum alloy in the original dataset as the x-axis and the predicted mechanical properties of aluminum alloy by the machine learning model as the y-axis. The plot is then calculated based on the coefficient of determination R... 2 The root mean square error (RMSE) determines the accuracy of the model's prediction of aluminum alloy properties. Everything else is the same as in Specific Implementation Method Four.
[0031] Specific Implementation Method Six: This implementation method differs from Specific Implementation Method Five in that the non-dominated sorting genetic algorithm described in S3, considering the continuous and limited range of aluminum alloy composition variables, employs a simulated binary crossover operator. Based on the predicted performance time, a specific population size and maximum generation number are set. Its collaborative optimization process involves using yield strength, tensile strength, and elongation as optimization objectives. It maintains the diversity of the solution set by rapidly decomposing the non-dominated sorting region into a hierarchy and combining it with crowding calculations, ultimately obtaining a uniformly distributed Pareto optimal front and outputting multiple sets of non-dominated solutions. Everything else is the same as in Specific Implementation Method Five.
[0032] Specific Implementation Method Seven: This implementation method differs from Specific Implementation Method Six in that the elemental composition range of the aluminum alloy described in S3 is as follows: Cu 2~8wt%, Mg 0~3wt%, Zn 0~3wt%, Mn 0~2wt%, Ti 0~1wt%, Cr 0~1wt%, Si 0~0.2wt%, Fe 0~0.2wt%, Sc 0~0.5wt%, Zr 0~0.5wt%, with the balance being aluminum. Everything else is the same as in Specific Implementation Method Six.
[0033] Specific Implementation Method Eight: This implementation method differs from Specific Implementation Method Seven in that the step size described in S3 is 0.05wt%. Everything else is the same as in Specific Implementation Method Seven.
[0034] Specific Implementation Method Nine: This implementation method differs from Specific Implementation Method Eight in that, in S3, after selecting an aluminum alloy composition that meets the requirements for mechanical properties under actual working conditions at the optimal Pareto front, the aluminum alloy is prepared by extrusion casting. Everything else is the same as in Specific Implementation Method Eight.
[0035] Specific Implementation Method Ten: This implementation method differs from Specific Implementation Method Nine in that the process of preparing aluminum alloy by extrusion casting is as follows:
[0036] S1: Pure aluminum and master alloy are weighed according to the weight fraction of each element in the aluminum alloy, while iron, as an impurity element, is not actively added through the master alloy.
[0037] The intermediate alloys are Al-10Si, Al-50Cu, Al-10Mg, Al-10Zn, Al-10Mn, Al-10Ti, Al-10Cr, Al-2Sc, and Al-5Zr;
[0038] S2: Place the above raw materials in a drying oven to remove moisture;
[0039] The drying temperature is 150℃ and the time is 4 hours;
[0040] S3: Melt the raw materials to obtain aluminum alloy molten liquid;
[0041] The smelting process is as follows: pure aluminum, Al-10Cr and Al-2Sc master alloys are melted at 800℃~850℃, and then Al-10Si, Al-50Cu, Al-5Zr and Al-10Mn master alloys are added in sequence. After all the alloys are melted, the temperature is held for 20 minutes to obtain the first smelting liquid. The temperature is then lowered to 740℃~760℃, and then Al-10Mg, Al-10Zn and Al-10Ti master alloys wrapped in aluminum foil are added. Argon gas is introduced above the smelting liquid for protection. After all the alloys are melted, the temperature is held for 30 minutes to obtain the second smelting liquid.
[0042] S4: Stir the secondary smelting liquid at a uniform speed of 60 r / min for 5 min using a graphite rod. After stirring, lower the temperature to 710℃~730℃ and hold for 20 min. Insert the ultrasonic amplitude transformer 5cm~7cm below the surface of the smelting liquid and start ultrasonication at a frequency of 20kHz~22kHz and a power of 2500W. Introduce argon gas at a rate of 0.6L / min~0.8L / min to remove hydrogen from the smelting liquid. The treatment time is 1 min~2 min. Remove the surface scum, raise the temperature to 730℃ and hold for 30 min.
[0043] S5: After keeping the molten liquid obtained in S4 at a constant temperature for a period of time, quickly pour it into the preheated mold, close the mold and perform extrusion casting, hold the pressure for a period of time, demold and quench.
[0044] The process conditions described in S5 are: casting temperature of 730℃, mold preheating temperature of 250℃, extrusion pressure of 200MPa, holding time of 60s, and quenching in 50℃ warm water after holding.
[0045] S6: Perform solution treatment and artificial aging treatment on the castings obtained in S5;
[0046] The solution treatment temperature is set to 440~480℃ and held for 2~8h to form a supersaturated solid solution, followed by room temperature water quenching;
[0047] The aging treatment temperature is set to 190℃ and held for 2-8 hours to precipitate the nano-precipitate phase. After removal, it is air-cooled. Other aspects are the same as in specific implementation method nine.
[0048] The invention was verified using the following experiments:
[0049] Experiment 1: This experiment demonstrates a method for designing high-mechanical-performance Al-Cu aluminum alloys based on machine learning. The specific steps are as follows:
[0050] S1. Establish the original dataset for Al-Cu alloy materials:
[0051] The original dataset was compiled by collecting 399 different aluminum alloy compositions and their corresponding mechanical properties from various publicly available data sources.
[0052] The aluminum alloy composition elements include Al, Cu, Mg, Zn, Mn, Ti, Cr, Si, Fe, Sc, and Zr (Fe is an impurity); the mechanical properties include yield strength, tensile strength, and elongation.
[0053] S2: Training the PSO-BP neural network model based on the original dataset:
[0054] 80% of the original dataset in S1 is divided into a training set and 20% into a test set. The PSO-BP neural network model is trained using the training set, and the trained model is used to predict the performance of the test set. Specifically, during training, the composition of aluminum alloy is used as input, and regression prediction is performed through the BP neural network. The final output is the yield strength, tensile strength and elongation of the aluminum alloy material.
[0055] The PSO-BP neural network model is a backpropagation neural network model optimized by the particle swarm optimization algorithm. The particle swarm optimization algorithm is used to optimize the initial weights and thresholds of the backpropagation neural network. The backpropagation neural network includes an input layer, at least one hidden layer and an output layer, wherein the nodes of the input layer correspond to the composition of the aluminum alloy and the nodes of the output layer correspond to the mechanical performance indicators to be predicted.
[0056] The training conditions are as follows: When predicting yield strength, the machine learning model randomly divides the original dataset into a training set and a test set, with a ratio of 8:2. The number of hidden levels is set to 16-18, the number of iterations is 3000-6000, the training function is the logsig function, the number of populations in the particle swarm optimization algorithm is set to 80-150 groups, and the number of evolutions is 200-300.
[0057] When predicting tensile strength, the machine learning model randomly divides the original dataset into a training set and a test set with a ratio of 8:2. The number of hidden levels is set to 16-18, the number of iterations is 3000-6000, the training function is the logsig function, the number of populations in the particle swarm optimization algorithm is set to 80-150 groups, and the number of evolutions is 200-300.
[0058] When predicting the elongation rate, the machine learning model randomly divides the data into a training set and a test set, with a ratio of 8:2. The number of hidden layers is set to 12-14, the number of iterations is 2000-5000, the training function is the tansig function, the number of populations in the particle swarm optimization algorithm is set to 80-150 groups, and the number of evolutions is 200-300.
[0059] Based on the original dataset in S1 and the model predictions derived from S2, a scatter plot of experimental measurements versus model predictions was plotted. The x-axis represents the measured mechanical properties of the aluminum alloy from the original dataset, and the y-axis represents the predicted mechanical properties of the aluminum alloy from the machine learning model. The plot was then analyzed using the coefficient of determination R0. 2 The root mean square error (RMSE) determines the accuracy of the model's predictions of aluminum alloy properties, such as... Figure 2As shown, (a) is the training set for predicted yield strength; (b) is the test set for predicted yield strength; (c) is the training set for predicted tensile strength; (d) is the test set for predicted tensile strength; (e) is the training set for predicted elongation; and (f) is the test set for predicted elongation. It can be seen that after optimizing the network parameters of the BP neural network model using the particle swarm optimization algorithm, the model has a high coefficient of determination R0 on both the training and test sets. 2 The value indicates that the model has good predictive performance.
[0060] S3: Using the model obtained in S2 as the fitness function, embed it into the Non-Dominated Sorting Genetic Algorithm (NSGA-II). After setting the range of elemental composition and step size in the aluminum alloy, the optimal Pareto front is obtained, such as... Figure 3 As shown, YS is the yield strength, UTS is the tensile strength, and EL is the elongation.
[0061] The elemental composition range of the aluminum alloy is as follows: Cu 2~8wt%, Mg 0~3wt%, Zn 0~3wt%, Mn 0~2wt%, Ti 0~1wt%, Cr 0~1wt%, Si 0~0.2wt%, Fe 0~0.2wt%, Sc 0~0.5wt%, Zr 0~0.5wt%, with the balance being aluminum; the step size is 0.05wt%.
[0062] The non-dominated sorting genetic algorithm, designed for the continuous and limited range of aluminum alloy composition variables, employs a simulated binary crossover operator and sets a specific population size and maximum number of generations based on the predicted performance time. Its collaborative optimization process is as follows: using yield strength, tensile strength, and elongation as optimization objectives, it maintains the diversity of the solution set by rapidly decomposing the non-dominated sorting region into a hierarchy and combining it with crowding calculation, ultimately obtaining a uniformly distributed Pareto optimal front and outputting multiple sets of non-dominated solutions.
[0063] The requirement is an Al-Cu-Mg-Zn alloy with a yield strength greater than or equal to 320 MPa, a tensile strength greater than or equal to 400 MPa, and an elongation greater than or equal to 6%. The aluminum alloy composition that meets the conditions is selected from the optimal Pareto front, resulting in six aluminum alloy materials with different chemical compositions, as shown in Table 1 (Si and Fe are both 0).
[0064] Table 1. Aluminum alloy composition table designed based on machine learning.
[0065]
[0066] S4: The six aluminum alloys in Table 1 were prepared by squeeze casting. The preparation process is as follows:
[0067] 1): Weigh out pure aluminum and intermediate alloy according to the weight fraction of each element in the aluminum alloy, and introduce iron as an impurity;
[0068] The intermediate alloys are Al-50Cu, Al-10Mg, Al-10Zn, Al-10Mn, Al-10Ti, Al-10Cr, Al-2Sc, and Al-5Zr;
[0069] 2): Place the above raw materials in a drying oven to remove moisture;
[0070] The drying temperature is 150℃ and the time is 4 hours;
[0071] 3): The raw materials are smelted to obtain aluminum alloy molten liquid;
[0072] The smelting process is as follows: pure aluminum, Al-10Cr and Al-2Sc master alloys are melted at 850℃, and then the temperature is lowered to 780℃ and held for 10 minutes; then Al-50Cu, Al-5Zr and Al-10Mn master alloys are added in sequence, and after all are melted, the temperature is lowered to 740℃ and held for 20 minutes to obtain the first smelting liquid; then Al-10Mg, Al-10Zn and Al-10Ti master alloys wrapped in aluminum foil are added, and argon gas is introduced above the smelting liquid for protection. After all are melted, the temperature is held for another 30 minutes to obtain the second smelting liquid.
[0073] 4): Stir the secondary smelting liquid at a uniform speed of 60 r / min for 5 min using a graphite rod. After stirring, lower the temperature to 730℃ and hold for 20 min. Insert the ultrasonic amplitude transformer 5 cm below the surface of the smelting liquid and start ultrasonication at a frequency of 20 kHz and a power of 2500 W. Introduce argon gas at 0.8 L / min to remove hydrogen from the smelting liquid for 2 min. Remove the surface scum, raise the temperature to 730℃ and hold for 30 min.
[0074] 5): After keeping the molten liquid obtained in 4) at a constant temperature for a period of time, quickly pour it into the preheated mold, close the mold and perform extrusion casting, hold the pressure for a period of time, demold and quench.
[0075] The process conditions described in 5) are: casting temperature of 730℃, mold preheating temperature of 250℃, extrusion pressure of 200MPa, holding time of 60s, and quenching in 50℃ warm water after holding pressure.
[0076] 6): The castings obtained in 5) are subjected to solution treatment and artificial aging treatment;
[0077] The solution treatment temperature is set to 460~480℃ and held for 2~6 hours to form a supersaturated solid solution, followed by room temperature water quenching;
[0078] The aging treatment temperature was set to 190℃ and held for 4~8 hours to precipitate nano-precipitates. After removal, the precipitates were air-cooled.
[0079] The heat treatment parameters for the six aluminum alloys in Table 1 are shown in Table 2.
[0080] Stress-strain curves of six aluminum alloys with different compositions were plotted under optimal heat treatment parameters, such as... Figure 4 As shown, the corresponding yield strength, tensile strength, and elongation were obtained. The alloy properties designed in this invention are all good, indicating that the alloy composition predicted by machine learning is reliable. The experimental results in this experiment also prove the rationality of the design scheme of this invention and the efficiency of the model prediction, which has guiding significance for actual industrial production.
[0081] Table 2 Optimal T6 heat treatment parameters for each alloy combination
[0082]
Claims
1. A method for designing an Al-Cu based aluminum alloy with high mechanical properties based on machine learning, characterized in that The method is carried out in the following steps: S1, establishing the original data set of Al-Cu alloy material: Collect a plurality of different aluminum alloy composition groups and their corresponding mechanical property data groups through various public data to form an original data set; The aluminum alloy composition elements include Al, Cu, Mg, Zn, Mn, Ti, Cr, Si, Fe, Sc and Zr; the mechanical properties include yield strength, tensile strength and elongation; S2: training of the PSO-BP neural network model based on the original data set: Divide 80% of the original data set in S1 into a training set and 20% into a test set, train the PSO-BP neural network model using the training set, and use the trained model to predict the performance of the test set composition, as follows: during training, the aluminum alloy composition is taken as input, regression prediction is performed through the BP neural network, and finally the yield strength, tensile strength and elongation of the aluminum alloy material are output; The PSO-BP neural network model is a back propagation neural network model optimized based on a particle swarm algorithm, and the particle swarm algorithm is used to optimize the initial weights and thresholds of the back propagation neural network; the back propagation neural network comprises an input layer, at least one hidden layer and an output layer, wherein the nodes of the input layer correspond to the aluminum alloy composition, and the nodes of the output layer correspond to the mechanical performance indicators to be predicted; S3: the model obtained in S2 is embedded into a non-dominated sorting genetic algorithm as a fitness function, the element composition range and step size in the aluminum alloy are set, and the optimal Pareto front is obtained; then, according to the requirements of the actual working conditions for the mechanical properties, the aluminum alloy composition that meets the conditions is selected from the optimal Pareto front.
2. The method for designing Al-Cu series aluminum alloy with high mechanical performance based on machine learning according to claim 1, characterized in that The training conditions in S2 include: when predicting the yield strength, the machine learning model randomly divides the original data set into a training set and a test set, the ratio of the training set to the test set is 8:2, the number of hidden layers is set to 16-18, the number of iterations is 3000-6000 times, the training function is selected as logsig function, the number of populations in the particle swarm algorithm is set to 80-150 groups, and the number of evolutions is 200-300 times.
3. The method for designing Al-Cu series aluminum alloy with high mechanical performance based on machine learning according to claim 2, characterized in that The training conditions in S2 include: when predicting the tensile strength, the machine learning model randomly divides the original data set into a training set and a test set, the ratio of the training set to the test set is 8:2, the number of hidden layers is set to 16-18, the number of iterations is 3000-6000 times, the training function is selected as logsig function, the number of populations in the particle swarm algorithm is set to 80-150 groups, and the number of evolutions is 200-300 times.
4. The method for designing Al-Cu series aluminum alloy with high mechanical properties based on machine learning according to claim 3, characterized in that The training conditions in S2 include: when predicting the elongation, the machine learning model randomly divides the data into a training set and a test set, wherein the ratio of the training set to the test set is 8:2, the number of hidden layers is set to 12-14, the number of iterations is 2000-5000 times, the training function is selected as tansig function, the number of populations in the particle swarm algorithm is set to 80-150 groups, and the number of evolutions is 200-300 times.
5. The method for designing Al-Cu series aluminum alloy with high mechanical properties based on machine learning according to claim 4, characterized in that According to the original data set in S1 and the model derived from S2, the experimental measurement-model prediction scatter plot is drawn, taking the aluminum alloy mechanical property measurement value in the original data set as the abscissa and the aluminum alloy mechanical property prediction value of the machine learning model as the ordinate, and determining the prediction accuracy of the model on the aluminum alloy performance according to the coefficient of determination R 2 and the root mean square error RMSE.
6. The method for designing Al-Cu series aluminum alloy with high mechanical performance based on machine learning according to claim 1, characterized in that The non-dominated sorting genetic algorithm described in S3 adopts a simulated binary crossover operator according to the continuous and range-limited characteristics of the aluminum alloy composition variable, and sets a specific population size and maximum generation number according to the time consumption of performance prediction, and the cooperative optimization process is as follows: taking the yield strength, tensile strength and elongation as the optimization objectives, the advantages and disadvantages of the decomposition are distinguished by fast non-dominated sorting, and the diversity of the solution set is maintained by combining the congestion calculation, and finally the uniformly distributed Pareto optimal frontier is obtained, and a plurality of non-dominated solutions are output.
7. The method for designing Al-Cu series aluminum alloy with high mechanical performance based on machine learning according to claim 6, characterized in that The element composition range of the aluminum alloy described in S3 is as follows: Cu is 2-8wt%, Mg is 0-3wt%, Zn is 0-3wt%, Mn is 0-2wt%, Ti is 0-1wt%, Cr is 0-1wt%, Si is 0-0.2wt%, Fe is 0-0.2wt%, Sc is 0-0.5wt%, Zr is 0-0.5wt%, and the balance is aluminum.
8. The method for designing Al-Cu series aluminum alloy with high mechanical performance based on machine learning according to claim 7, characterized in that The step size described in S3 is 0.05wt%.
9. The method for designing Al-Cu series aluminum alloy with high mechanical performance based on machine learning according to claim 8, characterized in that After selecting the aluminum alloy composition that meets the conditions in the optimal Pareto frontier according to the requirements of the actual working conditions for the mechanical properties in S3, the aluminum alloy is prepared by an extrusion casting method.
10. The method for designing Al-Cu series aluminum alloy with high mechanical performance based on machine learning according to claim 9, characterized in that The process of preparing the aluminum alloy by the extrusion casting method is as follows: S1: According to the weight fraction of each element in the aluminum alloy, pure aluminum and intermediate alloy are weighed, and iron is introduced as an impurity; The intermediate alloy is Al-10Si, Al-50Cu, Al-10Mg, Al-10Zn, Al-10Mn, Al-10Ti, Al-10Cr, Al-2Sc and Al-5Zr; S2: The above raw materials are placed in a drying box to remove moisture; The drying temperature is 150 DEG C, and the time is 4h; S3: The raw materials are melted to obtain an aluminum alloy molten liquid; The melting process is as follows: after the pure aluminum, Al-10Cr and Al-2Sc intermediate alloy are melted at 800 DEG C-850 DEG C, the Al-10Si, Al-50Cu, Al-5Zr and Al-10Mn intermediate alloy are sequentially added, and after all the melting, the temperature is maintained for 20min to obtain the first molten liquid; the temperature is reduced to 740 DEG C-760 DEG C, the Al-10Mg, Al-10Zn and Al-10Ti intermediate alloy wrapped in aluminum foil are added, and argon gas is introduced above the molten liquid for protection, and after all the melting, the temperature is maintained for 30min to obtain the second molten liquid; S4: The graphite rod is used to stir the second molten liquid at a uniform speed of 60r / min for 5min, after stirring, the temperature is reduced to 710 DEG C-730 DEG C and maintained for 20min; the ultrasonic amplitude bar is inserted into the molten liquid below the liquid surface by 5cm-7cm, the ultrasonic is started, the ultrasonic frequency is 20kHz-22kHz, the power is 2500W, and the argon gas is introduced at a flow rate of 0.6L / min-0.8L / min to remove hydrogen in the molten liquid, the treatment time is 1min-2min; remove the surface dross, heat to 730 DEG C and maintain for 30min; S5: The smelting liquid obtained in S4 is kept for a period of time and then rapidly poured into a preheated mold, the mold is closed for extrusion casting, the mold is opened after pressure maintaining for a period of time and then quenching is performed; The process conditions in S5 are as follows: the pouring temperature is 730 DEG C, the preheating temperature of the mold is 250 DEG C, the extrusion pressure is 200 MPa, the pressure maintaining time is 60 s, and the mold is placed in warm water at 50 DEG C for quenching after pressure maintaining; S6: The casting obtained in S5 is subjected to solid solution treatment and artificial aging treatment; The temperature for solid solution treatment is set to 440-480 DEG C and the over-saturated solid solution is formed after heat preservation for 2-8 h, and then room temperature water quenching is performed; The temperature for aging treatment is set to 190 DEG C and the nano precipitated phase is precipitated after heat preservation for 2-8 h, and then air cooling is performed after taking out.