Five-series aluminum alloy laser welding material component optimization method based on machine learning

Abstract

The invention relates to the technical field of welding material optimization design, in particular to a 5-series aluminum alloy laser welding material component optimization method based on machine learning, which comprises the following steps: on the basis of a 5-series aluminum alloy welding wire system, adding Ca, Ti and B elements to obtain a welding material component set for predicting the tensile strength and elongation of a laser welding joint; then, welding materials are prepared from corresponding components in the component set, laser welding and tensile testing are conducted, and a welding material component-joint strength / ductility data set is obtained; and then modeling is conducted on the two output characteristics, an agent model is constructed, component prediction is conducted through multiple optimization algorithms, a welding material is prepared from predicted optimized components, laser welding is conducted on the welding material, the mechanical property of a connector of the welding material is tested, and finally the aluminum alloy laser welding material component corresponding to the high-toughness welding connector is obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of welding material optimization design, and in particular to a method for optimizing the composition of 5-series aluminum alloy laser welding materials based on machine learning. Background Art

[0002] Laser welding is a key technology for the innovative upgrade of aluminum alloy car body manufacturing for high-speed trains, but the welding wire currently used follows the traditional arc welding material system, and the strength and toughness of the weld and the base material are not well matched. Optimizing the composition of the welding wire is a key method to achieve high strength and toughness matching between the weld and the base material. However, traditional welding material design methods are often based on trial and error, which is inefficient and time-consuming, and it is especially difficult to find better joint performance within a multi-component range. With the development of artificial intelligence, machine learning has begun to be widely used in the development of new materials. By establishing a relationship model between input features and target performance, rapid prediction of material composition and performance can be achieved, which provides ideas for designing new aluminum alloy welding wires. Summary of the invention

[0003] In view of the shortcomings and deficiencies of the prior art, the present invention provides a method for optimizing the composition of 5-series aluminum alloy laser welding materials based on machine learning, comprising the following steps:

[0004] S1: Construct a 5-series aluminum alloy welding material composition-laser welded joint mechanical properties dataset, and use the Latin hypercube sampling method to construct the welding material composition space. The joint strength and elongation of the welded joint corresponding to the welding material composition space are obtained, and the welding material composition-joint strength / elongation dataset is obtained;

[0005] S2: The welding material composition in the S1 data set is used as input data, and the strength and elongation of the laser welded joint are used as output data to construct input-output Kriging proxy models respectively;

[0006] S3: Combine the Kriging proxy model in S2 with the EGO multi-objective algorithm to carry out multi-objective optimization; use the multi-objective EI hypervolume criterion to obtain the improvement value of the target performance on the current Pareto frontier;

[0007] S4: Use the method in S3 to carry out multiple iterations, prepare the weld material for each iteration and perform laser welding, test the mechanical properties of the weld joint, add the updated weld material composition and joint mechanical properties after each iteration back into the data set, repeat S1 to S4, start a cyclic iteration, and obtain the optimized aluminum alloy laser welding material composition.

[0008] Furthermore, the number of data sets of welding material composition-joint strength / elongation described in S1 is 69 groups.

[0009] Furthermore, the data set is an initial data set of 69 rows x 7 columns, and the initial data set is randomly divided into a test set and a training set in a ratio of 2:5, wherein the test set is 20 groups and the training set is 49 groups.

[0010] Furthermore, the EGO multi-objective algorithms in S3 include KTA2, SMS-EGO, Par-EGO and EIM-EGO algorithms.

[0011] Furthermore, two different parameters CA and DA are constructed in the KTA2 algorithm to independently promote the convergence and diversity of the algorithm during the search process, and the values ​​of the CA and DA parameters are updated every 10 times.

[0012] Furthermore, the Par-EGO algorithm uses a nonlinear weight vector to integrate multiple objectives into a single objective, that is, formula (1)

[0013]

[0014] Where m is the number of target performance, m=2, λ i is the i-th value of the weight vector, is the i-th objective function normalized to [0, 1] according to the minimum and maximum values ​​of the target performance, ρ is a constant set to 0.05; x refers to the content corresponding to the input welding material composition, y is the objective function, and λ refers to the weight vector, λ = [1, 2].

[0015] Furthermore, the hypervolume improvement function in the hypervolume criterion described in S3 measures the improvement in target performance when the target vector is added to the initial data set, that is, formula (2)

[0016] I h (x) = H(S∪p) - H(S) (2);

[0017] Where S is the non-dominated front, H() is the hypervolume index of the non-dominated front; I h (x) is the Euler distance improvement function, and p is the target vector.

[0018] Beneficial Effects

[0019] The present invention provides a method for optimizing the composition of 5-series aluminum alloy laser welding materials based on machine learning. Compared with the traditional "trial and error" welding material development method, it can effectively reduce the design time and cost and improve the research and development efficiency; the present invention constructs a multi-objective EI improvement criterion, which can effectively search for improvement space and predict the optimized welding material composition for small sample space and low model accuracy. The welding material developed by the present invention can simultaneously improve the strength and plasticity of laser welded joints. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0021] In the figure:

[0022] Figure 1 It is a schematic diagram of the welding material optimization algorithm framework of the present invention;

[0023] Figure 2 This is a schematic diagram of the accuracy of the Kriging model used in the present invention. DETAILED DESCRIPTION

[0024] The following will be combined with the embodiments 1-5 and the attached Figures 1-2 , the technical solution of the present invention is clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] Example 1

[0026] S1: First, a dataset of 5-series aluminum alloy welding material composition and mechanical properties of laser welded joints was constructed: Based on the 5-series commercial welding material system and experimental results, a Latin hypercube sampling method was used to obtain a dataset for predicting the strength and plasticity of aluminum alloy laser welded joints; based on the Al-Mg binary aluminum alloy welding material system, 69 components (including six elements, Al, Mg, Mn, Ti, Ca, and Y) and their corresponding laser welded joint strength and elongation data were collected;

[0027] S2: The chemical composition (including six elements, Al, Mg, Mn, Ti, Ca, and Y) in the S1 data set is used as input data, and the strength and elongation of laser welded joints are used as output data. Multiple input-output Kriging proxy models are constructed respectively. The input features are the welding material composition, which is mainly 5 components, and the output features are the joint mechanical properties, including strength and elongation as target features, to form an initial data set; the initial data set is divided into a test set and a training set in a ratio of 2:5 and randomly allocated, with 20 test sets and 49 training sets; the Kriging model provides the predicted value and prediction error of the non-dominated solution, and the multi-objective EI criterion is constructed based on the prediction function and the error function; Figure 2 As shown, Figure 2For the model accuracy, it can be seen that the model accuracy of tensile strength and elongation is low;

[0028] S3: Combine the Kriging proxy model in S2 with four EGO multi-objective algorithms (KTA2, SMS-EGO, Par-EGO and EIM-EGO algorithms) to carry out multi-objective optimization; use the multi-objective EI hypervolume criterion to measure the improvement of the target performance on the current Pareto frontier;

[0029] In the KTA2 algorithm, two different parameters, CA and DA, are constructed to independently promote the convergence and diversity of the algorithm in the search process. The values ​​of CA and DA parameters are updated every 10 times. Par-EGO uses a nonlinear weight vector to integrate multiple objectives into a single objective, that is,

[0030]

[0031] Wherein, m is the number of target performance, and the target performance in the present invention is the tensile strength and elongation of the joint, m=2, and ρ is a small positive constant, which is set to 0.05 in the present invention;

[0032] x refers to the content corresponding to the input welding material composition, y is the objective function, in the present invention, the welding material composition to be designed is marked as Al-x1Mg-x2Mn-x3Ca-x4Ti-x5Y, x is a vector composed of five components, x = [x1, x2, x3, x4, x5], x1∈[1.6~7.0], x2∈[0.05~0.4], x3∈[0~0.15], x4∈[0.02~0.5], x5∈[0~0.5], y is the target vector of the unknown point x y = [y1(x), y2(x)], λ refers to the weight vector, λ = [1,2];

[0033] The update points obtained by the four EGO algorithms are added to the initial data set to start a new iteration. In each iteration, the hypervolume improvement function is used to measure the improvement in target performance when the target vector is added to the initial data set, that is:

[0034] I h (x) = H(S∪p) - H(S) (2);

[0035] Where S is the non-dominated front, H() is the hypervolume index of the non-dominated front; I h (x) is the Euler distance improvement function, which represents the minimum value of the Euler distance from the target vector [y1(x), y2(x)] of the unknown point x to the non-dominated point in front of the eye; p in the formula is the target vector, that is, y = [y1(x), y2(x)], x = [x1, x2, x3, x4, x5], h is the dimension of the unknown point x, and h = [1, 5] in the present invention;

[0036] Among them, the larger the value of the hypervolume index is, the more the target performance improvement obtained by the EGO algorithm is;

[0037] S4: Use the method in S3 to carry out multiple iterations, prepare the welding materials for each iteration and perform laser welding, test the mechanical properties of the welded joint, add the updated welding material composition and joint mechanical properties after each iteration back into the data set and repeat S1 to S4 to start the cyclic iteration. According to the data in Table 1, the solution quality obtained by the KTA2 algorithm in the first three iterations is the best, and the quality of the EIM algorithm solution is the best in the fourth iteration. The welding material compositions obtained by the KTA2 algorithm and the EIM algorithm in the fourth iteration are selected to prepare welding wires, and laser welding and tensile performance tests are performed to obtain the welding material compositions and their corresponding joint mechanical properties.

[0038] The welding material composition optimized in S4 was used for welding material preparation and laser welding, and the microstructure and mechanical properties of the joint were analyzed. The initial 69 data sets plus the sum of the data after four iterations, welding material composition-joint strength / elongation, a total of 85 data sets were analyzed for correlation. The correlation analysis was performed using the heat map analysis function in the ChiPlot software.

[0039] The hypervolume index after four iterations in Example 1 is shown in Table 1.

[0040] Table 1 Hypervolume index (HV) after four EGO algorithm iterations

[0041]

[0042] The welding material composition obtained in Example 1 and its corresponding joint mechanical properties are shown in Table 2

[0043] Table 2 Optimized welding material composition and joint mechanical properties

[0044]

[0045] The present invention is aimed at welding materials specially used for laser welding. At present, arc welding materials are still used for laser welding. Due to the violent metallurgical reaction of laser welding, serious evaporation and burning of elements, and extremely non-equilibrium solidification, arc welding materials are used for laser welding. The joint strength and plasticity are much lower than the parent material. There is still little research and development of laser welding materials at home and abroad. Therefore, the present invention aims at this problem and conducts research on welding materials specially used for laser welding. The machine learning algorithm used in the present invention can simultaneously improve the strength and elongation of the welded joint, which is greatly improved compared with the existing literature. The present invention uses machine learning for the research and development of laser welding materials, which greatly shortens the research and development time and saves costs, which is difficult to achieve using the traditional "trial and error method".

[0046] The 5-series aluminum alloy welding material optimized by the present invention is used for laser welding, and its joint strength coefficient reaches 84.6%, and its elongation reaches 9.2%, while the joint strength coefficient of existing references and commercial welding wire welding is only about 70% to 75%, and the elongation is only 5% to 6%. The use of the present invention can greatly improve the strength and elongation of the welding joint. In addition, the welding materials of commercial arc welding are still commonly used in laser welding, resulting in the lack of special welding materials for laser welding. The welding materials in the present invention are only for laser welding and are special welding materials for laser welding.

Claims

1. A method for optimizing the composition of 5-series aluminum alloy laser welding materials based on machine learning, characterized in that: The following steps are involved: S1: Construct a 5-series aluminum alloy welding material composition-laser welded joint mechanical properties dataset, and use the Latin hypercube sampling method to construct the welding material composition space. The joint strength and elongation of the welded joint corresponding to the welding material composition space are obtained, and the welding material composition-joint strength / elongation dataset is obtained; S2: The welding material composition in the S1 data set is used as input data, and the strength and elongation of the laser welded joint are used as output data to construct input-output Kriging proxy models respectively; S3: Combine the Kriging proxy model in S2 with the EGO multi-objective algorithm to carry out multi-objective optimization; use the multi-objective EI hypervolume criterion to obtain the improvement value of the target performance on the current Pareto frontier; S4: Use the method in S3 to carry out multiple iterations, prepare the weld material for each iteration and perform laser welding, test the mechanical properties of the weld joint, add the updated weld material composition and joint mechanical properties after each iteration back into the data set, repeat S1 to S4, start a cyclic iteration, and obtain the optimized aluminum alloy laser welding material composition.

2. The method for optimizing the composition of 5-series aluminum alloy laser welding materials based on machine learning according to claim 1 is characterized in that: The number of data sets of welding material composition-joint strength / elongation described in S1 is 69 sets.

3. The method for optimizing the composition of 5-series aluminum alloy laser welding materials based on machine learning according to claim 2 is characterized in that: The data set is an initial data set of 69 rows x 7 columns, and the initial data set is randomly divided into a test set and a training set in a ratio of 2:5, wherein the test set is 20 groups and the training set is 49 groups.

4. The method for optimizing the composition of 5-series aluminum alloy laser welding materials based on machine learning according to claim 1 is characterized in that: The EGO multi-objective algorithms in S3 include KTA2, SMS-EGO, Par-EGO and EIM-EGO algorithms.

5. The method for optimizing the composition of 5-series aluminum alloy laser welding materials based on machine learning according to claim 4 is characterized in that: Two different parameters CA and DA are constructed in the KTA2 algorithm to independently promote the convergence and diversity of the algorithm during the search process, and the values ​​of CA and DA parameters are updated every 10 times.

6. The method for optimizing the composition of 5-series aluminum alloy laser welding materials based on machine learning according to claim 4 is characterized in that: The Par-EGO algorithm uses a nonlinear weight vector to integrate multiple objectives into a single objective, that is, formula (1) Where m is the number of target performance, m=2, λi is the i-th value of the weight vector, is the i-th objective function normalized to [0, 1] according to the minimum and maximum values ​​of the target performance, ρ is a constant set to 0.05; x refers to the content corresponding to the input welding material composition, y is the objective function, and λ refers to the weight vector, λ=[1,2].

7. The method for optimizing the composition of 5-series aluminum alloy laser welding materials based on machine learning according to claim 1, characterized in that: The hypervolume improvement function in the hypervolume criterion described in S3 measures the improvement in target performance when the target vector is added to the initial data set, that is, formula (2) Ih(x)=H(Sp)-H(S)(2); Where S is the non-dominated front, H() is the hypervolume index of the non-dominated front; I h (x) is the Euler distance improvement function, and p is the target vector.

Images