A method for characterizing the temperature field in friction stir welding based on the least squares support vector machine algorithm

Abstract

The present invention belongs to the field of stir friction welding temperature detection, and relates to a stir friction welding temperature field characterization method based on a least squares support vector machine algorithm. The present invention uses a method combining experiments and theories to conduct a stir friction welding single-axis shoulder experiment, sets characteristic points at different distances from the center of the weld, and arranges thermocouples symmetrically on the forward side and the backward side along the weld. The least squares support vector machine algorithm is trained and tested using the peak temperature of the characteristic points obtained, and the temperature field distribution detection of stir friction welding is realized. The temperature field at different distances from the center of the weld during the welding process is characterized, thereby improving measurement efficiency and reducing research costs. The present invention can realize temperature field output, effectively solving the problem that the temperature field cannot be fully obtained in the experiment.

Description

Technical Field

[0001] The invention belongs to the field of friction stir welding temperature detection, and relates to a temperature field characterization method based on a friction stir welding process, and specifically to a 2219 aluminum alloy friction stir welding temperature characterization method using a thermocouple and a least squares support vector machine algorithm. Background Art

[0002] Heavy rockets are an important standard for measuring a country's aerospace level. At present, my country's CZ9 heavy rocket has entered the "deep control" stage. One of the core tasks is to assemble and manufacture super-large tanks using welding technology. 2219 aluminum alloy has good high-temperature mechanical properties and welding properties, high fracture toughness, and good stress corrosion resistance, which meet the performance requirements of tanks. Conventional fusion welding technology will produce defects such as holes, thermal cracks, and burns, which will reduce the mechanical properties of the material. Stir friction welding technology is a solid phase welding technology that uses the heat generated by the friction between the high-speed rotating welding tool and the weldment to make the welded material flow plastically, and form a weld through the extrusion of the shaft shoulder and the stirring needle. Unlike fusion welding technology, the temperature of stir friction welding is kept below the melting point of the weldment, the heat input is appropriate, the thermal stress is small, and the mechanical properties of the weldment are good. Therefore, stir friction welding technology is widely used in the welding process of 2219 aluminum alloy.

[0003] During the friction stir welding process, the material undergoes severe plastic deformation, the heat generation and heat transfer mechanisms are complex, and there are differences in the temperature fields of different welding areas. The temperature field distribution has an important influence on the welding quality. If the temperature is too low, defects will be caused. If the temperature gradient is too large, the microstructure morphology and performance will be different. The above factors will lead to a decrease in the connection strength. In order to understand the influence of different process parameters and temperatures on the welding quality, it is necessary to accurately and comprehensively obtain the temperature field distribution law during the friction stir welding process. At present, there are mainly the following methods for the temperature field detection of friction stir welding: simulation method, infrared thermal imaging method and thermocouple method. The heat generation mechanism and material flow mechanism of friction stir welding are not clear. The simulation method cannot simulate the welding process from a mechanistic perspective, and the accuracy is limited. It needs to be continuously verified by other methods. The infrared thermal imaging method can only measure the temperature of the welding surface and is affected by environmental factors such as light and reflectivity. The thermocouple is in direct contact with the temperature measurement area and generates a temperature signal in real time, which can realize efficient and accurate temperature detection during welding.

[0004] In the paper "Temperature Field of Double Shoulder Friction Stir Welding of 6082-T6 Aluminum Alloy Thin Plate" published in 2019 by Li Yupeng et al. from Jilin University, 8 characteristic points were symmetrically set on the forward and backward sides of the weld, and K-type armored thermocouples were arranged from the inside of the shoulder to the outside of the shoulder. The temperature field distribution law at different distances from the center of the weld and the influence of different process parameters on the temperature field were obtained through welding experiments. The above temperature measurement method provides a reference for stir friction welding temperature detection, but the temperature measurement method based on thermocouples can only obtain the temperature curve of fixed characteristic points, and cannot obtain the complete temperature field distribution at different distances from the center of the weld. Therefore, existing research cannot obtain the complete temperature field and the influence of temperature on the microstructure of different regions. Summary of the invention

[0005] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art, invent a temperature field characterization technology based on the stir friction welding process, introduce the least squares support vector machine algorithm into the temperature field detection process, realize the complete acquisition of the temperature field in the stir friction welding process, and provide support for the study of the temperature field distribution law; the present invention establishes a reliable stir friction welding temperature field characterization method based on the temperature measurement experiment based on thermocouples, provides a reference for the temperature control of stir friction welding, and provides support for the temperature and microstructure research of different areas of the weldment.

[0006] The technical solution of the present invention:

[0007] A temperature field characterization method based on the friction stir welding process uses a combination of experimental and theoretical methods to conduct a single-axis shoulder experiment of friction stir welding. Feature points are set at different distances from the center of the weld, and thermocouples are arranged symmetrically on the forward and backward sides along the weld. The peak temperature of the feature points is used to train and test the least squares support vector machine algorithm, and the temperature field distribution detection of friction stir welding is realized. The temperature field at different distances from the center of the weld during the welding process is characterized. The specific steps of the method are as follows:

[0008] Step 1: Select a thermocouple and matching hardware to form a signal transmission system, set characteristic points on the weld based on the distance from the weld center and symmetrical distribution, and determine the temperature measurement scheme;

[0009] Step 2: Process the weldment according to the determined temperature measurement scheme, embed the thermocouple, conduct the single-shaft shoulder temperature measurement experiment of the friction stir welding, and obtain the thermal cycle curve of the characteristic points in the welding process;

[0010] Step 3: Design a least squares support vector machine algorithm. In the regression model to be established, given the training data (x1, y1), ..., (x n ,y n ) constitutes the input space; where x iis the 1D input value, which is the distance from the weld center, i=1,…,n; y i is the 1-dimensional target value, is the temperature value; nonlinear mapping is used to map the input space to a high-dimensional feature space, and linear regression is performed in the high-dimensional feature space using the method of structural risk minimization to obtain the effect of nonlinear regression in the original space; SVM regresses the function through the following formula:

[0011] f(x)=ω*φ(x)+b (1)

[0012] In the formula, φ(x) is a function that maps the input space to a high-dimensional feature space; ω is a weight vector; b∈R is a threshold; according to the structural minimization principle, ω and b are determined by minimizing the function to solve the regression function:

[0013]

[0014]

[0015] Where: C is the penalty factor; ε is the insensitivity coefficient; ζ i is the error, which plays the role of a slack variable;

[0016] The least squares linear system is used as the loss function, and the high-dimensional inequality constraints are changed to equality constraints. The loss function based on the least squares method is as follows:

[0017] y i =ω T φ(x i )+b+ζ i (4)

[0018] Introduce the Lagrangian function:

[0019]

[0020] According to Mercer's condition, any semi-positive symmetric function can be used as a kernel function. i ,x j ) optimizes the regression function, and the LS-SVM regression estimation is expressed as follows:

[0021]

[0022] In the formula, α i is the Lagrange multiplier;

[0023] Step 4: Set the parameters of the least squares support vector machine algorithm, select the kernel function, input the peak temperature of the characteristic point obtained by the thermocouple, and distinguish between the training set and the test set;

[0024] Step 5: Use the root mean square error as the evaluation criterion to train the model;

[0025] Step 6: Use the trained model to obtain the complete temperature field distribution at different distances from the weld center.

[0026] The beneficial effects of the present invention are as follows: the temperature field of friction stir welding is predicted by combining experiments with theories, and a temperature field model considering different distances from the center of the weld is established, which improves work efficiency and reduces experimental costs; and solves the problem that the temperature field distribution cannot be fully obtained due to the limited number of feature points. In the established temperature field model, the temperature ranges of different regions of the microstructure can be obtained, which is conducive to studying the distribution law of the temperature field and the influence of the temperature field on the microstructure and welding quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a comparison chart of the characterization results based on the least squares support vector machine algorithm;

[0028] Figure 2 It is the temperature field distribution diagram of friction stir welding. DETAILED DESCRIPTION

[0029] The specific implementation of the present invention is described in detail below in combination with the technical scheme and the accompanying drawings. The temperature field of the 2219 aluminum alloy friction stir welding process is characterized using a thermocouple and a least squares support vector machine algorithm based on Matlab. The specific steps are as follows:

[0030] (1) K-type armored thermocouple is selected as the temperature measuring element, the temperature transmitter performs nonlinear correction on the temperature signal, the PCI-1747U data acquisition card collects the temperature signal and transmits it to the host computer; the characteristic points are selected at 6 mm from the upper surface and 8 mm, 9 mm, 10 mm, 11 mm, 12 mm, 13 mm, 14 mm, 15 mm, 18 mm, and 21 mm from the center of the weld, and are arranged symmetrically on the forward and backward sides.

[0031] (2) In the experiment, the workpiece is a cuboid of 300 mm × 75 mm × 18 mm. A hole is punched from the side, and the tip of the thermocouple probe is embedded in the feature point. A single-axis shoulder stir friction welding experiment is carried out. The stirring head speed is set to 400 r / min, the welding speed is 100 mm / min, and the pressing speed is 20 mm / min. The peak temperature of the feature point is obtained. The peak temperatures of the forward and backward sides are shown in the table.

[0032] Peak temperature on the forward side

[0033]

[0034] Peak temperature on the retreat side

[0035]

[0036] (3) Import the lib-svm toolbox in the Matlab root directory and write the least squares support vector machine algorithm. The algorithm code is as follows:

[0037] xn_train1=[8,9,11,13,14,18,21];

[0038] dn_train1=[458.9,432.7,406.2,392.7,377.9,330,316.2];

[0039] xn_train2=[-8,-9,-11,-13,-14,-18,-21];

[0040] dn_train2=[424.8,405.6,380,362.1,352.1,310.1,300];

[0041] xn_test1 = [10, 12, 15];

[0042] dn_test1=[424.2,392.9,373.9];

[0043] xn_test1 = [-10,-12,-15];

[0044] dn_test1=[394.7,369.4,340];

[0045] type='function estimation';

[0046] kernel = 'RBF_kernel';

[0047] gam = 10000;

[0048] sig2 = [4];

[0049] The above code inputs the peak temperature obtained by the thermocouple, selects 7 and 3 groups of data as training sets and test sets respectively, sets the basic parameters of the algorithm, and selects the kernel function as RBF Gaussian radial basis function.

[0050] model1=initlssvm(xn_train1',dn_train1',type,gam,sig2,kernel);

[0051] model1=trainlssvm(model1);

[0052] Yn1=simlssvm(model1,xn_test1');

[0053] Xz = [6:0.15:21]

[0054] Yz1=simlssvm(model1,Xz')

[0055] Model2=initlssvm(xn_train2',dn_train2',type,gam,sig2,kernel);

[0056] model2=trainlssvm(model2);

[0057] Yn2=simlssvm(model2,xn_test2');

[0058] Xz = [6:0.15:21]

[0059] Yz2=simlssvm(model2,Xz')

[0060] The above code uses the established model for prediction, obtains the temperature field distribution at different distances from the weld center, and solves the temperature model from the weld core area to the base metal area at intervals of 0.15 mm.

[0061] subplot(1,4,1)

[0062] plot(xn_test1',dn_test1','r+:',xn_test1',Yn,'go:')

[0063] title('+ is the experimental value, o is the predicted value')

[0064] subplot(1,4,2)

[0065] plot(Xz',Yz,'r+:')

[0066] subplot(1,4,3)

[0067] plot(xn_test1',dn_test1','r+:',xn_test1',Yn,'go:')

[0068] title('+ is the experimental value, o is the predicted value')

[0069] subplot(1,4,4)

[0070] plot(Xz',Yz,'r+:')

[0071] RMSE = ((Yn1(1)-dn_test1(1))^2+(Yn1(2)-dn_test1(2))^2+(Yn1(3)-dn_test1(3))^2)+(Yn2(1)-dn_test2(1))^2+(Yn2(2)-dn_test2(2))^2+(Yn2(3)-dn_test2(3))^2)^0.5 / 6 The above code plots the temperature field and compares it with the experimental temperature. Figure 1 As shown, the output RMS error is 3.9743, and the temperature peak curves at different distances from the weld center are obtained, such as Figure 2 shown.

[0072] For the friction stir welding process of 2219 aluminum alloy, a temperature field characterization model of friction stir welding was established using K-type thermocouple and least squares support vector machine algorithm. The peak temperature of the characteristic point measured by the experiment was compared with the model value. Figure 1 , we can see that the error is within 4%, with high accuracy. Figure 2 The temperature field distribution at the distance from the weld center shows the distribution law of the temperature field of the composite stir friction welding. The results show that the temperature value is highly consistent with the experimental value, which can prove that the temperature model is reliable in characterizing the temperature field of the 2219 aluminum alloy stir friction welding.

Claims

1. A method for characterizing the temperature field of friction stir welding based on the least squares support vector machine algorithm, characterized in that: The specific steps are as follows: Step 1: Select thermocouples and matching hardware to form a signal transmission system, set characteristic points on the weldment based on the distance from the weld center and symmetrical distribution, and determine the temperature measurement plan; Step 2: Process the weldment according to the determined temperature measurement scheme, embed the thermocouple, conduct the single-shaft shoulder temperature measurement experiment of the friction stir welding, and obtain the thermal cycle curve of the characteristic points in the welding process; Step 3: Design a least squares support vector machine algorithm. In the regression model to be established, given the training data (x1, y1), ..., (x n ,y n ) constitutes the input space; where x i is the 1D input value, which is the distance from the weld center, i=1,…,n; y i is the 1-dimensional target value, is the temperature value; nonlinear mapping is used to map the input space to a high-dimensional feature space, and linear regression is performed in the high-dimensional feature space using the method of structural risk minimization to obtain the effect of nonlinear regression in the original space; SVM regresses the function through the following formula: f(x)=ω*φ(x)+b (1) In the formula, φ(x) is a function that maps the input space to a high-dimensional feature space; ω is a weight vector; b∈R is a threshold; according to the structural minimization principle, ω and b are determined by minimizing the function to solve the regression function: Where: C is the penalty factor; ε is the insensitivity coefficient; ζ i is the error, which plays the role of a slack variable; The least squares linear system is used as the loss function, and the high-dimensional inequality constraints are changed to equality constraints. The loss function based on the least squares method is as follows: y i =ω T φ(x i )+b+z i (4) Introduce the Lagrangian function: According to Mercer's condition, any semi-positive symmetric function can be used as a kernel function. i ,x j ) optimizes the regression function, and the LS-SVM regression estimation is expressed as follows: In the formula, α i is the Lagrange multiplier; Step 4: Set the parameters of the least squares support vector machine algorithm, select the kernel function, input the peak temperature of the characteristic point obtained by the thermocouple, and distinguish between the training set and the test set; Step 5: Use the root mean square error as the evaluation criterion to train the model; Step 6: Use the trained model to obtain the complete temperature field distribution at different distances from the weld center.

Images