Welded joint strength prediction method and welding parameter optimization method

Abstract

The invention relates to the technical field of welding, and discloses a welded joint strength prediction method and a welding parameter optimization method.The welded joint strength prediction method comprises the steps that real welding parameters and corresponding welded joint strength data are collected; according to the welding parameters, a microstructure model is established, thermal-mechanical coupling simulation is conducted, and a corresponding welding joint strength simulation value is obtained; establishing a relation model between the welding joint strength simulation value and the welding parameters; and predicting the corresponding welding joint strength based on the relation model and the new welding parameters. The optimization method comprises the following steps: setting a value range of the strength of a welding joint; all the values of all the welding parameters are substituted into the welding joint strength prediction method, and corresponding welding joint prediction values are obtained; and selecting the corresponding welding parameter when the predicted value of the welding joint meets the value range, namely the optimized welding parameter. According to the method, the prediction precision of the welding joint strength can be improved, and the welding parameters corresponding to the required welding joint strength are obtained.

Description

Technical Field

[0001] The present invention relates to the field of welding technology, and in particular to a welding joint strength prediction method and a welding parameter optimization method. Background Art

[0002] With the rapid development of my country's economy, the welding industry has also developed rapidly. In various engineering constructions, welding technology is an indispensable process for connecting metal structures, and the mechanical properties of welded joints will directly determine the safety of the structure. However, during the welding process, the material undergoes thermal-mechanical coupling and the microstructure evolves in a complex manner, resulting in large fluctuations in the mechanical properties of the joints and poor controllability of product quality. At present, the optimization design of welding process parameters usually relies on experience, has a long cycle and high cost. Therefore, studying the prediction method of the mechanical properties of welded joints is of great significance for the optimization of welding structure processes and the improvement of product performance.

[0003] The mechanical properties of welded joints are greatly affected by microstructure, mainly including grain size, hard phase distribution, soft phase distribution, etc. At present, the traditional methods for analyzing joint microstructure mainly use experimental testing methods, including fracture analysis, electrolytic corrosion method, X-ray diffraction technology, etc. These methods are limited by factors such as experimental equipment, cost and manpower, and are only suitable for small-scale experimental research and process verification.

[0004] Machine learning technology provides an effective way to solve this problem. The prediction of microstructure parameters (such as grain size, hard phase volume fraction, etc.) can be achieved by using methods such as multivariate regression analysis, decision tree, neural network, support vector machine, etc. However, these methods all use statistical methods to establish a direct relationship between output and input, ignoring the physical process of the welding process, and the accuracy of the coupled prediction of welding process parameters, microstructure characteristics and mechanical properties of welded joints is limited. Summary of the invention

[0005] The present invention is made to solve the above-mentioned technical problems, and one of its purposes is to provide a method for predicting the strength of a welded joint, which can improve the accuracy of the prediction of the strength of a welded joint.

[0006] Another object of the present invention is to provide a welding parameter optimization method capable of obtaining welding parameters corresponding to the required welding joint strength.

[0007] According to one embodiment of the present invention, a method for predicting the strength of a welded joint is provided, comprising the following steps:

[0008] S1. Collect the real welding parameters and corresponding welding joint strength data;

[0009] S2. According to the welding parameters, a microstructure model is established, and a thermal-mechanical coupling simulation is performed to obtain a corresponding simulated value of the weld joint strength;

[0010] S3, establishing a relationship model between the welding joint strength simulation value and welding parameters;

[0011] S4. Based on the relationship model and the new welding parameters, predict the corresponding weld joint strength.

[0012] According to one embodiment of the present invention, a welding parameter optimization method is provided, comprising the following steps:

[0013] Set the value range of welding joint strength;

[0014] Substituting all values ​​of various welding parameters into the above-mentioned welding joint strength prediction method to obtain corresponding welding joint prediction values;

[0015] Select the welding parameters corresponding to the welding joint prediction value when it meets the value range, which is the optimized welding parameters.

[0016] According to the above description and practice, the weld joint strength prediction method described in the present invention combines the establishment of a microstructure model with thermal-mechanical coupling simulation to achieve a refined simulation of the welding process. The relationship model between welding parameters and weld joint strength is established based on a statistical high-order multi-scale analysis algorithm, which can accurately predict the weld joint strength corresponding to different welding parameters, providing a theoretical basis for the refined design and optimization of the welding process. In the case of insufficient test data, a prediction method for weld joint strength is provided, which can be used for the optimal design of welding parameters to improve the quality and efficiency of welding production. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 The present invention is a flowchart of a method for predicting weld joint strength according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, the exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and fully convey the concepts of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0019] In this embodiment, a method for predicting the strength of a welded joint is disclosed. Figure 1 As shown in Figure 2, the weld joint strength prediction method mainly includes the following four steps:

[0020] Step S1, collecting real welding parameters and corresponding welding joint strength data;

[0021] Step S2, establishing a microstructure model according to the welding parameters, performing a thermal-mechanical coupling simulation, and obtaining a corresponding weld joint strength simulation value;

[0022] Step S3, establishing a relationship model between the welding joint strength simulation value and welding parameters;

[0023] Step S4: predicting the corresponding weld joint strength based on the relationship model and the new welding parameters.

[0024] This weld joint strength prediction method combines the establishment of a microstructure model with thermal-mechanical coupling simulation to achieve a refined simulation of the welding process. The relationship model between welding parameters and weld joint strength is established based on a statistical high-order multi-scale analysis algorithm, which can accurately predict the weld joint strength corresponding to different welding parameters, providing a theoretical basis for the refined design and optimization of the welding process. In the case of insufficient test data, a prediction method for weld joint strength is provided, which can be used for the optimal design of welding parameters to improve the quality and efficiency of welding production.

[0025] In one embodiment, the actual welding parameters and corresponding welding joint strength data collected in step S1 are also preprocessed, feature engineered and normalized to facilitate subsequent calculations such as establishing a microstructure model, performing thermal-mechanical coupling simulation, and establishing a relationship model between welding joint strength simulation values ​​and welding parameters.

[0026] Specifically, in this embodiment, the welding parameters include: welding process parameter K, groove angle θ, welding speed v, welding current I, argon gas flow f, welding inclination angle α, filler material ratio λ, weld width l, gap e, interlayer temperature Ti and heat input Q, etc., which are parameters that need to be set during welding operations. Accordingly, preprocessing includes removing abnormal values ​​and filling missing values ​​in the welding parameters, and converting the data type and unit of the welding parameters into a preset format.

[0027] Among them, the outliers in these parameter data refer to those data points that deviate significantly from other values ​​in the data set. These outliers may be caused by measurement errors, data entry errors, or real variations. In practical applications, it is necessary to combine specific process requirements and historical data to determine outliers, and use statistical analysis methods (such as standard deviation, box plots, etc.) to identify data points that deviate significantly from the normal range. The handling of outliers is crucial to ensure the integrity and accuracy of the data, as they may have an adverse effect on the training and prediction results of the model.

[0028] For example, for the welding process parameter K, most of the data points are concentrated in a small range, such as 50-150, so a sudden high value such as 500 or a low value such as 0 is considered an outlier. For the groove angle θ, the groove angle of welding usually has a reasonable range, such as 30 degrees to 60 degrees. If a θ value of 10 degrees or 90 degrees appears in the data set, this is beyond the normal range and can be considered an outlier. For the welding speed v, if most of the data points of the welding speed v are in a specific range, such as 2 meters / minute to 5 meters / minute, then a v value of 0.5 meters / minute or 10 meters / minute is considered abnormal. For the welding current I, if the normal range of the welding current I is 100 amperes to 300 amperes, then a current value such as 50 amperes or 500 amperes is abnormal. For the argon gas flow rate f, if the normal range of the argon gas flow rate f is 5 liters / minute to 20 liters / minute, then a flow value such as 2 liters / minute or 30 liters / minute is considered abnormal. For the welding inclination angle α, if the normal range of the welding inclination angle α is 45 degrees to 85 degrees, then an α value such as 10 degrees or 90 degrees is considered abnormal. For the filler material ratio λ, if the normal range of the filler material ratio λ is 0.1 to 0.9, then a λ value such as 0 or 1 is considered abnormal. For the weld width l, if the normal range of the weld width l is 3 mm to 10 mm, then an l value such as 1 mm or 15 mm is considered abnormal. For the gap e, if the normal range of the gap e is 0.5 mm to 2 mm, then an e value such as 0.1 mm or 3 mm is considered abnormal. For the interlayer temperature Ti, if the normal range of the interlayer temperature Ti is 100 degrees to 300 degrees, then a Ti value such as 50 degrees or 400 degrees is considered abnormal. For heat input Q, if the normal range of heat input Q is 1 kJ / m to 10 kJ / m, then a Q value such as 0.5 kJ / m or 20 kJ / m is considered abnormal.

[0029] In one embodiment, for a welding parameter x, there is a reasonable minimum value x min and a reasonable maximum value x max ,when or When , the actual value of the welding parameter is an abnormal value.

[0030] The purpose of converting the data type and unit of welding parameters into a preset format is to ensure that all data is in a format suitable for subsequent analysis and calculation. In the data preprocessing stage, data type conversion and unit unification are two important steps that ensure that the data is consistent and standardized before analysis.

[0031] Among them, data type conversion refers to converting data from one format or type to another to meet the requirements of analysis tools. These include numeric type conversion, date type conversion, categorical data encoding, etc. Numeric type conversion refers to converting string type numbers to numeric types for mathematical operations. Date type conversion refers to converting dates from one format (such as "YYYY-MM-DD") to another format (such as "DD / MM / YYYY"), or to timestamps. Categorical data encoding refers to converting text categories into numerical representations, for example, using One-Hot Encoding or Label Encoding.

[0032] For example, the welding speed v exists in the data set in the form of a string, such as "5m / min". The data type conversion will convert this string into a numeric type, that is, 5, for subsequent calculation and analysis.

[0033] Unit unification refers to converting the same type of data into the same unit of measurement to ensure comparability between data. For example, if the groove angle θ appears in the data set in both degrees and radians, they need to be unified into one unit, such as degrees. If the welding speed v appears in the data set in both meters / minute and centimeters / second, they need to be unified into one unit, such as meters / minute. If the interlayer temperature Ti appears in the data set in both Celsius and Fahrenheit, they need to be unified into one unit, such as Celsius.

[0034] The above-mentioned format suitable for analysis means that the data has been processed into a format that can be directly read and processed by analysis tools (such as statistical software, machine learning algorithms, etc.). It usually includes: digitization, all data should be of numerical type for mathematical operations; consistency, all data should follow the same format and unit; missing value processing, missing values ​​should be filled or deleted; outlier processing, outliers should be identified and processed; feature encoding, categorical variables should be converted to numerical format so that the model can process them.

[0035] The feature engineering of the actual welding parameters and the corresponding welding joint strength data collected in step S1 includes:

[0036] The Pearson correlation coefficient method is used to perform correlation analysis on welding parameters and weld joint strength data. For strongly correlated welding parameters, only one of them is retained. The calculation formula is:

[0037]

[0038] In the formula, x i and iRespectively represent the two variables in the welding parameters and the welding joint strength data, and is the mean of the corresponding variable, n is the sample size, r xy represents the correlation coefficient between two variables, |r xy |≥0.6 indicates a strong correlation.

[0039] In this step, x i and i Refers to any two data of welding parameters and weld joint strength data, the correlation coefficient r xy It is a statistic that measures the degree of linear correlation between the two variables. The correlation coefficient r xy The value of is between -1 and 1. xy A value close to 1 or -1 indicates a strong correlation, that is, there is a significant linear relationship between the two variables; r xy A value close to 0 indicates no correlation, that is, there is no significant linear relationship between the two variables. xy |≥0.6 indicates a strong correlation between two variables, |r xy |<0.6 indicates that there is no correlation between the two variables. For example, if the absolute value of the correlation coefficient between welding speed and welding current is much greater than 0.6, it indicates that there is a strong linear relationship between the two features, and one of the parameters needs to be eliminated to avoid overfitting of the subsequent relationship model.

[0040] The actual welding parameters and corresponding welding joint strength data collected in step S1 are normalized, and the calculation formula for mapping the data to the interval [0,1] is:

[0041]

[0042] In the formula, x i is any of the original data of welding parameters and weld joint strength data, min(x) and max(x) are the minimum and maximum values ​​in the corresponding data set, respectively, x i ′ is the normalized data. For each type of welding parameter and corresponding welding joint strength data, a sample data set can be obtained after normalization.

[0043] In step S2, the normalized sample data set is used as an input parameter, and a weld microstructure model is established using a numerical simulation method, such as a phase diagram method, a line growth algorithm, etc. The Kinetic Monte Carlo model is used to achieve real-time update of the phase interface, simulate the evolution of the microstructure during welding, establish a periodic boundary for the weld microstructure, simulate the cooling process, calculate the thermal expansion coefficient, thermal conductivity, specific heat capacity and Young's modulus, and complete the thermal-mechanical coupling solution process of the weld microstructure.

[0044] In one embodiment, in step S3, a relationship model between the weld joint strength simulation value and the welding parameters is established, including:

[0045] The simulated value of welding joint strength and welding parameters are used as sample input, and Gaussian multivariate kernel function is used as the base kernel function to obtain the high-dimensional mapping of feature space and samples. The calculation formula is:

[0046]

[0047] The simulated value of welding joint strength and welding parameters are used as sample input, and the RBF kernel function is used as the base kernel function to perform nonlinear mapping on the sample and feature space. The calculation formula is:

[0048] K RBF (x,y)=exp(-γ||xy|| 2 ) (4)

[0049] Formula (3) and formula (4) are combined in a preset manner to obtain the combined kernel function value K combined (x, y), where x and y are input samples, σ is the parameter of the Gaussian multivariate kernel function, and γ is the parameter of the RBF kernel function;

[0050] Calculate the kernel function value K of all samples combined (x,y), form the kernel matrix, and calculate the kernel vector of each sample.

[0051] In this step, a complex model capable of handling nonlinear relationships was constructed using Gaussian multivariate kernel functions and RBF kernel functions, thereby improving the accuracy of weld joint strength prediction. The Gaussian multivariate kernel function is used to map data to a high-dimensional space, helping the model capture the complex nonlinear relationships between input features; the RBF kernel function, as a composite kernel function, further enhances the model's ability to handle nonlinear data. By combining the two kernel functions, the model can more effectively learn and predict the relationship between welding parameters and weld joint strength.

[0052] Both functions are kernel functions used to map data to high-dimensional feature space. The kernel function calculates the similarity between two samples in high-dimensional space without explicitly calculating the coordinates in high-dimensional space. The combination of the two functions means that the similarity of data in different aspects can be captured. For example, the Gaussian multivariate kernel function and the RBF kernel function can be combined to form a new kernel function, which can better capture the nonlinear relationship of the data. The specific combination method needs to be determined through experiments and verification of specific data to achieve the best prediction effect.

[0053] For example, in some embodiments, the combination of K combined The (x,y) methods include the following:

[0054] K combined (x,y)=K Gaussian (x,y);

[0055] K combined (x,y)=K RBF (x,y);

[0056] K combined (x,y)=K Gaussian (x,y)+K RBF (x,y);

[0057] K combined (x,y)=K Gaussian (x,y)×K RBF (x,y).

[0058] In other embodiments, the Gaussian multivariate kernel function and the RBF kernel function may be multiplied by a coefficient and then added to obtain K. combined (x, y). For example, in one embodiment, K combined (x,y)=pK Gaussian (x,y)+qK RBF (x, y), the values ​​of coefficients p and q can be set according to actual needs.

[0059] In this step, the kernel function value K of all samples is calculated combined (x, y), form a kernel matrix, and calculate the kernel vector of each sample. The kernel matrix is ​​composed of the kernel function values ​​K corresponding to all samples. combined Specifically, if there is a data set containing n samples {x1,x2,...,x n}, then the element K of the kernel matrix K ij is the kernel function K combined (x, y). Therefore, the kernel matrix is ​​an n×n matrix that represents the similarity between all sample pairs.

[0060] The kernel matrix represents the relationship between samples in the feature space, and the kernel function K combined (x, y) calculates the similarity between two samples in the feature space, and the kernel matrix is ​​the collection of these similarities. The kernel vector is calculated by calculating the kernel function value of a sample and all other samples in the data set to form a vector. Specifically, for a sample {x1, x2, ..., x n}, its kernel vector k i By K combined (x i ,x j ) is calculated for all j, where K combined (x i ,x j ) is the kernel function mentioned above, x j is the jth sample in the sample set. In other words, the kernel vector k i is calculated by taking a sample x i Compared with all other samples x in the dataset j (where j≠i) are calculated and these values ​​are arranged into a vector.

[0061] The purpose of calculating the kernel vector is to represent samples in a high-dimensional feature space. In machine learning, the kernel vector can be used to calculate the similarity of samples in the feature space. The original data can be mapped to a high-dimensional space through the kernel vector, and there is no need to explicitly calculate the coordinates in the high-dimensional space. In this embodiment, the calculation of the kernel vector is used to establish a relationship model between welding parameters and weld joint strength. The simulated values ​​of welding parameters and joint strength can be mapped to a high-dimensional feature space through the kernel vector, and then these mappings are used to train the model to predict the strength of the weld joint. Kernel vector calculation, as a key step in establishing the above-mentioned relationship model, can perform effective similarity calculations in high-dimensional space, thereby improving the predictive ability of the model.

[0062] After the relationship model between the weld joint strength simulation value and the welding parameters is established in step S3, the relationship model is further trained to improve the accuracy of the relationship model. The training method includes the following steps:

[0063] Step S301, dividing the welding parameters and corresponding welding joint strength data in step S1 into a training set and a test set;

[0064] Step S302, substituting the training set into the relationship model, and optimizing the relationship model using a generalized expectation-maximization algorithm;

[0065] Step S303, substitute the test set into the optimized relational model, and calculate the mean absolute percentage error MAPE between the predicted value of the weld joint strength obtained by the relational model and the simulated value of the weld joint strength. The calculation formula is:

[0066]

[0067] In the formula, n is the number of samples, is the predicted value of the strength of the ith weld joint, σ i is the simulated value of the strength of the ith welding joint;

[0068] Step S304, adjusting the hyperparameters in the relationship model, the hyperparameters include parameters σ and γ, selecting different σ and γ to train the relationship model, repeating steps S302 and S303, and selecting the hyperparameters corresponding to the minimum mean absolute percentage error MAPE to configure the relationship model.

[0069] Among them, for the distribution of welding parameters and corresponding weld joint strength data, the data set used to train the model usually accounts for 70% to 80% of the total data; the data set used to evaluate the performance of the model usually accounts for 20% to 30% of the total data. That is, the training set accounts for 70% to 80% of the total data, and the test set accounts for 20% to 30% of the total data, and the sum of the two is 100%.

[0070] When it comes to specific allocation, it can be done by random sampling or using specific data splitting techniques, such as k-fold cross validation. In k-fold cross validation, the data is divided into k equal parts, the model is trained on k-1 parts, and tested on the remaining parts. The process is repeated k times, each time choosing a different part as the test set.

[0071] The generalized expectation maximization algorithm is used to optimize the model, which is a common method of model optimization training. It mainly includes the following steps: initialization: select the initial parameter estimates, including σ and γ; E step: calculate the expected value of each sample based on the current parameter estimates, for example, calculate the expected value of the real joint strength under the conditions of the welding parameters and the joint strength simulation value; M step: maximize the likelihood function or log-likelihood function based on the expected value calculated in the E step to update the estimates of σ and γ; iteration: repeat the E step and the M step until the parameter estimates converge. As a prior art, it will not be described in detail here.

[0072] For the training of this relationship model, a more accurate one can be selected from several combinations of the above two kernel functions, and the parameters σ and γ can be optimized to form the most accurate relationship model between various welding parameters and the strength of the welded joint.

[0073] Based on the above relationship model, users only need to input new welding parameters to predict the corresponding weld joint strength.

[0074] Therefore, in this embodiment, a welding parameter optimization method is also disclosed, comprising the following steps:

[0075] First, set the value range of the welding joint strength and clarify the upper and lower limits of the required welding loudness.

[0076] Afterwards, all values ​​of each welding parameter are substituted into the welding joint strength prediction method as described above to obtain the corresponding welding joint prediction value. In this step, all values ​​of each welding parameter refer to all values ​​of any welding parameter within the standard range or reasonable range, which is equivalent to traversing all possible values ​​within these ranges to obtain the corresponding welding joint prediction value.

[0077] Finally, the welding parameters corresponding to the welding joint prediction values ​​satisfying the value range are selected as the optimized welding parameters.

[0078] The above-mentioned welding joint strength prediction method and welding description optimization method establish a relationship model between welding parameters and welding joint strength, realize the quantitative relationship between welding parameters and welding joint strength, and provide technical support for improving the mechanical properties of welding joints. Compared with traditional methods, the present invention considers the combination of simulation calculation and statistical analysis, improves the prediction accuracy, and provides a basis for welding parameter optimization and welding mechanical properties research.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for predicting weld joint strength, characterized in that: The steps include: S1. Collect the real welding parameters and corresponding welding joint strength data; S2. According to the welding parameters, a microstructure model is established, and a thermal-mechanical coupling simulation is performed to obtain a corresponding simulated value of the weld joint strength; S3, establishing a relationship model between the welding joint strength simulation value and welding parameters; S4. Predicting the corresponding weld joint strength based on the relationship model and the new welding parameters.

2. The method for predicting weld joint strength according to claim 1, characterized in that: In step S1, the welding parameters and welding joint strength data are also preprocessed, feature engineered and normalized.

3. The method for predicting weld joint strength according to claim 2, characterized in that: The welding parameters include: welding process parameter K, groove angle θ, welding speed v, welding current I, argon gas flow f, welding inclination angle α, filling material ratio λ, weld width l, gap e, interlayer temperature Ti and heat input Q; The preprocessing includes: removing abnormal values ​​and filling missing values ​​in the welding parameters, and converting the data type and unit of the welding parameters into a preset format.

4. The method for predicting weld joint strength according to claim 2, characterized in that: Feature engineering of the welding parameters and weld joint strength data includes: The Pearson correlation coefficient method is used to perform correlation analysis on the welding parameters and the weld joint strength data. For the strongly correlated welding parameters, only one of them is retained. The calculation formula is: In the formula, x i and i Respectively represent the two variables in the welding parameters and the welding joint strength data, and is the mean of the corresponding variable, n is the sample size, r xy represents the correlation coefficient between two variables, |r xy |≥0.6 indicates a strong correlation.

5. The method for predicting weld joint strength according to claim 2, characterized in that: The welding parameters and welding joint strength data are normalized and mapped to the [0,1] interval. The calculation formula is: In the formula, x i is any one of the original data of the welding parameters and weld joint strength data, min(x) and max(x) are the minimum and maximum values ​​in the corresponding data set, respectively, x i ′ is the normalized data.

6. The method for predicting weld joint strength according to claim 5, characterized in that: In step S2, the normalized data is used as input parameters, and a weld microstructure model is established using a numerical simulation method. The Kinetic Monte Carlo model is used to achieve real-time updating of the phase interface, simulate the evolution of the microstructure during welding, establish a periodic boundary for the weld microstructure, simulate the cooling process, calculate the thermal expansion coefficient, thermal conductivity, specific heat capacity and Young's modulus, and complete the thermal-mechanical coupling solution process of the weld microstructure.

7. The method for predicting weld joint strength according to claim 6, characterized in that: In step S3, a relationship model between the welding joint strength simulation value and welding parameters is established, including: The simulated value of the weld joint strength and the welding parameters are used as sample inputs, and the Gaussian multivariate kernel function is used as the base kernel function to obtain a high-dimensional mapping of the feature space and the sample. The calculation formula is: The simulated value of the weld joint strength and the welding parameters are used as sample inputs, and the RBF kernel function is used as the base kernel function to perform nonlinear mapping on the sample and the feature space. The calculation formula is: K RBF (x,y)=exp(-γ||x-y|| 2 ) (4) Formula (3) and formula (4) are combined in a preset manner to obtain the combined kernel function value K combined (x, y), where x and y are input samples, σ is the parameter of the Gaussian multivariate kernel function, and γ is the parameter of the RBF kernel function; Calculate the kernel function value K of all samples combined (x,y), form the kernel matrix, and calculate the kernel vector of each sample.

8. The method for predicting weld joint strength according to claim 7, characterized in that: After establishing the relationship model between the welding joint strength simulation value and the welding parameters, the relationship model is also trained, and the training method includes: The welding parameters and corresponding welding joint strength data in step S1 are divided into a training set and a test set; Substituting the training set into the relational model, and optimizing the relational model using a generalized expectation-maximization algorithm; Substitute the test set into the optimized relational model, and calculate the mean absolute percentage error (MAPE) between the predicted value of the weld joint strength obtained by the relational model and the simulated value of the weld joint strength. The calculation formula is: In the formula, n is the number of samples, is the predicted value of the strength of the ith weld joint, σ i is the simulated value of the strength of the ith welding joint; The hyperparameters in the relationship model are adjusted, wherein the hyperparameters include parameters σ and γ, different σ and γ are selected to train the relationship model, and the hyperparameters corresponding to the minimum mean absolute percentage error (MAPE) are selected to configure the relationship model.

9. The method for predicting weld joint strength according to claim 7, characterized in that: The K combined (x,y)=K Gaussian (x,y); or The K combined (x,y)=K RBF (x,y); or The K combined (x,y)=K Gaussian (x,y)+K RBF (x,y); or The K combined (x,y)=K Gaussian (x,y)×K RBF (x,y).

10. A welding parameter optimization method, characterized in that: The steps include: Set the value range of welding joint strength; Substituting all values ​​of various welding parameters into the welding joint strength prediction method as described in any one of claims 1 to 9 to obtain corresponding welding joint prediction values; The welding parameters corresponding to the welding joint prediction value satisfying the value range are selected as the optimized welding parameters.

Images