Component design method of marine structural steel welding material

Abstract

The invention provides a component design method for a marine structural steel welding material. The component design method comprises the steps that S1, a uniform mechanical property data sample of the marine structural steel welding material is obtained, and a model is built; s2, multiple sets of input data related to alloy element components of the welding material are obtained through complete set crossover operation, mechanical property predicted values corresponding to the input data are obtained on the basis of the constructed model after standardization processing, and a database containing the mechanical property predicted values corresponding to the input data is established; and S3, target mechanical property parameters are set, and alloy element composition design data meeting preset property parameter conditions are retrieved and output in the database. According to the method, modeling is conducted through uniform mechanical property data samples, the interaction among alloy elements in the welding material can be accurately reflected, and the mechanical property of the welding material is accurately predicted within the temperature range of-60 DEG C to 20 DEG C under the pressure of 300 MPa to 1000 MPa; and blindness in component design of the marine high-strength steel welding material is reduced, and the research and development period is greatly shortened.

Description

Technical Field

[0001] The invention relates to the technical field of metal material welding, and in particular to a composition design method for ship structural steel. Background Art

[0002] As more and more new ships with better performance continue to emerge, there is an urgent need for the research and development of high-performance ship materials. Ship structures are mainly made by welding various specifications of structural steel, so welded joints are often the weakest parts of ship structures and are prone to material failures such as fracture and fatigue cracks.

[0003] For ship high-strength steel weld metal, alloying elements are the most critical factors affecting the mechanical properties of weld metal, and even determine the service performance of ship structures. Since weld metal is a cast structure, it cannot be rolled and heat treated like steel plates to improve performance; in addition, there are often complex interactions and nonlinear effects between the various alloying elements in the welding material. The effect of each element on the weld metal depends not only on itself, but also on the presence and addition of other elements. As a result, there is no fixed functional relationship between the performance of the welding material and its alloying elements. Therefore, the research and development of welding materials for ship structural steel is extremely difficult.

[0004] If the optimal design is sought by conducting comprehensive tests on all alloying elements, it is obvious that the number of tests is too large and almost impossible to achieve. For example, to examine the influence of 6 main alloying elements on the performance of welding materials, each element takes 5 levels, then 15625 tests are required. For this reason, domestic and foreign scholars usually use the traditional empirical method and single factor rotation method. The empirical method is to adjust the element content based on the known alloy composition of the welding material by experience. It has relatively large limitations and is somewhat blind; while the single factor rotation method is to first fix several factors unchanged, and only change one factor each time to test, and obtain its optimal content until a relatively satisfactory test result is produced. For example, the Chinese patent with application number 202210498755.3 discloses a design method for aluminum alloy welding wire, including: S1, welding wire composition design; S2, smelting; S3, machining; S4, welding; S5, sample preparation; S6, microstructure observation; S7, mechanical properties test; S8, data feedback; S9, welding wire preparation. This method directly uses ingots to produce aluminum alloy welding wire, which can omit the extrusion and drawing processes during the research and development process to improve research and development efficiency; however, this method cannot examine the interaction between various factors, and the extracted experimental information is highly one-sided; if the interaction between alloy elements is large, it is difficult to obtain the optimal result.

[0005] In view of this, the present invention is proposed. Summary of the invention

[0006] The problem solved by the present invention is that the existing composition design method of marine welding materials has limitations and certain blindness, and it is difficult to consider the interaction between various alloy elements, resulting in a large amount of experiments and a long research and development cycle.

[0007] To solve the above problems, the present invention provides a composition design method for ship structural steel welding materials, comprising the following steps: S1, obtaining uniform mechanical property data samples of ship structural steel welding materials and constructing a model; S2, obtaining multiple groups of input data related to the alloy element composition of the welding materials through a full set cross operation, and then obtaining mechanical property prediction values ​​corresponding to the input data based on the constructed model after standardization, and establishing a database containing mechanical property prediction values ​​corresponding to the input data; S3, setting target mechanical property parameters, retrieving and outputting alloy element composition design data that meets the preset performance parameter conditions in the database.

[0008] Preferably, the alloy element composition value range used as input data in step S2 is: W(C): 0.03%-0.12%, w(Si): 0.02%~0.6%, w(Mn): 0.4%~2.4%, w(Ni): 0.4%~9.6%, w(Cr): 0.1%~1.0%, w(Mo): 0.1%~1.2%, w(V): 0%~0.02%, w(Ti): 0%~0.05%, and the progressive length of the input data is W(C): 0.01%, w(Si): 0.1%, w(Mn): 0.1%, w(Ni): 0.2%, w(Cr): 0.1%, w(Mo): 0.1%, w(V): 0.01%, w(Ti): 0.02%.

[0009] Preferably, in step S2, the input data is standardized using the mean value μ and standard deviation σ of the sample data used for modeling in step S1.

[0010] Preferably, in step S3, an error function is set and the error value between the target performance and the predicted performance is calculated, and the input variable corresponding to the predicted performance with the smallest error is selected as the design composition of the alloy element.

[0011] Preferably, the error function is a weighted error Error = |yield prediction value - yield target value|*W1+|tensile prediction value - tensile target value|*W2+|impact prediction value - impact target value|*W1, wherein W1 = Rmse(tensile)*Rmse(impact) / [Rmse(yield)*Rmse(tensile)]+Rmse(yield)*Rmse(impact)]+Rmse(tensile)*Rmse(impact)]; W2 = Rmse(yield)*Rmse(impact) / [Rmse(yield)* Rmse(tensile)]+Rmse(yield)*Rmse(impact)]+Rmse(tensile)*Rmse(impact)]; W3=Rmse(yield)*Rmse(tensile) / [Rmse(yield)*Rmse(tensile)]+Rmse(yield)*Rmse(impact)]+Rmse(tensile)*Rmse(impact)]; where Rmse(yield), Rmse(tensile) and Rmse(impact) are the root mean square errors of the yield strength model, tensile strength model and impact work model, respectively.

[0012] Preferably, the mechanical properties include yield strength Rp 0.2 , tensile strength Rm, elongation, and impact energy.

[0013] Preferably, step S1 comprises:

[0014] S11. Obtain uniform data samples of mechanical properties of high-strength marine steel welding materials;

[0015] On the basis of historical test data, orthogonal test was adopted to design and prepare a series of welding materials with different compositions. Welding operation was carried out under different line energy. The alloy elements (C, Si, Mn, Ni, Cr, Mo, V, Ti, Cu), yield strength, tensile strength, elongation and impact energy at different temperatures of weld metal were measured. The above data were taken as data samples, and the maximum value, minimum value, average value and standard deviation of the sample data were calculated.

[0016] S12. Determine the input characteristic variables of the model through correlation analysis and significance t test;

[0017] S13, standardize the sample data according to the formula z=(x-μ) / σ, where μ and σ represent the mean value and standard deviation of the characteristic parameter respectively, and z represents the processed value;

[0018] S14, model establishment;

[0019] Randomly select 80%-90% of the sample data as the training set, and the remaining 10%-20% of the sample data as the test set, use the XGBoost algorithm to build the model and adjust the parameters; use the test set to calculate the determination coefficient R of the constructed model2 , average error and root mean square error, when the determination coefficient R 2 When >0.92 and the average error percentage ≤12%, it means that the modeling is successful.

[0020] Preferably, in step S12, the correlation coefficients between alloy elements, line energy E and temperature and mechanical properties of welding materials are analyzed, and the input characteristic variables of the model are determined according to the significance level of the correlation between the two.

[0021] Preferably, in step S14, the XGBoost model is adjusted by random grid search and five-fold cross validation, the accuracy value is calculated for each division of the cross validation, the average validation accuracy is calculated for each hyperparameter setting, and the hyperparameter with the highest average validation accuracy is selected as the model hyperparameter. Preferably, the welding material is a welding rod.

[0022] Preferably, after the standardization processing in step S2, the line energy and temperature of the input data are obtained based on the constructed model as fixed values, for example, the line energy is set to the conventional welding line energy value of the electrode 13.3kJ / cm, and the temperature is set to the typical assessment temperature of the ship steel electrode, that is, -50°C.

[0023] Compared with the prior art, the composition design method of ship structural steel welding materials described in the present invention has the following beneficial effects: 1) uniform mechanical property data samples are obtained for modeling, which can more accurately reflect the interaction between alloy elements in the welding materials, and accurately predict the mechanical properties of the welding materials in the range of 300MPa to 1000MPa and temperature in the range of -60℃ to 20℃; 2) the alloy composition design of ship high-strength steel welding materials can be quickly realized through the massive sample grid search method, effectively reducing the blindness in the composition design of ship high-strength steel welding materials, and the research and development cycle is short and efficient; 3) by weighted processing of mechanical property data, the impact energy obtains a higher weight, thereby improving the accuracy of alloy design. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a diagram showing the prediction results of mechanical properties based on the constructed model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. The technical features of the embodiments of the present invention can be combined with each other without conflict.

[0026] Since the welded joint is the weak part of the ship structure, the optimization design of the alloy elements of the welding material is a complex problem with multiple factors. For example, the Ni element in the weld metal of the high-strength steel for ships can increase the stacking fault energy, thereby promoting the cross slip of screw dislocations at low temperatures, which is beneficial to the low-temperature toughness of the weld metal; but when the Mn content in the weld is high, excessive increase in Ni content will form more martensite and coarse bainite structure, which will seriously reduce the low-temperature toughness of the weld.

[0027] Traditional empirical methods and single-factor optimization methods are difficult to consider the interaction between various alloy elements, have great limitations and are somewhat blind. With the rapid development of computer technology, computing power has been greatly improved, and computer models and algorithms have gradually been used in the field of material performance prediction, which is conducive to the design of welding material alloy composition and thus greatly shortens the research and development cycle. To this end, the applicant proposes the following technical solution:

[0028] A method for designing alloy composition of a marine structural steel welding rod comprises the following steps:

[0029] S1. Obtain uniform mechanical properties data samples of high-strength marine steel welding materials

[0030] 64 groups of orthogonal experiments were designed, and a series of welding rods with different compositions were prepared. Welding operations were carried out under different line energies to determine the alloy elements (C, Si, Mn, Ni, Cr, Mo, V, Ti, Cu), yield strength, tensile strength, elongation and impact energy of the weld metal at different temperatures. Combined with the above historical data, a total of 276 groups of data samples related to yield strength, tensile strength and elongation, and 393 groups of data samples related to impact energy were obtained. The maximum, minimum, average and standard deviation of the sample data were calculated. The results are shown in Table 1.

[0031] Table 1 Minimum, maximum, average and standard deviation of sample data

[0032]

[0033] S2. Determine input feature variables

[0034] SPSS and other software were used to analyze the correlation coefficients between alloy elements, line energy E and temperature and the mechanical properties of welding materials. The input characteristic variables of the model were determined according to the significance level of the correlation between the two. The results are shown in Table 2.

[0035] Table 2 Correlation analysis between input characteristic variables and mechanical properties

[0036]

[0037] Note: The calculation formula of t-test statistic is: Where n is the total number of samples.

[0038] According to the significance level α = 0.05 and the degree of freedom n-2, we can check the t distribution table and get: t α / 2 =1.96. If the t-test value of an input characteristic variable is greater than 1.96, it indicates that there is a significant correlation with the mechanical properties. As shown in Table 2, the input characteristic variables for yield strength and tensile strength that pass the t-test are line energy E and C, Si, Mn, Ni, Cr, Mo, V and Ti elements, the input characteristic variables for elongation that pass the t-test are C, Si, Mn, Ni, Cr, Mo, V and Ti elements, and the input characteristic variables for impact energy that pass the t-test value are temperature T and C, Si, Mn, Ni, Cr, Mo, V and Ti elements. In summary, the input characteristic variables of each model are determined according to the correlation coefficient r and t-test results as shown in Table 3:

[0039] Table 3 Input characteristic variables and output variables of the model

[0040]

[0041] S3. Sample data preprocessing

[0042] The sample data is standardized according to the formula z = (x-μ) / σ, where μ and σ represent the mean and standard deviation of the characteristic parameters, respectively, and z represents the value after processing. Since data such as line energy, alloy element content, and test temperature have different physical meanings and different dimensions, the sample data should also be standardized before modeling. By transforming the input features into a standard normal distribution (i.e., the mean is 0 and the standard deviation is 1), it can ensure that all features have the same importance in subsequent modeling, and avoid the numerical values ​​of some characteristic variables being too large or too small to dominate the model learning process.

[0043] S4. Construct a prediction model for the mechanical properties of marine high-strength steel welding materials, including:

[0044] S41. Randomly select 80%-90% of the sample data as the training set, and the remaining 10%-20% of the sample data as the test set. Adjust the parameters of the XGBoost model through random grid search and five-fold cross validation. The specific range and step size of the hyperparameters are shown in Table 4. The accuracy value is calculated for each division of the cross validation. The average validation accuracy is calculated for each hyperparameter setting. The hyperparameter with the highest average validation accuracy is selected as the model hyperparameter to obtain the optimized hyperparameter combination, as shown in Table 5. The mechanical properties prediction model of high-strength steel welding materials is established using the optimized hyperparameter combination.

[0045] Table 4 Hyperparameter search range and step size

[0046]

[0047] Table 5. Optimized XGBOOST model hyperparameters

[0048]

[0049] Since the prediction of mechanical properties of welding materials is a complex problem with the interaction of multiple factors, traditional regression fitting methods are difficult to solve this complex problem, while machine learning algorithms such as BP neural network, multi-layer perceptron (MLP), support vector machine (SVM), random forest (BF), gradient boosting machine (GBDT) and extreme gradient boosting machine (XGBOOST) can be used to deal with such problems. After research, it was found that XGBOOST is more effective than other algorithms in modeling.

[0050] S42. Use the test set to calculate the coefficient of determination R of the constructed model 2 , average error and root mean square error, determine whether the coefficient R 2 >0.92 and the average error percentage is ≤12%. If so, it means that the model is successfully constructed; if not, the hyperparameter search range is expanded, including increasing the hyperparameter range and refining the step size, for example, the hyperparameter range is expanded to 120% of the original range and the step size is refined to 1 / 2-2 / 3 of the original step size, and then the process returns to step S41.

[0051] The coefficient of determination R 2 , average error and root mean square error and other indicators are used to analyze the prediction accuracy of the constructed model, and the prediction performance of the model is verified using the test set data. 2 >0.92 and the average error percentage ≤12%, indicating that the prediction accuracy of the constructed model is good; when the determination coefficient R 2 When <0.92 or the average error percentage is >12%, it means that the prediction accuracy of the model is poor. Then the hyperparameter search range is expanded (including increasing the hyperparameter range and refining the step size), and the five-fold cross validation and grid search methods are continued to obtain the best hyperparameters until a prediction model with ideal prediction accuracy is obtained. The results are shown in Table 6. Figure 1 .

[0052] Table 6 Model prediction accuracy indicators

[0053]

[0054] The mechanical property model of the ship's high-strength steel welding material has a wide range of application, a large span of strength levels (300MPa to 1000MPa), and a wide temperature range (-60°C to 20°C); all test data are derived from the measured data of ship structural steel welding rods, with high prediction accuracy, overcoming the shortcoming that empirical formulas cannot reflect interactions, and is suitable for predicting the mechanical properties of ship steel welding rods.

[0055] S5, obtaining multiple sets of input data through full set crossover operation, and establishing a database including the input data and corresponding mechanical property prediction values;

[0056] Specifically, the progressive length is set within the value range of the input characteristic variables, multiple sets of input data are obtained through full set crossover operation, the input data are standardized and the corresponding output variables, namely yield strength, tensile strength, elongation and impact energy, are obtained using the model, and a database containing input data and corresponding mechanical property prediction values ​​is obtained. For example, the line energy is set to the conventional welding line energy value of the welding rod 13.3kJ / cm, the temperature is set to the typical assessment temperature of the marine steel welding rod, namely -50℃, and the value range and progressive length of other input characteristic variables (alloy elements) are shown in Table 7, and a total of 26231040 sets of input data are obtained.

[0057] Table 7 Value range and incremental length of input characteristic variables (alloying elements)

[0058]

[0059]

[0060] It should be noted that when the input data is standardized, the values ​​of μ and σ are consistent with those in step S3, that is, the mean value and standard deviation of the sample data during modeling are adopted.

[0061] S6. Set target performance parameters and retrieve alloy element composition design data that meets the preset performance parameter conditions from the database.

[0062] Specifically, if the target performance of the electrode is set to yield strength Rp 0.2 =920MPa-930MPa, tensile strength Rm≥970MPa, elongation A≥17%, -50℃ impact energy KV2≥100J. The design components that meet the conditions are obtained through retrieval, and the results are shown in Table 8.

[0063] Table 8 Design components to meet target performance

[0064]

[0065] When precise design is required, the error function is set to weighted error and the error between the target performance and the predicted performance is calculated. The input variable corresponding to the predicted performance with the smallest error is selected as the design component.

[0066] The weighted error Error = |yield prediction value - yield target value|*W1+|tensile prediction value - tensile target value|*W2+|impact prediction value - impact target value|*W1;

[0067] W1 = Rmse (tensile) * Rmse (impact) / [Rmse (yield) * Rmse (tensile)] + Rmse (yield) * Rmse (impact)] + Rmse (tensile) * Rmse (impact)];

[0068] W2 = Rmse(yield)*Rmse(impact) / [Rmse(yield)*Rmse(tensile)]+Rmse(yield)*Rmse(impact)]+Rmse(tensile)*Rmse(impact)];

[0069] W3 = Rmse (yield) * Rmse (tensile) / [Rmse (yield) * Rmse (tensile)] + Rmse (yield) * Rmse (impact)] + Rmse (tensile) * Rmse (impact)];

[0070] Among them, Rmse(yield), Rmse(tensile) and Rmse(impact) are the root mean square errors of the yield strength model, tensile strength model and impact work model respectively.

[0071] Since the values ​​of yield strength, tensile strength and impact energy are quite different, the above weighting process can give the impact energy a higher weight, thereby improving the accuracy of alloy design. 0.2 =925MPa, tensile strength R m =980MPa, -50℃ impact energy KV2=100J, calculate the error between the target performance and the predicted performance, the input variable corresponding to the predicted performance with the smallest error is the design component, and the results are shown in Table 9.

[0072] Table 9 Design results of alloy elements in welding rods

[0073]

[0074] Although the present invention is disclosed as above, the present invention is not limited thereto. Any person skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the scope defined by the claims.

Claims

1. A composition design method for marine structural steel welding materials, characterized in that: The method comprises the following steps: S1, obtaining uniform mechanical property data samples of marine structural steel welding materials and constructing a model; S2, obtaining multiple groups of input data related to the alloy element composition of the welding materials through full set cross operation, obtaining mechanical property prediction values ​​corresponding to the input data based on the constructed model after standardization, and establishing a database containing mechanical property prediction values ​​corresponding to the input data; S3, setting target mechanical property parameters, retrieving and outputting alloy element composition design data that meets the preset performance parameter conditions in the database.

2. The composition design method of marine structural steel welding materials according to claim 1, characterized in that: The alloy element composition value range used as input data in step S2 is: W(C): 0.03%-0.12%, w(Si): 0.02%~0.6%, w(Mn): 0.4%~2.4%, w(Ni): 0.4%~9.6%, w(Cr): 0.1%~1.0%, w(Mo): 0.1%~1.2%, w(V): 0%~0.02%, w(Ti): 0%~0.05%, and the progressive length of the input data is W(C): 0.01%, w(Si): 0.1%, w(Mn): 0.1%, w(Ni): 0.2%, w(Cr): 0.1%, w(Mo): 0.1%, w(V): 0.01%, w(Ti): 0.02%.

3. The composition design method of marine structural steel welding materials according to claim 1, characterized in that: In step S2, the input data is standardized using the mean value μ and standard deviation σ of the sample data used for modeling in step S1.

4. The composition design method of marine structural steel welding materials according to claim 1, characterized in that: In step S3, an error function is set and the error value between the target performance and the predicted performance is calculated, and the input variable corresponding to the predicted performance with the smallest error is selected as the design composition of the alloy element.

5. The composition design method of marine structural steel welding materials according to claim 4, characterized in that: The error function is weighted error Error = |yield prediction value - yield target value |*W1 + |tensile prediction value - tensile target value |*W2 + |impact prediction value - impact target value |*W1, wherein W1 = Rmse (tensile) * Rmse (impact) / [Rmse (yield) * Rmse (tensile)] + Rmse (yield) * Rmse (impact)] + Rmse (tensile) * Rmse (impact)]; W2 = Rmse (yield) * Rmse (impact) / [Rmse (yield) * Rmse (tensile)] + Rmse (yield) * Rmse (impact)] + Rmse (tensile) * Rmse (impact)]; W3 = Rmse(yield)*Rmse(tensile) / [Rmse(yield)*Rmse(tensile)]+Rmse(yield)*Rmse(impact)]+Rmse(tensile)*Rmse(impact)]; where Rmse(yield), Rmse(tensile) and Rmse(impact) are the root mean square errors of the yield strength model, tensile strength model and impact work model, respectively.

6. The composition design method of marine structural steel welding materials according to claim 1, characterized in that: The mechanical properties include yield strength Rp 0.2 , tensile strength Rm, elongation, and impact energy.

7. The composition design method of marine structural steel welding materials according to claim 1, characterized in that: Step S1 includes: S11. Obtain uniform data samples of mechanical properties of high-strength marine steel welding materials; On the basis of historical test data, orthogonal test was adopted to design and prepare a series of welding materials with different compositions. Welding operation was carried out under different line energy. The alloy elements (C, Si, Mn, Ni, Cr, Mo, V, Ti, Cu), yield strength, tensile strength, elongation and impact energy at different temperatures of weld metal were measured. The above data were taken as data samples, and the maximum value, minimum value, average value and standard deviation of the sample data were calculated. S12. Determine the input characteristic variables of the model through correlation analysis and significance t test; S13, standardize the sample data according to the formula z=(x-μ) / σ, where μ and σ represent the mean value and standard deviation of the characteristic parameter respectively, and z represents the processed value; S14, model establishment; Randomly select 80%-90% of the sample data as the training set, and the remaining 10%-20% of the sample data as the test set, use the XGBoost algorithm to build the model and adjust the parameters; use the test set to calculate the determination coefficient R of the constructed model 2 , average error and root mean square error, when the determination coefficient R 2 When >0.92 and the average error percentage ≤12%, it means that the modeling is successful.

8. The composition design method of marine structural steel welding materials according to claim 7, characterized in that: In step S12, the correlation coefficients between alloy elements, line energy E and temperature and mechanical properties of welding materials are analyzed, and the input characteristic variables of the model are determined according to the significance level of the correlation between the two.

9. The composition design method of marine structural steel welding materials according to claim 7, characterized in that: In step S14, the XGBoost model is adjusted through random grid search and five-fold cross validation. The accuracy value is calculated for each division of cross validation, the average validation accuracy is calculated for each hyperparameter setting, and the hyperparameter with the highest average validation accuracy is selected as the model hyperparameter.

Images