A method and device for predicting the impact toughness of a multi-pass weld heat-affected zone

Abstract

The present application relates to the technical field of material performance prediction, and particularly relates to a multi-pass welding heat affected zone impact toughness prediction method and device. The method comprises the following steps: obtaining impact toughness data of a double-pass heat affected zone of a welded material; preprocessing the impact toughness data of the double-pass heat affected zone to obtain target feature variables; performing dimension reduction processing on the target feature variables, and inputting the dimension reduction processed data into a machine learning algorithm for training and verification to obtain a machine learning model; and predicting the impact toughness of a multi-pass welding heat affected zone of a target welded material based on the machine learning model. The present application is a multi-pass welding heat affected zone impact toughness prediction method based on low alloy steel double-pass heat simulation experiment data and machine learning technology, which considers the chemical composition, original microstructure and impact performance of the welded material (base material) to improve the efficiency of welding process design and the reliability of the welded joint.

Description

Technical Field

[0001] This invention relates to the field of material property prediction technology, and in particular to a method and apparatus for predicting the impact toughness of the heat-affected zone in multi-pass welding. Background Technology

[0002] In the engineering field, the manufacturing and installation of critical metal components often require multi-pass welding. During multi-pass welding, the base material undergoes multiple thermal cycles, and the impact toughness of its heat-affected zone (HAZ) is one of the key factors affecting the performance of the welded joint. However, the chemical composition, original microstructure, and mechanical properties of the base material all have a significant impact on the impact toughness of the weld heat-affected zone.

[0003] The impact toughness of the weld heat-affected zone (HAZ) of materials is typically assessed using three methods: actual welding, post-thermal simulation impact testing, and predicting the performance of the HAZ by establishing a neural network model. Chinese patent CN 106513925 A uses the actual welding method to assess the impact toughness of the HAZ and to screen or optimize welding process parameters. However, this method relies on destructive testing of the welded component, and because the HAZ is very narrow, the location of the impact notch greatly affects the accuracy of the test results, making it difficult to comprehensively reflect the influence of welding process parameters. Furthermore, this method is costly and time-consuming.

[0004] The lack of a quantitative correlation model between the simulation parameters of laboratory two-pass thermal simulation tests and actual welding parameters in engineering (such as interpass temperature and heat input) remains a technical challenge in predicting the impact toughness of the actual multi-pass weld heat-affected zone (HAZ) using thermal simulation data. Chinese patent CN 112632721 A discloses "A Method for Predicting the Performance of MAG Welded Joints Based on a Combined Model," which uses welded joint performance data as training data to establish a BP neural network model, and then establishes an RBF neural network model based on the training data; a Kriging interpolation model is also established based on the training data; a linear weighted method is used to combine the BP neural network model, the RBF neural network model, and the Kriging interpolation model to construct a combined model; and the combined model is used to predict the performance of MAG welded joints. This aims to improve the accuracy and stability of predicting MAG welded joint performance (weld reinforcement, joint tensile strength, and impact energy). However, this method not only requires destroying the welded structure but also does not consider the influence of the original chemical composition, microstructure, and original mechanical properties of the base material on the weld heat-affected zone. Therefore, for new welded materials, it is still difficult to accurately predict the performance of the welded joint.

[0005] Therefore, there is an urgent need to propose a method that can efficiently and accurately predict the impact toughness of the heat-affected zone in multi-pass welding. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method and apparatus for predicting the impact toughness of the heat-affected zone in multi-pass welding.

[0007] This invention provides the following technical solution:

[0008] A method for predicting the impact toughness of the heat-affected zone in multi-pass welding is provided, the method comprising,

[0009] Obtain impact toughness data of the double-pass secondary heat-affected zone of the welded material;

[0010] The impact toughness data of the dual-channel secondary heat-affected zone are preprocessed to obtain the target characteristic variables;

[0011] The target feature variables are dimensionality reduced, and the dimensionality-reduced data is input into a machine learning algorithm for training and validation to obtain a machine learning model.

[0012] Based on the machine learning model, the impact toughness of the multi-pass heat-affected zone of the target welded material is predicted.

[0013] Specifically, the impact toughness data affecting the dual-channel secondary heat-affected zone is preprocessed, including:

[0014] The impact toughness data of the dual-channel heat-affected zone were cleaned, missing and outlier values ​​were removed, and then standardized to obtain the feature variables that significantly affect the impact toughness of the heat-affected zone. These feature variables include multiple sub-feature variables.

[0015] Based on the principles of metallurgy, the interaction between two or more sub-feature variables is used to obtain the target feature variable.

[0016] Specifically, the characteristic variables that significantly affect the impact toughness of the heat-affected zone include the chemical element content, carbon equivalent, cold crack sensitivity coefficient, original microstructure parameters, impact performance parameters, and thermal cycling parameters of each welded material in a two-pass thermal simulation.

[0017] Specifically, the chemical element content includes the content of carbon, silicon, manganese, sulfur, phosphorus, chromium, molybdenum, nickel, copper, niobium, vanadium, and / or titanium; and / or,

[0018] The original microstructure parameters include the microstructure type of the material being welded, the proportion of each microstructure type, grain size, and banding level; and / or,

[0019] Impact performance parameters include impact absorbed energy, lateral expansion rate, and shear ratio; and / or,

[0020] The thermal cycling parameters for dual-track thermal simulation include: peak temperature of dual-track thermal cycle, heating rate of dual-track thermal cycle, high-temperature residence time of dual-track thermal cycle, cooling rate of dual-track thermal cycle, and inter-track temperature of dual-track thermal cycle.

[0021] Specifically, the sub-characteristic variables include the content of each chemical element, carbon equivalent and cold crack sensitivity coefficient, percentage of granular bainite, ferrite, pearlite and martensite-austenite, grain size of the original structure, banded structure level, peak temperature of the two-pass thermal cycle, heating rate of the two-pass thermal cycle, high-temperature residence time of the two-pass thermal cycle, cooling rate of the two-pass thermal cycle, interpass temperature of the two-pass thermal cycle, impact absorption energy and shear cross-sectional area.

[0022] Specifically, the target feature variables are dimensionality reduced, and the dimensionality-reduced data is input into a machine learning algorithm for training and validation to obtain the optimal machine learning model, including:

[0023] The target feature variables are dimensionality reduced to obtain the dimensionality-reduced data.

[0024] The data after dimensionality reduction is divided into a training set and a test set, and the training set is input into a machine learning algorithm for training to build a machine learning model.

[0025] The parameters in the machine learning model were optimized using the five-fold cross-validation method.

[0026] Based on the mean squared error and coefficient of determination of the optimization results in the test set, the machine learning algorithm corresponding to the optimization results when the mean squared error is less than or equal to the first threshold and the coefficient of determination is greater than or equal to the second threshold is taken as the optimal machine learning model.

[0027] A device for predicting the impact toughness of the heat-affected zone in multi-pass welding is also provided, the device comprising,

[0028] The acquisition unit is used to acquire impact toughness data of the double-pass secondary heat-affected zone of the welded material.

[0029] The preprocessing unit is used to preprocess the impact toughness data of the dual-channel secondary heat-affected zone to obtain the target feature variables;

[0030] The obtained unit is used to reduce the dimensionality of the target feature variables, and the dimensionality-reduced data is input into the machine learning algorithm for training and validation to obtain the machine learning model;

[0031] The prediction unit is used to predict the impact toughness of the multi-pass heat-affected zone of the target welded material based on the machine learning model.

[0032] Specifically, the impact toughness data affecting the dual-channel secondary heat-affected zone is preprocessed, including:

[0033] The impact toughness data of the dual-channel heat-affected zone is cleaned, missing values ​​and outliers are removed, and then standardized to obtain the feature variables that have a significant impact on the impact toughness of the heat-affected zone. The feature variables that have a significant impact on the impact toughness of the heat-affected zone include multiple sub-feature variables.

[0034] Based on the principles of metallurgy, the interaction between two or more sub-feature variables is used to obtain the target feature variable.

[0035] Specifically, the sub-characteristic variables include the content of each chemical element, carbon equivalent and cold crack sensitivity coefficient, percentage of granular bainite, ferrite, pearlite and martensite-austenite, grain size of the original structure, banded structure level, peak temperature of the two-pass thermal cycle, heating rate of the two-pass thermal cycle, high-temperature residence time of the two-pass thermal cycle, cooling rate of the two-pass thermal cycle, interpass temperature of the two-pass thermal cycle, impact absorption energy and shear cross-sectional area.

[0036] Specifically, the target feature variables are dimensionality reduced, and the dimensionality-reduced data is input into a machine learning algorithm for training and validation to obtain the optimal machine learning model, including:

[0037] The target feature variables are dimensionality reduced to obtain the dimensionality-reduced data.

[0038] The data after dimensionality reduction is divided into a training set and a test set, and the training set is input into a machine learning algorithm for training to build a machine learning model.

[0039] The parameters in the machine learning model were optimized using the five-fold cross-validation method.

[0040] Based on the mean squared error and coefficient of determination of the optimization results in the test set, the machine learning algorithm corresponding to the optimization results when the mean squared error is less than or equal to the first threshold and the coefficient of determination is greater than or equal to the second threshold is taken as the optimal machine learning model.

[0041] The technical effects and advantages of this invention are as follows:

[0042] This invention utilizes two-pass thermal simulation experiments to accurately reproduce the complex thermal cycling process of multi-pass welded HAZs and establishes an efficient and reliable impact toughness prediction model. Using this method, through two-pass thermal simulation experiments with materials of different compositions, parameters such as peak temperature (Tp1, Tp2), cooling time rate (t8 / 5), interval time (Δt) / interpass temperature, etc., of the two thermal cycles can be precisely controlled. This realistically reproduces the tempering / reheating effect (such as softening zone, critical coarse grain zone, etc.) of subsequent weld passes on the HAZ of preceding weld passes in actual multi-pass welding, overcoming the deficiency of single-pass thermal simulation in reflecting the heat accumulation effect of multi-pass welding. Furthermore, based on the two-pass thermal simulation experimental data, a machine learning model is constructed that quantitatively maps "composition → process parameters (heat input, interpass temperature, etc.) → microstructure (grain size, MA components, etc.) → impact toughness," achieving multi-parameter collaborative prediction of the impact toughness of multi-pass welded HAZs. This method requires only a small number of thermal simulation samples to cover multiple welding process windows (such as different heat inputs and interpass temperature combinations), avoiding the cumbersome process of repeated full-size welding, sampling, and impact testing required by traditional methods, shortening the R&D cycle by more than 50%, and significantly reducing test costs by 60% to 80%.

[0043] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description and the drawings. Attached Figure Description

[0044] Figure 1 This is a flowchart of the multi-pass weld heat-affected zone impact toughness prediction method provided in the embodiments of this application;

[0045] Figure 2 This is a flowchart illustrating the specific implementation of the multi-pass weld heat-affected zone impact toughness prediction method provided in this application embodiment;

[0046] Figure 3 This is a comparison chart of the predicted impact value of the heat-affected zone of the X80 pipe GMAW ring weld and the actual Charpy impact value of the ring weld joint provided in the embodiments of this application. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] To address the shortcomings of existing technologies, this invention discloses a method for predicting the impact toughness of the heat-affected zone in multi-pass welding, such as... Figure 1As shown, the method includes,

[0049] Step 1: Analyze multiple welded materials with the same steel grade but different chemical element contents to obtain the impact toughness data of the double heat-affected zone of the welded materials; wherein, the impact toughness data of the double heat-affected zone includes the chemical element content, carbon equivalent, cold cracking sensitivity coefficient, original microstructure parameters, impact performance parameters and thermal cycling parameters of each welded material.

[0050] For example, in step 101, select welding materials of the same grade (steel) but different chemical compositions, and record the chemical composition (such as the content of elements such as C, Mn, Cr, Mo, Ni, etc., carbon equivalent CE), initial microstructure and impact properties of the materials.

[0051] Step 102: Then, thermal cycling parameters for different chemical compositions are obtained through a two-stage thermal simulation experiment, including the peak temperature of the two-stage thermal cycle, the heating rate of the two-stage thermal cycle, the high-temperature residence time of the two-stage thermal cycle, and the cooling rate of the two-stage thermal cycle (t). 8 / 5 and t 8 / 3 And the interlayer (channel) temperature of the dual-channel thermal cycle, etc. t 8 / 5 The time t represents the cooling time from 800℃ to 500℃. 8 / 3 This indicates the time required to cool from 800℃ to 300℃. Step 103: Obtain the impact toughness values ​​(such as impact absorbed energy, shear ratio, lateral expansion rate, etc.) of the double-pass thermal simulation specimen using impact testing.

[0052] Step 2: The impact toughness data affecting the dual-channel secondary heat-affected zone is preprocessed;

[0053] Step 3: Perform dimensionality reduction on the target feature variables, and input the dimensionality-reduced data into the machine learning algorithm for training and validation to obtain the machine learning model;

[0054] Step 4: Based on the machine learning model, predict the impact toughness of the multi-pass weld hardening zone (HAZ) of the target weld material. For example, optimize the welding process parameters based on the prediction results to improve the impact toughness of the welded joint and ensure the reliability of the welded structure.

[0055] In a specific embodiment of the present invention, for step 2, the impact toughness data of the dual-channel secondary heat-affected zone is preprocessed to obtain target characteristic variables, including:

[0056] Step 201: Clean the impact toughness data of the dual-channel heat-affected zone, remove missing and outlier values, perform standardization processing to eliminate dimensional differences, and obtain the feature variables that have a significant impact on the impact toughness of the heat-affected zone.

[0057] In one specific embodiment of the present invention, the characteristic variables that significantly affect the impact toughness of the heat-affected zone include the chemical element content, carbon equivalent, cold cracking sensitivity coefficient, original microstructure parameters, impact performance parameters, and thermal cycling parameters from a two-pass thermal simulation of each welded material. The chemical element content includes the contents of carbon, silicon, manganese, sulfur, phosphorus, chromium, molybdenum, nickel, copper, niobium, vanadium, and titanium; and / or,

[0058] The original microstructure parameters include the microstructure type of the material being welded, the proportion of each microstructure type, grain size, and banding level; and / or,

[0059] Impact performance parameters include impact absorbed energy, lateral expansion rate, and shear ratio; and / or,

[0060] The thermal cycling parameters for dual-track thermal simulation include: peak temperature of dual-track thermal cycle, heating rate of dual-track thermal cycle, high-temperature residence time of dual-track thermal cycle, cooling rate of dual-track thermal cycle, and inter-track temperature of dual-track thermal cycle.

[0061] Among them, the characteristic variables that significantly affect the impact toughness of the heat-affected zone include several sub-characteristic variables, including the content of each chemical element, carbon equivalent and cold crack sensitivity coefficient, the percentage of granular bainite, ferrite, pearlite and martensite-austenite, the grain size of the original structure, the banded structure level, the peak temperature of the two-pass thermal cycle, the heating rate of the two-pass thermal cycle, the high-temperature residence time of the two-pass thermal cycle, the cooling rate of the two-pass thermal cycle, the interpass temperature of the two-pass thermal cycle, the impact absorbed energy and the shear cross-sectional area.

[0062] For example, the sub-feature variables are divided into chemical composition features k. h Original organizational state characteristics k z Thermal cycle parameter characteristics k r Original impact performance characteristics k c Among them, chemical composition characteristic k h This includes the content of each chemical element, carbon equivalent (CE), and cold cracking sensitivity coefficient (Pcm). Original microstructure characteristics k z Including the percentages of granular bainite, ferrite, pearlite, and martensite-austenite (MA components), the grain size of the original structure, and the banding level; thermal cycling parameter characteristics k. r Including peak temperature, heating rate, high-temperature residence time, and cooling rate (t) 8 / 5 and t 8 / 3 ), interlayer (channel) temperature.

[0063] Step 202 involves using metallurgical principles to interact between two or more sub-feature variables to obtain the target feature variable. Among these steps...

[0064] The interactions between the sub-characteristic variables include at least the complex carbide strengthening effect of interactions between chemical elements, the sensitivity of the interaction between microstructure and cooling rate to the formation of martensite-austenite microstructure (MA component), the effect of peak temperature of the interaction between process and microstructure on the dissolution of MA component, the higher-order interaction of the nonlinear relationship between cooling rate and MA component, and the comprehensive index of ductile-brittle phase ratio balance.

[0065] The target feature variable K obtained from the interaction between the two sub-feature variables includes the chemical composition interaction feature variable K. hh Tissue-component interaction characteristic variable K zh Process-organization interaction characteristic variable K rz And the higher-order comprehensive interactive feature variable K obtained by interacting three or more sub-feature variables. gj Among them, the chemical composition interaction characteristic variable K hh This indicates that the chemical composition characteristic k h The sub-feature variables in the text interact pairwise; the tissue-component interaction feature variable K zh This indicates that the chemical composition characteristic k h Sub-feature variables and original organizational state features k z The sub-feature variables interact pairwise; the process-organization interaction sub-feature variable K rz This indicates that the chemical composition characteristic k z Sub-feature variables and thermal cycle parameter feature k r The sub-feature variables interact pairwise; for example, the target feature variables and their expressions are shown in Tables 1, 2 and 3 respectively.

[0066] The target feature variable K also includes the higher-order integrated interaction feature variable K. gj High-order integrated interactive feature variable K gj This refers to obtaining new variables by fusing three or more original features that significantly affect the impact toughness of the heat-affected zone (HAZ) and introducing nonlinear transformations (such as exponential, logarithmic, piecewise functions, etc.) to quantitatively characterize complex physical / metallurgical mechanisms. In the prediction of welded HAZ impact toughness, these features are not simply multiplied or added / subtracted, but rather multi-parameter coupled expressions constructed based on metallurgical principles.

[0067] For example, the higher-order integrated interaction feature variable K gjThe target characteristic variable and its expression are obtained by interacting three or more of the following sub-characteristic variables: cold crack sensitivity coefficient Pcm, MA component and cooling rate; impact performance of the original parent material; ferrite percentage and carbon equivalent CE and martensite-austenite percentage; bainite and pearlite percentage and cooling rate; sulfur (S) and phosphorus (P) elements and band structure. Table 4 shows the high-order integrated interactive characteristic variable K described in this embodiment of the invention. gj The variables include the comprehensive index of cold crack risk, toughness reserve index, microstructure gradient control factor, and impurity segregation-banded microstructure interaction. These variables are obtained by nonlinear transformation of three or more variables in Tables 1 to 4.

[0068] Table 1. Interactive characteristic variables and expressions of chemical components

[0069]

[0070] Note: h1, h2, h3, and h4 represent the calculation coefficients, respectively.

[0071] Table 2. Tissue-component interaction characteristic variables and expressions

[0072]

[0073] Note: z1 and z2 represent the calculated coefficients of tissue-component interaction, respectively.

[0074] Table 3. Process-organization interaction characteristic variables and expressions

[0075]

[0076] Note: r1 represents the calculation coefficient of process-structure interaction, t represents the cooling time, and T represents the deviation from the optimal layer temperature.

[0077] Table 4. High-order integrated interaction characteristic variables and expressions

[0078]

[0079] Note: g1 represents the calculation coefficient of the integrated interaction.

[0080] High-order integrated interactive feature variables refer to new variables obtained by fusing three or more original features that significantly affect the impact toughness of the heat-affected zone and introducing nonlinear transformations (such as exponential, logarithmic, piecewise functions, etc.) to quantitatively characterize complex physical / metallurgical mechanisms. In the prediction of welded HAZ impact toughness, these features are not simply multiplied or added / subtracted, but rather multi-parameter coupled expressions constructed based on metallurgical principles.

[0081] In a specific embodiment of the present invention, step 3, which involves dimensionality reduction of the target feature variables and inputting the dimensionality-reduced data into a machine learning algorithm for training and validation to obtain the optimal machine learning model, includes:

[0082] Step 301: Perform dimensionality reduction on the target feature variable to obtain the dimensionality-reduced data. For example, Principal Component Analysis (PCA) can be used to reduce the dimensionality of the target feature variable K, forming five principal components (PC1-PC5). Then, the principal components PC1-PC5 and the t value from the original feature are combined... 8 / 3 The martensite-austenite content (MA_pct) and the original impact absorbed energy are used as the final model inputs for predicting the impact toughness of the heat-affected zone in multi-pass welds.

[0083] The dimensionality reduction of the target feature variable K in the interactive tables 1-4 using principal component analysis (PCA) is based on the following method:

[0084] (a) Based on the dimensionality reduction of target characteristic variables such as carbon equivalent-microalloy interaction, microalloy carbide formation potential, MA component cooling rate sensitivity index, critical cooling rate threshold, interlayer temperature fluctuation sensitivity, and cold crack risk comprehensive index, the principal component PC1 (cooling rate-hardening principal component) is formed.

[0085] (b) Based on the interaction of the target feature variables, such as nickel-chromium equivalent ratio, manganese-sulfur ratio, phosphorus embrittlement sensitivity index, MA component cooling rate sensitivity index, bainite stability factor, ferrite nucleation driving force, pearlite-impurity interaction term, toughness reserve index, microstructure gradient control factor, and impurity segregation-banded microstructure interaction, the principal component PC2 (microstructure toughness principal component) is formed.

[0086] (c) Based on the dimensionality reduction of target characteristic variables such as carbon equivalent-microalloy interaction, microalloy carbide formation potential, impurity segregation-banded structure interaction, and peak temperature-grain size coupling term, principal component PC3 (microalloy interaction principal component) is formed.

[0087] (d) Based on the dimensionality reduction of target characteristic variables such as carbon equivalent-microalloy interaction, ferrite nucleation driving force, peak temperature-grain size coupling term, and critical cooling rate threshold, principal component PC4 (heat input principal component) is formed.

[0088] (e) Based on the dimensionality reduction of target characteristic variables such as manganese-sulfur ratio, phosphorus embrittlement sensitivity index, toughness reserve index, and impurity segregation-banding structure interaction, the principal component PC5 (impurity embrittlement principal component) is formed.

[0089] Step 302: Divide the dimensionality-reduced data into a training set and a test set, and input the training set into a machine learning algorithm for training to build a machine learning model; for example, the machine learning algorithm includes random forest, gradient boosting tree (GBM), support vector regression (SVR) or neural network.

[0090] Step 303: Optimize the parameters in the machine learning model based on the five-fold cross-validation method;

[0091] Step 304: Based on the mean squared error and coefficient of determination of the optimization results in the test set, obtain the mean squared error (MSE) less than or equal to the first threshold and the coefficient of determination (R²). 2 The machine learning algorithm corresponding to the optimization result that is greater than or equal to the second threshold is taken as the optimal machine learning model. For example, MSE ≤ 100 J. 2 R 2 ≥0.90. Evaluate model performance using the test set, calculating the mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R²). 2 The accuracy of the model is verified by comparing predicted and actual values ​​using indicators such as [insert indicators here] and visualization methods. The criteria for evaluating model accuracy are: mean absolute error (MAE) of impact absorbed energy ≤ 10 J; root mean square error (RMSE) of impact absorbed energy ≤ 20 J; and coefficient of determination (R²) [insert R² value here]. 2 )≥0.90.

[0092] This invention provides a method for predicting the impact toughness of the heat-affected zone (HAZ) in multi-pass welds based on two-pass thermal simulation. By utilizing experimental data from two-pass thermal simulation and machine learning techniques, it achieves efficient prediction of the HAZ impact toughness of actual materials to be welded in multi-pass welds. This method comprehensively considers the influence of welding process parameters and material parameters, improving prediction accuracy. It provides a scientific basis for welding process design, helps improve the performance of welded joints, and ensures the reliability of welded structures.

[0093] The present invention also provides a device for predicting the impact toughness of the heat-affected zone in multi-pass welding, the device comprising,

[0094] The acquisition unit is used to acquire data on the impact toughness of the double-pass heat-affected zone of the welded material; the impact toughness data of the double-pass heat-affected zone includes the chemical element content, carbon equivalent, cold cracking sensitivity coefficient, original microstructure parameters, impact performance parameters and thermal cycling parameters of each welded material.

[0095] The preprocessing unit is used to preprocess the impact toughness data of the dual-channel secondary heat-affected zone to obtain target feature variables; including: cleaning the impact toughness data of the dual-channel secondary heat-affected zone, removing missing values ​​and outliers, and then performing standardization processing;

[0096] The impact toughness data of the dual-channel heat-affected zone were cleaned, missing and outlier values ​​were removed, and then standardized to obtain the feature variables that significantly affect the impact toughness of the heat-affected zone. Among them, the feature variables that significantly affect the impact toughness of the heat-affected zone include chemical composition feature k. h Original organizational state characteristics k z Thermal cycle parameter characteristics k r Original impact performance characteristics k c ;

[0097] Interact with the feature variables to obtain the target feature variable.

[0098] The obtained unit is used to reduce the dimensionality of the target feature variables, and the dimensionality-reduced data is input into the machine learning algorithm for training and validation to obtain the machine learning model;

[0099] The prediction unit is used to predict the impact toughness of the multi-pass heat-affected zone of the target welded material based on the machine learning model.

[0100] The impact toughness data affecting the dual-channel secondary heat-affected zone are preprocessed.

[0101] Regarding the apparatus in the above embodiments, the specific manner in which each unit performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0102] The technical solution of the present invention will be further described below with reference to specific embodiments.

[0103] Taking the circumferential welding of a low-alloy steel X80 pipe with an outer diameter of 1219 mm and a wall thickness of 22 mm as an example, the welding method adopted is gas metal arc welding (GMAW) multi-layer multi-pass welding. A multi-pass weld heat-affected zone impact toughness prediction method based on double-pass secondary heat simulation is used to predict the impact toughness of the coarse-grained zone of the actual X80 pipe to be welded in GMAW circumferential welding. Figure 2 As shown, it includes the following steps:

[0104] S1, Data Acquisition: Acquire characteristic variables that significantly affect the impact toughness of the dual-channel secondary heat-affected zone, including chemical composition characteristics k. h Original organizational state characteristics k z Thermal cycle parameter characteristics k r Original impact performance characteristics k c .

[0105] S1-1, Select welding materials of the same grade or steel, but with different chemical and physical properties, and record the chemical composition characteristics k of the materials. hThis includes the content of elements such as carbon (C), silicon (Si), manganese (Mn), sulfur (S), phosphorus (P), chromium (Cr), molybdenum (Mo), nickel (Ni), copper (Cu), niobium (Nb), vanadium (V), and titanium (Ti), as well as the carbon equivalent (CE), initial microstructure, and impact performance. The chemical composition of the X80 pipe to be welded is shown in Table 5. Its microstructure is granular bainite + martensitic austenitic island structure (GB+MA), with a grain size of 11.5 and a banded structure of 2.0. The initial impact absorption energy is 182 J. The formulas for calculating the carbon equivalent (CE) and cold crack sensitivity coefficient (Pcm) are as follows:

[0106]

[0107]

[0108] Table 5 Chemical composition and carbon equivalent of X80 pipe materials

[0109]

[0110] Then, thermal cycling parameters of the coarse-grained region with different chemical compositions were obtained through a two-pass thermal simulation experiment, including the peak temperature, heating rate, high-temperature residence time, and cooling rate (t) of the two passes. 8 / 5 and t 8 / 3 Temperature between layers (channels), etc.

[0111] Finally, Charpy impact tests were used to obtain the impact absorption energy and shear fraction of the X80 coarse-grained region dual-pass thermal simulation specimens. The thermal simulation was performed using a Gleeble 3500 thermodynamic simulation machine, and the impact test was a V-notch Charpy impact absorption test.

[0112] S2, Data Preprocessing

[0113] The Python programming language is used to clean and standardize the data, handling missing and outlier values.

[0114] The features are then standardized to eliminate dimensional differences.

[0115] Finally, the dataset was divided into training and test sets in an 8:2 ratio.

[0116] S3, Feature Engineering

[0117] S3-1 Select characteristics that significantly affect the impact toughness of the heat-affected zone, such as main chemical composition, original microstructure and impact properties, peak temperature, cooling time, heating rate, and pass interval. Among them, the original microstructure characteristics that significantly affect the impact toughness of the heat-affected zone refer to the microstructure type of the welded material, the proportion of each component phase, grain size, and banded microstructure level; the original impact properties characteristics that significantly affect the impact toughness of the heat-affected zone refer to impact energy absorption and shear ratio.

[0118] Among them, the thermal cycling parameter characteristic k has a significant impact on the impact toughness of the heat-affected zone. r Including peak temperature, heating rate, high-temperature residence time, and cooling rate (t) 8 / 5 and t 8 / 3 ), interlayer (channel) temperature.

[0119] The original impact performance characteristics k that significantly affect the impact toughness of the heat-affected zone c This includes impact energy absorption and shear cross-sectional area.

[0120] S3-2, then consider the interactions between features to generate new features.

[0121] The interactions between these features include at least the composite carbide strengthening effect of niobium, vanadium, and titanium elements, the sensitivity of the interaction between microstructure and cooling rate to the formation of martensite-austenite microstructure (MA component), the influence of peak temperature of the interaction between process and microstructure on the dissolution of MA component, the higher-order interaction between cooling rate and the nonlinear relationship of MA component, and the comprehensive index of ductile-brittle phase ratio balance.

[0122] The target feature variable K includes the chemical composition interaction feature variable K. hh Tissue-component interaction characteristic variable K zh Process-organization interaction characteristic variable K rz

[0123] and higher-order integrated interaction feature variable K gj The target feature variables and their expressions are shown in Tables 6, 7, 8 and 9, respectively.

[0124] Table 6. Interactive characteristic variables and expressions of chemical components

[0125]

[0126] Table 7. Tissue-component interaction characteristic variables and expressions

[0127]

[0128] Table 8. Process-organization interaction characteristic variables and expressions

[0129]

[0130] Table 9. High-order integrated interaction characteristic variables and expressions

[0131]

[0132] S3-3 Finally, Principal Component Analysis (PCA) was used to reduce the dimensionality to 5 principal components.

[0133] The dimensionality reduction of the target feature variable K in the interactive tables 1-4 using principal component analysis (PCA) is based on the following method:

[0134] (a) Based on the dimensionality reduction of target characteristic variables such as carbon equivalent-microalloy interaction, microalloy carbide formation potential, MA component cooling rate sensitivity index, critical cooling rate threshold, interlayer temperature fluctuation sensitivity, and cold crack risk comprehensive index, the principal component PC1 (cooling rate-hardening principal component) is formed.

[0135] (b) Based on the interaction of the target feature variables, such as nickel-chromium equivalent ratio, manganese-sulfur ratio, phosphorus embrittlement sensitivity index, MA component cooling rate sensitivity index, bainite stability factor, ferrite nucleation driving force, pearlite-impurity interaction term, toughness reserve index, microstructure gradient control factor, and impurity segregation-banded microstructure interaction, the principal component PC2 (microstructure toughness principal component) is formed.

[0136] (c) Based on the dimensionality reduction of target characteristic variables such as carbon equivalent-microalloy interaction, microalloy carbide formation potential, impurity segregation-banded structure interaction, and peak temperature-grain size coupling term, principal component PC3 (microalloy interaction principal component) is formed.

[0137] (d) Based on the dimensionality reduction of target characteristic variables such as carbon equivalent-microalloy interaction, ferrite nucleation driving force, peak temperature-grain size coupling term, and critical cooling rate threshold, principal component PC4 (heat input principal component) is formed.

[0138] (e) Based on the dimensionality reduction of target characteristic variables such as manganese-sulfur ratio, phosphorus embrittlement sensitivity index, toughness reserve index, and impurity segregation-banding structure interaction, the principal component PC5 (impurity embrittlement principal component) is formed.

[0139] The five principal components (PC1~PC5) after dimensionality reduction are shown in Table 10, and their cumulative variance contribution rate is ≥85%. The principal component loading matrix is ​​shown in Table 11, where an absolute value >0.7 indicates strong correlation, and 0.4-0.7 indicates moderate correlation.

[0140] Table 10 Principal components and key loading characteristics after dimensionality reduction

[0141]

[0142] Table 11 Principal Component Loading Matrix

[0143]

[0144] Then, the principal components PC1-PC5 and the t in the original features are... 8 / 3 The martensite-austenite content (MA_pct) and the original impact absorbed energy are used as the final model inputs for predicting the impact toughness of the heat-affected zone in multi-pass welds.

[0145] S4, Model Building and Training

[0146] S4-1, Select the random forest regression machine learning algorithm to build the model, and set the hyperparameters n_estimators=100 and max_depth=None.

[0147] S4-2 uses the training set to train the model and optimizes the model parameters through 5-fold cross-validation.

[0148] S5, Model Evaluation

[0149] Evaluate model performance using the test set, and calculate MSE and R. 2 .

[0150] Plot a comparison graph of the predicted and actual values ​​to verify the accuracy of the model.

[0151] S6, Model Application

[0152] The trained model is saved as a file and used for predicting the impact toughness of the heat-affected zone in the GMAW multi-pass welding of actual X80 pipes to be welded.

[0153] The chemical composition, original microstructure, impact properties, and GMAW two-pass coarse-grained zone thermal cycling parameters of the X80 pipeline shown in Table 1 were input into a trained model to predict the impact toughness of the coarse-grained zone of the X80 pipeline to be welded to be 182 J. Multi-layer, multi-pass welding of the X80 pipeline was performed using solid wire gas metal arc welding (GMAW). The welding process parameters used were consistent with the predicted thermal cycling parameters. After welding, samples of the X80 pipeline weld joint were taken for impact testing. The V-notch location was in the coarse-grained zone of the heat-affected zone, allowing for the measurement of the actual test value of the X80 pipeline GMAW multi-pass ring weld joint. Three parallel tests were performed, yielding three actual test values ​​of 181 J, 186 J, and 176 J, respectively. Figure 3 As shown, the values ​​are close to the average of the actual test values ​​of the GMAW multi-ring weld joint of the X80 pipe.

[0154] Based on the prediction results, welding process parameters can be further optimized to improve the impact toughness of welded joints and ensure the reliability of welded structures.

[0155] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting the impact toughness of the heat-affected zone in multi-pass welding, characterized in that, The method includes, Obtain impact toughness data of the double-pass secondary heat-affected zone of the welded material; The impact toughness data affecting the dual-channel secondary heat-affected zone are preprocessed to obtain target feature variables; wherein, the preprocessing of the impact toughness data affecting the dual-channel secondary heat-affected zone includes: The impact toughness data of the dual-channel heat-affected zone is cleaned, missing values ​​and outliers are removed, and then standardized to obtain the feature variables that have a significant impact on the impact toughness of the heat-affected zone. The feature variables that have a significant impact on the impact toughness of the heat-affected zone include multiple sub-feature variables. Based on the principles of metallurgy, the interaction between two or more sub-feature variables is used to obtain the target feature variable; Among them, the characteristic variables that significantly affect the impact toughness of the heat-affected zone include the chemical element content, carbon equivalent, cold crack sensitivity coefficient, original microstructure parameters, impact performance parameters, and thermal cycling parameters of each welded material in the two-pass thermal simulation. The chemical element content includes the content of carbon, silicon, manganese, sulfur, phosphorus, chromium, molybdenum, nickel, copper, niobium, vanadium, and / or titanium; and / or, The original microstructure parameters include the microstructure type of the material being welded, the proportion of each microstructure type, grain size, and banding level; and / or, Impact performance parameters include impact absorbed energy, lateral expansion rate, and shear ratio; and / or, The thermal cycling parameters for the two-pass thermal simulation include: peak temperature of the two-pass thermal cycle, heating rate of the two-pass thermal cycle, high-temperature residence time of the two-pass thermal cycle, cooling rate of the two-pass thermal cycle, and inter-pass temperature of the two-pass thermal cycle. The sub-characteristic variables include the content of each chemical element, carbon equivalent and cold crack sensitivity coefficient, percentage of granular bainite, ferrite, pearlite and martensite-austenite, grain size of the original structure, banded structure level, peak temperature of the two-pass thermal cycle, heating rate of the two-pass thermal cycle, high-temperature residence time of the two-pass thermal cycle, cooling rate of the two-pass thermal cycle, interpass temperature of the two-pass thermal cycle, impact absorption energy and shear cross-sectional area. The target feature variables are dimensionality reduced, and the dimensionality-reduced data is input into a machine learning algorithm for training and validation to obtain a machine learning model. Based on the machine learning model, the impact toughness of the multi-pass heat-affected zone of the target welded material is predicted.

2. The method for predicting the impact toughness of the heat-affected zone in multi-pass welding according to claim 1, characterized in that, The target feature variables are dimensionality reduced, and the dimensionality-reduced data is input into a machine learning algorithm for training and validation to obtain the optimal machine learning model, including: The target feature variables are dimensionality reduced to obtain the dimensionality-reduced data. The data after dimensionality reduction is divided into a training set and a test set, and the training set is input into a machine learning algorithm for training to build a machine learning model. The parameters in the machine learning model were optimized using the five-fold cross-validation method. Based on the mean squared error and coefficient of determination of the optimization results of the test set, the machine learning algorithm corresponding to the optimization results when the mean squared error is less than or equal to the first threshold and the coefficient of determination is greater than or equal to the second threshold is taken as the optimal machine learning model.

3. A device for predicting the impact toughness of the heat-affected zone in multi-pass welding, characterized in that, The device includes, The acquisition unit is used to acquire impact toughness data of the double-pass secondary heat-affected zone of the welded material. A preprocessing unit is used to preprocess the impact toughness data affecting the dual-channel secondary heat-affected zone to obtain target feature variables; wherein, the preprocessing of the impact toughness data affecting the dual-channel secondary heat-affected zone includes: The impact toughness data of the dual-channel heat-affected zone is cleaned, missing values ​​and outliers are removed, and then standardized to obtain the feature variables that have a significant impact on the impact toughness of the heat-affected zone. The feature variables that have a significant impact on the impact toughness of the heat-affected zone include multiple sub-feature variables. Based on the principles of metallurgy, the interaction between two or more sub-feature variables is used to obtain the target feature variable; Among them, the characteristic variables that significantly affect the impact toughness of the heat-affected zone include the chemical element content, carbon equivalent, cold crack sensitivity coefficient, original microstructure parameters, impact performance parameters, and thermal cycling parameters of each welded material in the two-pass thermal simulation. The chemical element content includes the content of carbon, silicon, manganese, sulfur, phosphorus, chromium, molybdenum, nickel, copper, niobium, vanadium, and / or titanium; and / or, The original microstructure parameters include the microstructure type of the material being welded, the proportion of each microstructure type, grain size, and banding level; and / or, Impact performance parameters include impact absorbed energy, lateral expansion rate, and shear ratio; and / or, The thermal cycling parameters for the two-pass thermal simulation include: peak temperature of the two-pass thermal cycle, heating rate of the two-pass thermal cycle, high-temperature residence time of the two-pass thermal cycle, cooling rate of the two-pass thermal cycle, and inter-pass temperature of the two-pass thermal cycle. The sub-characteristic variables include the content of each chemical element, carbon equivalent and cold crack sensitivity coefficient, percentage of granular bainite, ferrite, pearlite and martensite-austenite, grain size of the original structure, banded structure level, peak temperature of the two-pass thermal cycle, heating rate of the two-pass thermal cycle, high-temperature residence time of the two-pass thermal cycle, cooling rate of the two-pass thermal cycle, interpass temperature of the two-pass thermal cycle, impact absorption energy and shear cross-sectional area. The obtained unit is used to reduce the dimensionality of the target feature variables, and the dimensionality-reduced data is input into the machine learning algorithm for training and validation to obtain the machine learning model; The prediction unit is used to predict the impact toughness of the multi-pass heat-affected zone of the target welded material based on the machine learning model.

4. The multi-pass weld heat-affected zone impact toughness prediction device according to claim 3, characterized in that, The target feature variables are dimensionality reduced, and the dimensionality-reduced data is input into a machine learning algorithm for training and validation to obtain the optimal machine learning model, including: The target feature variables are dimensionality reduced to obtain the dimensionality-reduced data. The data after dimensionality reduction is divided into a training set and a test set, and the training set is input into a machine learning algorithm for training to build a machine learning model. The parameters in the machine learning model were optimized using the five-fold cross-validation method. Based on the mean squared error and coefficient of determination of the optimization results of the test set, the machine learning algorithm corresponding to the optimization results when the mean squared error is less than or equal to the first threshold and the coefficient of determination is greater than or equal to the second threshold is taken as the optimal machine learning model.

Images