A temperature distribution real-time high-precision monitoring method for tank structure FSW

Abstract

The application discloses a kind of real-time high-precision monitoring methods for temperature distribution of friction stir welding of storage tank structure.First, according to Latin hypercube sampling to determine the combination of welding process parameters, establish high-precision simulation model of temperature field of friction stir welding of storage tank structure based on Abaqus, based on the way of path point extraction for monitoring surface feature points, storage tank surface two-dimensional temperature field and core area three-dimensional temperature field data.Then based on Kriging surrogate model, the correlation between surface feature points, surface two-dimensional temperature field and core area three-dimensional temperature field is established, and a three-dimensional temperature field prediction model of the core area of friction stir welding based on Kriging surrogate model is established, which realizes real-time high-precision monitoring of temperature distribution of friction stir welding of storage tank structure, guides the temperature control of storage tank welding process, and can be integrated into the digital twin system of friction stir welding.

Description

Technical Field

[0001] This invention belongs to the field of temperature distribution monitoring in friction stir welding (FSW), and relates to a real-time high-precision monitoring method for temperature distribution in tank structure FSW. It comprehensively uses infrared thermal imager measurement, finite element simulation model and Kriging proxy model to monitor the two-dimensional temperature field on the surface of tank structure FSW and the three-dimensional temperature field in the core area. Background Technology

[0002] 2219 aluminum alloy is an age-hardening Al-Cu-Mn aluminum alloy. Due to its high specific strength, excellent high / low temperature load-bearing capacity, superior weldability, and outstanding corrosion resistance, it is frequently used in aerospace tank structural components. Friction stir welding (FSW), as a solid-state joining method, works by using a high-speed rotating stirring head to generate heat through friction with the workpiece. This heat, combined with the heat generated by the plastic deformation of the workpiece material, causes the material to soften. The shoulder of the stirring head generates an upsetting force while preventing the softened material from overflowing. Under the thermo-mechanical coupling, a dense bond is formed, which solidifies upon cooling. Compared to traditional fusion welding, FSW offers higher joint strength and tensile strength, requires no welding wire or shielding gas, and results in lower residual stress after welding, making it the preferred method for welding 2219 aluminum alloy tank structural components.

[0003] In the welding process of storage tanks, the welding heat input is large, the high temperature duration is long, and the temperature field is complex. The temperature field in the welding core area directly affects the plastic flow and microstructure of the material, thus affecting the mechanical properties of the weld joint and even causing welding defects such as cavities. Moreover, the temperature field in the welding core area is also an important basis for studying the plastic flow of materials and optimizing process parameters. Currently, the main methods for obtaining the temperature field in the welding core area are thermocouple measurement, infrared thermal imaging measurement, and numerical simulation. In actual welding processes, thermocouple temperature measurement requires drilling holes to embed thermocouples inside the weldment or stirring head, which limits the number of thermocouples and is not suitable for actual production. Infrared thermal imaging measurement is a non-contact measurement method that can achieve long-distance monitoring of the welding area without damaging the weldment or stirring head, but it can only obtain the temperature field on the outer side of the shoulder on the surface of the weldment. Numerical simulation provides more comprehensive results and can obtain data that is difficult to measure with sensors, but the simulation process is time-consuming and complex, which is not conducive to real-time monitoring of the temperature field in the welding core area. Summary of the Invention

[0004] The technical problem this invention aims to solve is to overcome the shortcomings of existing technologies and to develop a real-time, high-precision monitoring method for the FSW temperature distribution of tank structures, obtaining the temperature distribution in the welding core area for welding mechanism research and temperature control scheme implementation. This invention combines infrared thermal imager surface temperature measurement, FSW temperature field simulation, and a temperature field characterization method based on the Kriging surrogate model. First, welding process parameter combinations are determined based on Latin hypercube sampling, and a high-precision simulation model of the tank structure's FSW temperature field based on Abaqus is established. Surface feature points, the two-dimensional temperature field of the tank surface, and the three-dimensional temperature field of the core area are extracted using a path-based sampling method. Then, the correlation between surface feature points, the two-dimensional surface temperature field, and the three-dimensional core area temperature field is established based on the Kriging surrogate model. During the friction stir welding process of the storage tank, an infrared thermal imager measures the temperature of surface feature points in real time. Combined with the two-dimensional temperature field prediction model of the FSW surface based on the Kriging surrogate model, the two-dimensional temperature field of the storage tank surface is obtained. Based on the three-dimensional temperature field prediction model of the FSW core area based on the Kriging surrogate model, real-time high-precision monitoring of the FSW temperature distribution of the storage tank structure is realized, which guides the temperature control of the storage tank welding process and can be integrated into the friction stir welding digital twin system.

[0005] The technical solution of the present invention:

[0006] A method for real-time, high-precision monitoring of temperature distribution in a storage tank structure FSW includes the following steps:

[0007] Step 1: Based on Abaqus, a simulation model of the FSW temperature field of the tank structure was established using the coupled Euler-Lagrange method and then verified.

[0008] Step 2: Perform Latin hypercube sampling in the sampling space composed of process parameters, and perform FSW temperature field simulation based on the sampling results; extract the dataset of surface feature points for monitoring, the two-dimensional temperature field of the tank surface and the three-dimensional temperature field of the core area from the simulation model based on the path sampling method.

[0009] Step 3: Design a temperature field prediction model for the core region of the FSW of the tank structure based on the Kriging surrogate model, including a two-dimensional temperature field prediction model for the FSW surface of the tank structure and a three-dimensional temperature field prediction model for the core region of the FSW of the tank structure.

[0010] The Kriging surrogate model is an interpolation-based unbiased estimation model with high nonlinear analysis capabilities. Its core idea is that the attribute value of a point in space is related to the attribute values ​​of its surrounding points and can be obtained by interpolation from the attribute values ​​of those surrounding points. The model establishment process involves solving for the weighting factors. The established expression for the temperature field prediction model of the FSW core area of ​​the tank structure can be expressed as:

[0011]

[0012] In the formula, Let z be the output temperature field response value under the desired welding environment. i Given the output temperature field data under a known welding environment, w i is the weight factor, i is the training dataset group number for building the model, and n is the number of training dataset groups for building the model.

[0013] Assuming the sampling space is stationary, a first-order linear function is chosen as the regression model to represent the relationship between the true output temperature field value and the input temperature value, i.e., z = βx + R(x), where z is the true output temperature field value, x is the input temperature data, β is the regression model parameter, and R(x) is a random variable with an expected value of 0 and a variance σ. 2 It is a constant. Based on the above assumptions, to ensure that the error of the output temperature field response value under the welding environment to be determined meets the unbiased requirement, we obtain:

[0014]

[0015] In the formula, x0 represents the input temperature data under the welding environment to be determined, x i Given the input temperature data under the known welding environment, and taking the minimization of the variance of the output temperature field response value under the desired welding environment as the objective function, the expression is established using the Lagrange multiplier method:

[0016]

[0017] In the formula, L(w,λ) is the constructor, γ represents the semivariance, and γ ij =(z i -z j ) 2 / 2 is used to describe the differences between different temperature fields, i and j are the training dataset group numbers for building the model, and λ is the Lagrange multiplier. The following expression is obtained by solving:

[0018]

[0019] Given a dataset As the training set, x i (i = 1, ..., n) represents the input temperature data for a certain set of core area temperature field prediction models, z i (i=1,…,n) represents the output temperature field data of a certain core area temperature field prediction model. The semivariance between the input temperature data under each known welding environment is calculated.

[0020] The commonly used Gaussian model is chosen as the correlation function. The Gaussian model is shown below:

[0021]

[0022] The correlation function parameter θ is solved by semivariance fitting using a pattern search method. θ is a vector with the same dimension as the input temperature data. The semivariance between the input temperature data under the welding environment to be determined and the input temperature data under each known welding environment is solved using the correlation function. Substituting this into equation (4), the weighting factor is solved, and the correlation between the input temperature data and the output temperature field data can be established according to equation (1).

[0023] Step 4: Using the surface feature points and two-dimensional temperature field data of the tank surface obtained in Step 2 as the training set, establish a two-dimensional temperature field prediction model for the FSW surface of the tank structure according to Step 3.

[0024] Step 5: Select m sets of process parameters different from the Latin hypercube sampling results to perform FSW temperature field simulation, extract the datasets of surface feature points, two-dimensional temperature field of tank surface and three-dimensional temperature field of core area for monitoring, and divide them into verification dataset and test dataset.

[0025] Step 6: Using a genetic algorithm based on cross-validation, the initial reference values ​​of the correlation function parameters θ from Step 3 are obtained. The minimum maximum relative error is used as the evaluation criterion. The reference values ​​are evaluated based on the validation dataset from Step 5 to obtain the optimal initial values ​​of the correlation function parameters of the FSW surface two-dimensional temperature field prediction model based on the Kriging surrogate model. These initial values ​​are used for the FSW surface two-dimensional temperature field prediction model for the tank structure established in Step 4, and the model is validated using the test dataset from Step 5.

[0026] Step 7: Similarly, the two-dimensional temperature field data of the tank surface and the three-dimensional temperature field data of the core area obtained in Step 2 are used as training sets. A three-dimensional temperature field prediction model of the FSW core area based on the Kriging surrogate model for the tank structure is established according to Step 3, and the model is validated using the test dataset in Step 5.

[0027] Step 8: Conduct friction stir welding experiments on the tank. Use an infrared thermal imager to measure the temperature of characteristic points on the tank surface in real time. Input the temperature into the two-dimensional temperature field prediction model of the FSW surface established in Step 4 to obtain the two-dimensional temperature field of the tank surface. Then input the two-dimensional temperature field of the surface into the three-dimensional temperature field prediction model of the FSW core area established in Step 7. This will enable in-situ characterization of the three-dimensional temperature field of the FSW core area of ​​the tank structure, and thus realize the monitoring of the two-dimensional temperature field of the FSW surface and the three-dimensional temperature field of the core area of ​​the tank structure based on the infrared thermal imager.

[0028] The beneficial effects of this invention are as follows: It combines infrared thermal imager measurement and numerical simulation methods, and establishes the correlation between surface feature points, surface two-dimensional temperature field and core three-dimensional temperature field based on the Kriging surrogate model. This enables the monitoring of the two-dimensional temperature field of the FSW surface and the three-dimensional temperature field of the core area of ​​the tank structure. The model is verified by simulation data and experimental data, which proves the effectiveness of the method. Attached Figure Description

[0029] Figure 1 This is a basic flowchart of the present invention;

[0030] Figure 2 This is the assembled three-dimensional simulation model in the embodiment. In the figure, 1 is the Lagrange reference stirring head, 2 is the Euler body weldment, 3 is the Lagrange reference body, and 4 is the Euler body empty layer.

[0031] Figure 3 These are the material parameters of 2219 aluminum alloy that change with temperature;

[0032] Figure 4 This is a schematic diagram of the Latin hypercube sampling results;

[0033] Figure 5 This is a schematic diagram of surface feature points used for monitoring;

[0034] Figure 6 This is a schematic diagram of the characteristic points representing the two-dimensional temperature field on the surface of the tank;

[0035] Figure 7(a) is a schematic diagram of the characteristic point cross section representing the three-dimensional temperature field of the core area in the direction perpendicular to the weld seam, and Figure 7(b) is a schematic diagram of the characteristic point cross section representing the three-dimensional temperature field of the core area 2 mm away from the upper surface of the weldment.

[0036] Figure 8 This is the predicted result of the two-dimensional temperature field on the FSW surface of the tank structure. Detailed Implementation

[0037] The specific embodiments of the present invention will now be described in detail with reference to the technical solutions and accompanying drawings.

[0038] In this embodiment, the rocket fuel tank is used as the monitoring object, and the tank material is 2219 aluminum alloy. Due to the large overall size of the tank and the small curvature at the weld seams, the tank welding can be approximated as flat plate welding. By establishing a temperature field prediction model for the FSW core area of ​​the tank structure, the monitoring and in-situ characterization of the temperature field in the FSW core area of ​​the tank structure are realized. The specific steps are as follows:

[0039] (1) Establish a simulation model of the FSW temperature field of the tank structure.

[0040] A geometric model of the stirring head was established using SolidWorks. The structural dimensions and detailed morphological parameters of the stirring head are shown in Table 1. The welding material is 2219 aluminum alloy. A simulation model was established using the coupled Euler-Lagrange method. The Lagrange body dimensions are 150mm × 100mm × 18mm, and the Euler body dimensions are 150mm × 100mm × 24mm. The assembled geometric model is shown below. Figure 2 As shown.

[0041] Table 1. Dimensional parameters of the stirring head

[0042]

[0043] The mass fractions of chemical elements in the 2219 aluminum alloy used were obtained using X-ray fluorescence spectroscopy, as shown in Table 2. The data were imported into JMat-Pro software to obtain the dynamic parameters of the material as a function of temperature, as follows: Figure 3 As shown.

[0044] Table 2. Chemical element mass fraction (wt%) of 2219 aluminum alloy

[0045]

[0046] The Johnson-Cook constitutive model is used to describe the relationship between the rheological stress and temperature, strain, and strain rate of 2219 aluminum alloy, as shown in the following form:

[0047]

[0048] In the formula, Equivalent stress; For equivalent change; T represents the relative equivalent rate of change. * denoted as dimensionless temperature. A represents the yield stress of the material; n represents the influence coefficient of strain hardening; B represents a material-related constant; C represents the strain rate sensitivity coefficient; and m represents the temperature sensitivity coefficient.

[0049] The stirring head is set as a rigid body with a constant temperature, and the heat conduction from the stirring head to the welding machine tool and the heat convection and radiation to the surrounding air are neglected. The heat conduction formula from the workpiece to the backing plate and fixture is set as follows:

[0050]

[0051] In the formula, h w-p The convective heat transfer coefficient between each surface of the weldment and the backing plate and fixture is set to 100 W / m. 2 ℃; T w T represents the temperature of the air heat exchange surface (°C). p This represents the temperature (°C) of the pad and fixture surfaces.

[0052] The formulas for heat dissipation from the weldment via air convection and radiation are as follows:

[0053]

[0054] In the formula: ε is the Stefan-Boltzmann constant; b The emissivity of the object's surface is set to 0.09; γ w-r The convective heat transfer coefficient between the weldment and the air is set to 30 W / m. 2 ℃; T w The temperature of the surface exchanging heat with air; T r The room temperature is set to 20℃.

[0055] When 2219 aluminum alloy is heated above a certain temperature, it becomes viscous and then undergoes plastic flow under the stirring action of the stirring head. At this point, shear friction between the stirring head and the workpiece is the main heat generation mechanism. The contact state between the stirring head and the workpiece is represented by a modified Coulomb friction model that varies with temperature, as shown in equation (9). The friction coefficients are shown in Table 3.

[0056]

[0057] In the formula, τ friction It is frictional shear stress; τ shear Slip shear stress; μ f It is the coefficient of friction; p is the contact pressure; σ is the friction coefficient. s It is the equivalent flow stress.

[0058] Table 3. Friction coefficient as a function of temperature

[0059]

[0060] The Euler body uses a 10-node thermally coupled Euler mesh type from EC3D10MT, specifically designed for Euler bodies. Since the reference body only participates in material assignment and is not used in actual simulation, an 8-node thermally coupled cubic mesh type from C3D8RT with a size of 5mm is selected. Because the stirring head is a three-dimensional solid with threaded and irregularly shaped stirring pins, a 10-node thermally coupled second-order tetrahedral mesh type from C3D10MT is used, with a mesh size of 1mm.

[0061] A single-shoulder FSW experiment was conducted on a gantry-type FSW machine. Thermocouples were placed on the forward and backward sides at a distance of 10 mm from the weld center and 3 mm from the upper surface of the workpiece. The thermal cycle curves were measured at stirring head speeds of 350, 400, and 450 r / min, welding speed of 100 mm / min, and pressing speed of 20 mm / min. The relative error between the measured and simulated values ​​did not exceed 4.5%, verifying the effectiveness of the established tank structure FSW temperature field simulation model.

[0062] (2) Latin hypercube sampling is performed in the sampling space composed of process parameters, and FSW temperature field simulation is performed based on the sampling results. Surface feature points for monitoring, two-dimensional temperature field of tank surface, and three-dimensional temperature field of core area are extracted from the simulation model based on the path sampling method.

[0063] Latin hypercube sampling is a method for generating random sample points. It can uniformly and randomly distribute sample points in the sampling space, achieving high sampling accuracy with fewer sampling attempts and producing more representative sample points. Using Latin hypercube sampling, 20 sample points were extracted from a sampling space defined by the stirring head rotation speed and welding speed, with the rotation speed ranged from 300 r / min to 500 r / min and the welding speed from 60 mm / min to 150 mm / min. The Latin hypercube sampling results were generated using the `lhsdesign` function in MATLAB, as shown below. Figure 4 As shown.

[0064] Based on the Latin hypercube sampling results, FSW temperature field simulation was performed using the tank structure FSW temperature field simulation model established based on the above steps. Twenty sets of surface feature point data, two-dimensional temperature field data of the tank surface, and three-dimensional temperature field data of the core area were extracted for monitoring. Two surface feature points were selected for monitoring, with locations as shown below. Figure 5 As shown in Table 4, the extracted temperature data and corresponding welding process parameters are presented. Twelve feature points characterizing the two-dimensional temperature field on the tank surface were selected, and their locations are shown below. Figure 6 As shown in the figure, the extracted temperature data is shown in Table 5. 48 feature points characterizing the three-dimensional temperature field of the core area were selected, and their locations are shown in Figures 7(a) and 7(b). Partial temperature data extracted from these points is shown in Table 6.

[0065] Table 4 shows the surface feature point temperature data used for monitoring.

[0066]

[0067] Table 5. Temperature data (°C) of characteristic points on the tank surface based on the Latin hypercube characterization of the two-dimensional temperature field.

[0068]

[0069]

[0070] Table 6. Temperature data (°C) of feature points in the three-dimensional temperature field of the core region based on Latin hypercube characterization.

[0071]

[0072] (3) Select several sets of process parameters different from the Latin hypercube sampling results to simulate the FSW temperature field, extract the data sets of surface feature points, two-dimensional temperature field of tank surface and three-dimensional temperature field of core area for monitoring, and divide them into verification dataset and test dataset.

[0073] Eleven sets of welding process parameters, different from those obtained by Latin hypercube sampling, were designed for simulation and temperature field data were extracted. The number and location of surface feature points used for monitoring, representing both the two-dimensional temperature field of the tank surface and the three-dimensional temperature field of the core area, remained unchanged. The temperature data of the surface feature points used for monitoring and the corresponding welding process parameters are shown in Table 7. The temperature data of the feature points representing the two-dimensional temperature field of the tank surface are shown in Table 8. The first seven sets of data were used as the validation dataset, and the last four sets were used as the test dataset. Only the temperature data of the feature points representing the three-dimensional temperature field of the core area from the last four sets were extracted as the test dataset; a portion of the extracted temperature data is shown in Table 9.

[0074] Table 7 shows the surface feature point temperature data and corresponding welding process parameters used for monitoring.

[0075]

[0076] Table 8. Temperature data (°C) of characteristic points representing the two-dimensional temperature field on the tank surface.

[0077]

[0078] Table 9. Feature point temperature test dataset (°C) characterizing the three-dimensional temperature field of the core area.

[0079]

[0080]

[0081] (4) The optimal values ​​of the initial values ​​of the relevant function parameters are obtained by using a genetic algorithm based on cross-validation. A two-dimensional temperature field prediction model for the FSW surface of the tank structure based on the Kriging surrogate model is established and the model is validated.

[0082] In the Kriging surrogate model, the initial values ​​of the correlation function parameters are model hyperparameters. When using pattern search to solve the correlation function, the solution process may get stuck in local optima due to the small dimensionality of the input, directly affecting the fitting accuracy of the correlation function. The Latin hypercube sampling results can be considered representative of each region in the process parameter space. The candidate hyperparameters obtained based on each group of training data are applicable to each region of the sampling space, and the optimal hyperparameters applicable to the entire sampling space also exist among these candidate hyperparameters. Leave-one-out cross-validation is used, dividing the dataset obtained based on the Latin hypercube into the same number of parts as the dataset groups. One part is used as the test set, and the remaining parts as the training set, ensuring that each set of datasets obtained based on the Latin hypercube can serve as a test set. With the minimum maximum relative error as the optimization objective, the initial value of any element of the correlation function parameter is set to range from 0.001 to 30. The genetic algorithm toolbox in MATLAB is used to optimize the hyperparameters used to predict each set of datasets obtained based on the Latin hypercube, obtaining 20 sets of hyperparameter reference values.

[0083] Using the surface feature point temperatures for monitoring as input and the feature point temperatures characterizing the two-dimensional temperature field of the tank surface as output, the surface feature point temperature data for monitoring and the feature point temperature data characterizing the two-dimensional temperature field of the tank surface given in Tables 4 and 5 are used as training sets to establish a prediction model for the two-dimensional temperature field of the FSW surface of the tank structure. Using the maximum relative error as the evaluation index, the initial values ​​of 20 sets of correlation function parameters are compared using the validation datasets given in Tables 7 and 8. The results are shown in Table 10.

[0084] Table 10 Initial Values ​​and Alternative Values ​​for Relevant Function Parameters

[0085]

[0086]

[0087] As shown in Table 10, the initial values ​​of the correlation function parameters in group 10 have the lowest mean evaluation index in the validation set, and can be considered as the optimal hyperparameters applicable to the entire process parameter space. The two-dimensional temperature field prediction model for the FSW surface of the tank structure using the optimal hyperparameters was validated using the test dataset given in Table 8. The maximum relative error between the model predictions and the simulation data did not exceed 5%, and the average relative error did not exceed 3%, demonstrating the effectiveness of the model prediction.

[0088] (5) A three-dimensional temperature field prediction model for the core area of ​​the FSW tank structure based on the Kriging surrogate model was developed and the model was validated.

[0089] Using the surface feature point temperatures for monitoring and the feature point temperatures characterizing the two-dimensional temperature field of the tank surface as inputs, and the feature point temperatures characterizing the three-dimensional temperature field of the core area as outputs, the surface feature point temperature data for monitoring, the feature point temperature data characterizing the two-dimensional temperature field of the tank surface, and the feature point temperature data characterizing the three-dimensional temperature field of the core area given in Tables 4, 5, and 6 are used as training sets. The initial values ​​of all relevant function parameters are set to 15. A prediction model for the three-dimensional temperature field of the FSW core area for the tank structure based on the Kriging surrogate model is established. The prediction model for the three-dimensional temperature field of the FSW core area for the tank structure is validated using the test dataset given in Table 9. The maximum relative error between the model prediction values ​​and the simulation data does not exceed 6%, and the average relative error does not exceed 2%, proving the effectiveness of the model prediction.

[0090] (6) Conduct FSW experiments on the tank, use an infrared thermal imager to measure the surface temperature of the tank, and use the experimental data to verify the FSW temperature field prediction model used for the tank structure.

[0091] FSW experiments were conducted on an FSW-5M FSW experimental machine tool. The welding parameters used in the experiment are shown in Table 11. The surface feature point temperatures were measured using an FLIRA615 thermal imager. Four feature points located 30 mm from both sides of the weld center were measured using an infrared thermal imager. These feature points correspond to feature points 1, 2, 11, and 12, which characterize the two-dimensional temperature field of the tank surface. The prediction model for the two-dimensional temperature field of the FSW surface was validated, and the results are shown in Table 11. Figure 8 .

[0092] Table 11 Experimental welding process parameters

[0093]

[0094] For the 2219 aluminum alloy tank FSW, the welding process parameter combination was first determined based on Latin hypercube sampling, and a high-precision simulation model of the tank structure FSW temperature field based on Abaqus was established. Surface feature points, the two-dimensional temperature field of the tank surface, and the three-dimensional temperature field of the core area were extracted for monitoring using a path sampling method. Then, the correlation between surface feature points, the two-dimensional temperature field of the surface, and the three-dimensional temperature field of the core area was established based on the Kriging surrogate model. Finally, tank FSW experiments were conducted, using the surface feature point temperature monitored by an infrared thermal imager as input, to achieve real-time high-precision monitoring of the two-dimensional temperature field of the surface and the three-dimensional temperature field of the core area. The average relative error between the model prediction and the measured value did not exceed 4%, and the maximum relative error did not exceed 7%, demonstrating the effectiveness of this method for real-time high-precision monitoring of FSW temperature distribution in tank structures.

Claims

1. A method for real-time, high-precision monitoring of temperature distribution during friction stir welding of storage tank structures, characterized in that, Includes the following steps: Step 1: Establish and verify the temperature field simulation model of the tank structure for friction stir welding using the coupled Euler-Lagrange method based on Abaqus; Step 2: Perform Latin hypercube sampling in the sampling space composed of process parameters, and simulate the temperature field of friction stir welding based on the sampling results; Data sets of surface feature points, two-dimensional temperature field of tank surface and three-dimensional temperature field of core area are extracted from simulation model based on path point sampling method; Step 3: Design a temperature field prediction model for the core area of ​​friction stir welding of tank structure based on Kriging proxy model, including a two-dimensional temperature field prediction model for the surface of friction stir welding of tank structure and a three-dimensional temperature field prediction model for the core area of ​​friction stir welding of tank structure. In the Kriging proxy model, the attribute value of a point in space is related to the attribute values ​​of its surrounding points and can be obtained by interpolation of the attribute values ​​of its surrounding points; the model building process is the process of solving for the weight factors. The established expression for the temperature field prediction model of the core area of ​​friction stir welding of the tank structure is as follows: In the formula, z represents the output temperature field response value under the desired welding environment. i The output temperature field data is given under a known welding environment; w i As a weighting factor; i represents the training dataset group number for building the model; n is the number of training dataset groups used to build the model; Assuming the sampling space is stationary, a first-order linear function is chosen as the regression model to represent the relationship between the true value of the output temperature field and the input temperature value, i.e., z = βx + R(x), where R(x) is a random variable with an expected value of 0 and a variance σ. 2 The value is a constant; based on the above assumptions, to ensure that the error of the output temperature field response value under the welding environment to be determined meets the unbiased requirement, we obtain: In the formula, x0 represents the input temperature data under the welding environment to be determined; x i Given the input temperature data under the known welding environment, and taking the minimization of the variance of the output temperature field response value under the desired welding environment as the objective function, the expression is established using the Lagrange multiplier method: In the formula, L(w,λ) is the constructor, γ represents the semivariance, and γ ij =(z i -z j ) 2 / 2 is used to describe the differences between different temperature fields; i and j are the group numbers of the training dataset used to build the model, and λ is the Lagrange multiplier. The following expression is obtained by solving: In the formula, Given a dataset As the training set, x i The input temperature data for a certain set of core area temperature field prediction models; z i For the output temperature field data of a certain core area temperature field prediction model, calculate the semivariance between the input temperature data under various known welding environments; A Gaussian model is chosen as the correlation function. The Gaussian model is shown below: The correlation function parameter θ is solved by using the pattern search method to fit the semivariance. θ is a vector with the same dimension as the input temperature data. The semivariance between the input temperature data under the welding environment to be determined and the input temperature data under each known welding environment is solved by using the correlation function. Substitute it into equation (4) to solve the weight factor. Then, the correlation between the input temperature data and the output temperature field data can be established according to equation (1). Step 4: Using the surface feature points and two-dimensional temperature field data of the tank surface obtained in Step 2 as the training set, establish a two-dimensional temperature field prediction model for the friction stir welding surface of the tank structure according to Step 3. Step 5: Select m sets of process parameters different from the Latin hypercube sampling results to simulate the temperature field of friction stir welding, extract the datasets of surface feature points, two-dimensional temperature field of tank surface and three-dimensional temperature field of core area for monitoring, and divide them into verification dataset and test dataset. Step 6: Using a genetic algorithm based on cross-validation, the initial reference values ​​of the correlation function parameters θ from Step 3 are obtained. The minimum maximum relative error is used as the evaluation criterion. The reference values ​​are evaluated based on the validation dataset from Step 5 to obtain the optimal initial values ​​of the correlation function parameters of the two-dimensional temperature field prediction model of the friction stir welding surface based on the Kriging surrogate model. These initial values ​​are used for the two-dimensional temperature field prediction model of the friction stir welding surface for the tank structure established in Step 4, and the model is validated using the test dataset from Step 5. Step 7: Use the two-dimensional temperature field data of the tank surface and the three-dimensional temperature field data of the core area obtained in Step 2 as the training set, and establish a three-dimensional temperature field prediction model of the core area of ​​friction stir welding for tank structure based on the Kriging surrogate model according to Step 3, and verify the model using the test dataset in Step 5. Step 8: Conduct friction stir welding experiments on the storage tank. Use an infrared thermal imager to measure the temperature of characteristic points on the surface of the storage tank in real time, and input it into the two-dimensional temperature field prediction model of the friction stir welding surface to obtain the two-dimensional temperature field of the storage tank surface. Then input the two-dimensional temperature field of the surface into the three-dimensional temperature field prediction model of the core area of ​​the friction stir welding to obtain the three-dimensional temperature field of the core area of ​​the storage tank structure. This realizes the monitoring of the two-dimensional temperature field of the friction stir welding surface and the three-dimensional temperature field of the core area of ​​the storage tank structure based on the infrared thermal imager.

Images