Laser welding three-dimensional temperature field reconstruction and all-weld forming prediction method and system based on multi-source data driving

Abstract

The invention belongs to the field of laser welding numerical simulation, and discloses a laser welding three-dimensional temperature field reconstruction and all-weld forming prediction method and system based on multi-source data driving. Multi-source experimental data are obtained by adopting experimental means such as high-energy synchrotron radiation, a high-speed camera, an infrared camera and a thermocouple, three-dimensional numerical reconstruction of the morphology and the temperature field of the laser welding pool is realized, and a relation between real experimental data and a virtual numerical space is established; a parallel solving thought is adopted, and efficient three-dimensional reconstruction of the morphology and the temperature field of the welding pool at different observation moments is achieved through the MPI parallel computing technology; introducing a reconstruction result into the numerical model as an initial condition, and predicting the evolution behavior of the molten pool temperature field at a subsequent moment; and the results of different parallel processes are integrated to obtain a forming prediction result of all welding seams, so that high-precision and high-efficiency prediction of laser welding seam forming is realized, and a convenient and effective forming prediction method is provided for optimization of a laser welding process.

Description

Technical Field

[0001] The present invention belongs to but is not limited to the field of laser processing numerical simulation technology, and in particular relates to a method and system for laser welding three-dimensional temperature field reconstruction and full weld formation prediction based on multi-source data drive. Background Art

[0002] Predicting laser welding weld formation is crucial for optimizing welding process parameters and improving weld quality. Numerical simulation, with its low cost, high efficiency, and visual description of the multi-physics field of the welding process, is the leading technology of choice for studying the evolution of laser welding physical fields and predicting weld formation. By considering complex coupled physical phenomena such as the interaction between the laser and the metal material, the melting and solidification of the metal, and heat transfer, numerical simulation technology can simulate the dynamic behavior of the laser welding keyhole-molten pool temperature field and predict the geometric morphology of the weld during the laser welding process.

[0003] The establishment of numerical models requires the simplification of many complex physical phenomena to improve the model's solvability, which results in differences between the simulation process and the actual welding situation, leading to large deviations in the weld formation prediction results. To eliminate this difference, some researchers have used methods to adjust model parameters (such as laser absorption coefficient, grid size, material physical properties, etc.) for specific welding processes to make the simulation results consistent with the actual weld formation effect. However, their prediction efficiency is low and they cannot achieve rapid prediction of weld formation. Another group of researchers proposed a data-driven weld formation prediction method, but its prediction accuracy is greatly affected by the quality and quantity of data, and its prediction accuracy is low and cannot meet the requirements of high-precision weld formation prediction.

[0004] In view of the above analysis, the technical problems that need to be solved urgently in the existing technology are:

[0005] How to establish the connection between the numerical model and the actual welding conditions, map the actual physical field data into the virtual numerical space to compensate for the deviation of the prediction results, and take into account the prediction efficiency to achieve rapid prediction of weld formation, and provide a "high-efficiency" and "high-precision" laser welding weld formation prediction method for the optimization of laser welding process parameters has become an urgent problem to be solved. Summary of the Invention

[0006] In response to the problems existing in the prior art, the present invention provides a method and system for reconstructing the three-dimensional temperature field of laser welding and predicting the entire weld formation based on multi-source data driving.

[0007] The present invention is achieved by providing a method for reconstructing the three-dimensional temperature field of laser welding and predicting the full weld shape based on multi-source data. The method is characterized in that the method comprises:

[0008] S1: During the laser welding process, high-energy synchrotron radiation technology is used to capture the laser welding pool from multiple angles, and a high-speed camera is used to capture the upper surface of the laser welding pool in the welding direction. Through image processing, the contours of the keyhole and the molten pool extension at various angles are extracted. Voxel carving technology is used to restore the 3D structure of the keyhole and the molten pool's outer contours. The 3D structure of the welding pool is obtained by removing the keyhole structure from the outer contour structure.

[0009] S2: Infrared temperature measurement technology is used to measure the temperatures of the upper and lower surfaces of the weld pool. Combined with the three-dimensional structure of the weld pool, the three-dimensional boundary temperature distribution of the weld pool is obtained. Embedded thermocouples are used to extract the actual internal temperature data of the weld pool. The temperature data of an appropriate number of grids in the high-temperature zone of the numerical model are randomly selected to obtain the three-dimensional internal temperature distribution of the weld pool.

[0010] S3: Uniformly select boundary temperatures and 3D internal temperature distributions to form a temperature field reconstruction dataset. A radial basis function interpolation method is used to construct the 3D temperature field of laser welding. Based on the 3D structure of the molten pool, 3D grid points are generated to restrict the interpolation result to the interior of the molten pool. The remaining boundary data points are used to evaluate the model prediction error. By adjusting the basis function parameters, a 3D temperature field reconstruction model with high prediction accuracy is obtained.

[0011] S4: Extract laser welding experimental data and reconstruct the 3D temperature field of each cross-sectional observation window in parallel. For the i-th observation window, use the reconstructed temperature field as the initial condition of the laser welding heat conduction model, extract the 3D temperature field structure information to construct a heat source model, and calculate the temperature field change between observation windows i to i+1. Extract the historical temperature information of the thermocouple, compare it with the model prediction results, and adjust the model parameters.

[0012] S5: Merge the prediction results between each observation window to achieve efficient calculation of the three-dimensional temperature of the entire weld seam; extract model results such as penetration depth and width to achieve full weld seam formation prediction in laser welding.

[0013] Furthermore, the step S1, using voxel carving technology to restore the three-dimensional structure of the welding pool, specifically includes the following steps:

[0014] S11: Using ImageJ software, the orthogonal melt pool observation image is subjected to noise reduction, contrast enhancement, and binarization processing to extract the morphological features of the keyhole and the outer contour of the melt pool. In the orthogonal melt pool observation image, the melt pool is dark and the background is light, so the keyhole morphology information can be easily extracted through image processing. There is a shallow boundary between the melt pool boundary and the base material. After contrast enhancement processing, the solid-liquid boundary of the melt pool is highlighted to obtain the outer contour information of the melt pool (i.e., the solid-liquid boundary of the melt pool).

[0015] S12: Using voxel carving technology, the three-dimensional space is discretized into voxel space, and the initial value of the voxel is 1;

[0016]

[0017] S13: Convert the binary images of the three orthogonal projection planes into voxel matrices:

[0018]

[0019] Among them, the voxel value at the keyhole and the outer contour of the molten pool is 1, and the rest is 0;

[0020] S14: Perform the following operations on each projection direction of the three-dimensional voxel space to obtain the voxel space distribution of the three-dimensional structure of the melt pool:

[0021] V`(x,y,z)=V(x,y,z)·M

[0022] Eliminate all voxels in the voxel space that do not meet the projection constraints, and finally restore the three-dimensional structure of the keyhole and the outer contour of the molten pool;

[0023] S15: The three-dimensional structure of the welding pool can be obtained by removing the keyhole structure from the outer contour structure of the molten pool.

[0024] Furthermore, the S2, using infrared camera technology to construct the laser welding molten pool boundary temperature distribution, and using embedded thermocouples combined with simulation data to construct the laser welding molten pool internal temperature distribution, specifically includes the following steps:

[0025] S21: using infrared camera technology to extract the temperature distribution of the upper surface of the welding pool except the keyhole area, and obtain the boundary temperature distribution of the upper surface of the welding pool;

[0026] S22: The temperature distribution of the high-temperature area is extracted using image processing technology. Assuming that the keyhole temperature is Gaussian, the three-dimensional temperature distribution is as follows:

[0027]

[0028] x i is the three-dimensional coordinate of the data point, u i is the mean of the temperature data in each direction, is the variance of the temperature data in each direction. Here, it is assumed that the variables in each direction are independent. In the height direction, the mean and variance in the height direction are obtained by combining the temperature distribution of the lower surface with the temperature distribution of the upper surface. Combined with the three-dimensional structure of the keyhole, the three-dimensional temperature distribution of the keyhole area of ​​the molten pool is obtained.

[0029] S23: Assuming that the remaining boundaries of the molten pool are the solid-liquid interface, the temperature is set to the melting point of the base material, and the above temperature information is integrated with the three-dimensional structure of the molten pool to obtain the three-dimensional boundary temperature distribution of the welding molten pool;

[0030] S24: Use embedded thermocouples to extract the true internal temperature data of the weld pool. Assuming that the laser welding molten pool is symmetrical, more internal temperature data of the molten pool can be supplemented by embedding thermocouples at different depths on both sides of the material.

[0031] S25: Randomly select the temperature data of an appropriate number of grids inside the laser welding numerical model molten pool, and combine them with the thermocouple temperature data to form the three-dimensional internal temperature data of the laser welding molten pool.

[0032] Furthermore, the S3 uses a radial basis function method to construct a three-dimensional temperature field reconstruction model for laser welding, which specifically includes the following steps:

[0033] S31: uniformly select some boundary temperature grid points and form a temperature field reconstruction data set with the three-dimensional internal temperature data of the welding pool;

[0034] S32: Use radial basis function interpolation method to construct the three-dimensional temperature field of laser welding, assuming that there are N known data points The temperature calculation formula for a data point is as follows:

[0035]

[0036] In the formula, ∈ is the shape parameter of the basis function, which affects the accuracy of temperature field reconstruction; λ is the weight coefficient, which is obtained by solving the weight vector. The formula is as follows:

[0037] Φλ=T

[0038] Φ is a symmetric matrix, which is an N×N matrix composed of known data points, and its elements are the distance r between two points. λ is the weight vector, which represents the weight of each known data point; T is the temperature matrix, which represents the temperature value of each known data point; according to the spatial distribution of voxels in the three-dimensional structure of the melt pool, the interpolation result is restricted to the inside of the melt pool, and the formula is as follows:

[0039] T V =T(x,y,z)·V`(x,y,z)

[0040] The voxel value inside the molten pool is 1, and the interpolation process proceeds normally; the voxel value outside the molten pool is 0, and the interpolation result is 0;

[0041] S33: The remaining boundary data points are used to evaluate the model prediction error. By adjusting the shape parameters of the basis function, the radial basis function is iteratively optimized to obtain a three-dimensional temperature field reconstruction model with high prediction accuracy.

[0042] Furthermore, the S4, parallel reconstruction of the three-dimensional temperature field of each time node, specifically includes the following steps:

[0043] S41: Conduct welding experiments, obtain full weld seam experimental data, construct a finite element laser welding temperature field evolution model, and obtain full weld seam temperature field simulation data;

[0044] S42: Using MPI parallel computing technology, construct a temperature field reconstruction data set at the corresponding moment for each observation window using the data processing methods described in S1 and S2, and realize three-dimensional reconstruction of the temperature field using the method described in S3;

[0045] S43: The reconstructed temperature field is used as the initial condition of the laser welding temperature field evolution model. The historical temperature data of the embedded thermocouple is used to evaluate the prediction results of each time step. The model prediction deviation is corrected by dynamically adjusting model parameters such as laser absorption coefficient and thermal conductivity.

[0046] S44: Finally, the reconstructed temperature field evolution data between each observation window is obtained.

[0047] Furthermore, the S5 extracts the prediction results of each parallel branch through MPIGather and outputs them as a dat file, merges the prediction results between each observation window together, and realizes the efficient calculation of the three-dimensional temperature field evolution behavior of the entire weld; extracts model results such as penetration depth and penetration width to realize the prediction of the entire weld formation in laser welding.

[0048] Another object of the present invention is to provide a laser welding three-dimensional temperature field reconstruction and full weld formation prediction system based on multi-source data drive, the system specifically comprising:

[0049] The data acquisition module uses high-energy synchrotron radiation, high-speed cameras, infrared cameras, thermocouples and other experimental methods to obtain multi-source experimental data;

[0050] The three-dimensional numerical reconstruction module adopts a parallel solution approach and uses MPI parallel computing technology to achieve efficient three-dimensional reconstruction of the welding pool morphology and temperature field at different observation times;

[0051] The prediction module introduces the reconstruction results as initial conditions into the numerical model, predicts the evolution behavior of the molten pool temperature field at subsequent moments, and integrates the results of different parallel processes to obtain the formation prediction results of the entire weld.

[0052] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0053] The present invention uses a variety of experimental methods to obtain multi-source experimental data, including keyhole-molten pool morphology and temperature information. Combining voxel carving and radial basis interpolation, the method achieves high-precision reconstruction of the three-dimensional temperature field of the laser welding molten pool. This establishes a connection between the actual welding conditions and the virtual numerical space, laying the foundation for high-precision laser welding weld formation prediction.

[0054] The present invention adopts MPI parallel computing technology to reconstruct the temperature field of the laser welding molten pool in different observation windows, and uses the historical temperature data of embedded thermocouples to correct the numerical model, thus achieving efficient and high-precision prediction of the entire laser welding weld formation.

[0055] This invention provides a feasible digital twin approach for the laser welding field. Previous research employed independent data-driven approaches and mechanism-based modeling, lacking an effective connection between real experimental data and virtual numerical space. This method utilizes multi-source experimental data to reconstruct the three-dimensional temperature field of the laser welding molten pool and uses historical temperature information to correct the mechanism model, achieving a digital twin approach that can simultaneously map physical and virtual space.

[0056] The technical solution of this invention fills a gap in experimental data-driven prediction technology for full-weld seam formation in laser welding, both domestically and internationally. Specifically, it provides a three-dimensional reconstruction method for the laser welding molten pool and temperature field based on experimental observation data. This is then incorporated into a numerical model using this initial condition, enabling efficient prediction of full-weld seam formation using parallel computing.

[0057] The technical solution of the present invention overcomes the technical bias of large prediction errors in laser welding full-weld prediction technology based on numerical simulation. This is specifically manifested in the fact that the establishment of numerical models requires the simplification of many complex physical phenomena to improve the model's solvability, resulting in discrepancies between the simulation process and actual welding conditions, leading to large deviations in weld formation prediction results. The present invention obtains multi-source experimental data through different experimental methods, achieving a three-dimensional reconstruction of the laser welding molten pool temperature field, and introduces this data as the initial condition into the numerical model, compensating for this deviation and resolving this technical bias. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 Schematic diagram of the experimental equipment and observation direction provided by an embodiment of the present invention;

[0059] Figure 2 This is a module diagram of a laser welding three-dimensional temperature field reconstruction and full weld formation prediction system based on multi-source data drive provided by an embodiment of the present invention;

[0060] Figure 3 This is a flow chart of a method for reconstructing a three-dimensional temperature field and predicting full weld formation in laser welding based on multi-source data drive, provided by an embodiment of the present invention;

[0061] Figure 4 is a schematic diagram of three-dimensional temperature field reconstruction provided by an embodiment of the present invention;

[0062] Figure 5 is a flowchart of parallel solution provided by an embodiment of the present invention;

[0063] Figure 6 This is a comparison chart of prediction results of a certain example provided by an embodiment of the present invention;

[0064] In the figure: 1. Embedded thermocouple; 2. Molten pool; 3. High-speed camera; 4. Infrared camera; 5. Laser beam; 6. High-energy synchrotron radiation beam in XY direction; 7. XY observation window; 8. Keyhole. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0066] The proposed method for three-dimensional laser welding temperature field reconstruction and full weld formation prediction, driven by multi-source data, aims to address key technical challenges in existing laser welding processes, including insufficient dimensionality in temperature field acquisition, low reconstruction accuracy, inaccurate molten pool modeling, and poor continuity in weld formation prediction. Traditional welding temperature field monitoring methods often rely on single thermocouple measurement points or infrared array imaging, which cannot fully capture the complex temperature gradients and structural evolution within the molten pool. This is particularly true during high-power laser rapid welding, where spatial non-uniformity caused by the keyhole effect, evaporation-driven flow, and intense heat conduction interactions significantly impairs the practical applicability and predictive capabilities of numerical models.

[0067] During the 3D melt pool modeling process, this method incorporates an orthogonal observation mechanism using high-energy synchrotron radiation and high-speed videography to acquire keyhole and melt pool boundary morphology data from multiple perspectives. After extracting the 2D contours through image processing, voxel carving is used to restore the 3D structure of the melt pool and keyhole. Unlike approximate surface projection reconstruction methods, voxel carving simultaneously satisfies three orthogonal projection constraints, eliminating voxel information that does not conform to the actual structural projection rules. This results in a high-resolution, highly realistic 3D representation of the structure, providing physical boundary constraints for subsequent temperature field interpolation modeling.

[0068] To acquire the melt pool temperature, a multi-source boundary and internal temperature fusion model was constructed. The upper surface temperature field was acquired using infrared thermal imaging, and the internal temperature distribution in the keyhole region was simulated using a three-dimensional Gaussian function. The lower surface temperature was fitted using thermocouple signals. The boundary temperature was fixed to the material melting point and converted into a three-dimensional boundary constraint through spatial mapping. The internal melt pool temperature data was enhanced using thermocouple arrays placed at different depths, taking into account the symmetry of the welding process. The data was then fused with data from representative temperature points in the finite element mesh to form a dense internal temperature sample set.

[0069] To address the nonlinear and localized mutation characteristics of temperature field reconstruction, this paper uses radial basis function (RBF) interpolation to construct a three-dimensional temperature field model. This method uses a high-dimensional Euclidean distance function to achieve nonlinear fitting of known temperature points. A voxel mask is introduced to restrict the interpolation results to the melt pool voxel space, thereby ensuring physical rationality and spatial coherence. The shape parameters in the interpolation process have a significant impact on prediction accuracy. Through iterative adjustment and the introduction of a boundary error feedback mechanism, the numerical stability and reconstruction accuracy of the model are further improved.

[0070] To achieve a full-time temperature evolution description of the welding process, this paper constructs a parallel temperature reconstruction framework based on sliding observation windows. An MPI parallel computing strategy is used to reconstruct the local three-dimensional temperature field within each observation window. The reconstruction results serve as the initial conditions for the subsequent heat conduction model. The model's thermophysical properties, such as laser absorptivity, thermal conductivity, and material specific heat, are dynamically corrected using historical thermocouple data. The time-step temperature field output by the heat conduction model is compared with experimental data and automatically corrected for any deviations, achieving continuous and accurate reconstruction of the temperature field across the entire weld area.

[0071] Ultimately, the MPIGather mechanism integrates simulated data from multiple observation windows to generate complete three-dimensional weld temperature evolution information. Based on the reconstruction results, key weld formation parameters such as penetration depth, weld width, and heat-affected zone contour are extracted, enabling a comprehensive visual prediction of weld formation quality. This method overcomes the limitations of traditional temperature field acquisition and discrete prediction results, providing a new technical path for high-precision welding control and process inversion.

[0072] like Figure 1 As shown, an embodiment of the present invention provides a method for reconstructing a three-dimensional temperature field of laser welding and predicting full weld formation based on multi-source data. The method specifically includes:

[0073] S1: During the laser welding process, high-energy synchrotron radiation technology is used to shoot the laser welding molten pool from multiple angles, and a high-speed camera is used to shoot the upper surface of the laser welding molten pool in the welding direction (the equipment observation direction is as follows Figure 1 By image processing, the contours of the keyhole and the molten pool at various angles are extracted; the voxels culpting technology is used to restore the three-dimensional structure of the keyhole and the outer contour of the molten pool. The three-dimensional structure of the welding pool can be obtained by removing the keyhole structure from the outer contour structure;

[0074] S2: Infrared temperature measurement technology is used to measure the temperatures of the upper and lower surfaces of the weld pool. Combined with the three-dimensional structure of the weld pool, the three-dimensional boundary temperature distribution of the weld pool is obtained. Embedded thermocouples are used to extract the actual internal temperature data of the weld pool. The temperature data of an appropriate number of grids in the high-temperature zone of the numerical model are randomly selected to obtain the three-dimensional internal temperature distribution of the weld pool.

[0075] S3: Uniformly select boundary temperatures and 3D internal temperature distributions to form a temperature field reconstruction dataset. A radial basis function (RBF) interpolation method is used to construct the 3D temperature field of laser welding. Based on the 3D structure of the molten pool, 3D grid points are generated to restrict the interpolation result to the interior of the molten pool. The remaining boundary data points are used to evaluate the model prediction error. By adjusting the basis function parameters, a 3D temperature field reconstruction model with high prediction accuracy is obtained.

[0076] S4: Extract laser welding experimental data and reconstruct the three-dimensional temperature field of each cross-sectional observation window in parallel. For the i-th observation window, use the reconstructed temperature field as the initial condition of the laser welding heat conduction model, extract the three-dimensional temperature field structure information to construct a heat source model, and calculate the temperature field change between observation windows i to i+1. Extract the historical temperature information of the thermocouple, compare it with the model prediction results, and adjust the model parameters to ensure the accuracy of the prediction results.

[0077] S5: Merge the prediction results between each observation window to achieve efficient calculation of the three-dimensional temperature of the entire weld seam; extract model results such as penetration depth and width to achieve full weld seam formation prediction in laser welding.

[0078] The above equipment observation direction is as follows Figure 1 As shown in the figure, the weld center plane and a set cross-section of the molten pool are observed using high-energy synchrotron radiation. High-speed camera technology supplements information on the upper surface of the molten pool and the keyhole opening, forming three orthogonal molten pool observation images. Cross-sectional molten pool information is acquired within a set observation window. Therefore, accurate three-dimensional reconstruction of the temperature field occurs only within a specific observation window. To ensure the correctness of the temperature field reconstruction, the experimental data is processed in a unified space-time, and the numerical model is time-space correspondence with the physical space by calculating the simulation time step.

[0079] S1, using voxel carving technology to restore the three-dimensional structure of the welding pool, specifically includes the following steps:

[0080] S11: Using ImageJ software, the orthogonal melt pool observation image is subjected to noise reduction, contrast enhancement, and binarization processing to extract the morphological features of the keyhole and the outer contour of the melt pool. In the orthogonal melt pool observation image, the melt pool is dark and the background is light, so the keyhole morphology information can be easily extracted through image processing. There is a shallow boundary between the melt pool boundary and the base material. After contrast enhancement processing, the solid-liquid boundary of the melt pool is highlighted to obtain the outer contour information of the melt pool (i.e., the solid-liquid boundary of the melt pool).

[0081] S12: Using voxel carving technology, the three-dimensional space is discretized into voxel space, and the initial value of the voxel is 1;

[0082]

[0083] S13: Convert the binary images of the three orthogonal projection planes into voxel matrices:

[0084]

[0085] The voxel value at the keyhole and the outer contour of the molten pool is 1, and the rest is 0.

[0086] S14: Perform the following operations on each projection direction of the three-dimensional voxel space to obtain the voxel space distribution of the three-dimensional structure of the melt pool:

[0087] V`(x,y,z)=V(x,y,z)·M

[0088] Eliminate all voxels in the voxel space that do not meet the projection constraints, and finally restore the three-dimensional structure of the keyhole and the outer contour of the molten pool;

[0089] S15: The three-dimensional structure of the welding pool can be obtained by removing the keyhole structure from the outer contour structure of the welding pool;

[0090] S2, using infrared camera technology to construct the temperature distribution at the boundary of the laser welding molten pool, and using embedded thermocouples combined with simulation data to construct the temperature distribution inside the laser welding molten pool, specifically includes the following steps:

[0091] S21: using infrared camera technology to extract the temperature distribution of the upper surface of the welding pool except the keyhole area, and obtain the boundary temperature distribution of the upper surface of the welding pool;

[0092] S22: The temperature distribution of the high-temperature area is extracted using image processing technology. Assuming that the keyhole temperature is Gaussian, the three-dimensional temperature distribution is as follows:

[0093]

[0094] x i is the three-dimensional coordinate of the data point, u i is the mean of the temperature data in each direction, is the variance of the temperature data in each direction, assuming that the variables in each direction are independent. In the height direction, the mean and variance in the height direction are obtained by combining the temperature distribution of the lower surface with the temperature distribution of the upper surface. Combined with the three-dimensional structure of the keyhole, the three-dimensional temperature distribution of the keyhole area of ​​the molten pool is obtained;

[0095] S23: Assuming that the remaining boundaries of the molten pool are the solid-liquid interface, the temperature is set to the melting point of the base material, and the above temperature information is integrated with the three-dimensional structure of the molten pool to obtain the three-dimensional boundary temperature distribution of the welding molten pool;

[0096] S24: Use embedded thermocouples to extract the true internal temperature data of the weld pool. Assuming that the laser weld pool is symmetrical (many literatures show that the laser weld pool of the same metal has a standard symmetrical distribution), more internal temperature data of the weld pool can be supplemented by embedding thermocouples at different depths on both sides of the material.

[0097] S25: Randomly select the temperature data of an appropriate number of grids inside the laser welding numerical model molten pool, and combine them with the thermocouple temperature data to form the three-dimensional internal temperature data of the laser welding molten pool;

[0098] S3, using radial basis function method to construct a three-dimensional temperature field reconstruction model for laser welding, specifically includes the following steps:

[0099] S31: uniformly select some boundary temperature grid points and form a temperature field reconstruction data set with the three-dimensional internal temperature data of the welding pool;

[0100] S32: Use radial basis function (PBF) to construct the three-dimensional temperature field of laser welding. Assuming there are N known data points The temperature calculation formula for a data point is as follows:

[0101]

[0102] In the formula, ∈ is the shape parameter of the basis function, which affects the accuracy of temperature field reconstruction; λ is the weight coefficient, which is obtained by solving the weight vector. The formula is as follows:

[0103] Φλ=T

[0104] Φ is a symmetric matrix, which is an N×N matrix composed of known data points, and its elements are the distance r between two points. λ is the weight vector, which represents the weight of each known data point; T is the temperature matrix, which represents the temperature value of each known data point; according to the spatial distribution of voxels in the three-dimensional structure of the melt pool, the interpolation result is restricted to the inside of the melt pool, and the formula is as follows:

[0105] T V =T(x,y,z)·V`(x,y,z)

[0106] The voxel value inside the molten pool is 1, and the interpolation process proceeds normally; the voxel value outside the molten pool is 0, and the interpolation result is 0, thereby ensuring that the interpolation process value occurs inside the molten pool.

[0107] S33: The remaining boundary data points are used to evaluate the model prediction error. By adjusting the basis function shape parameters, the radial basis function is iteratively optimized to obtain a three-dimensional temperature field reconstruction model with high prediction accuracy.

[0108] Said S4, parallel reconstruction of the three-dimensional temperature field of each time node, specifically comprises the following steps:

[0109] S41: Conduct welding experiments using Figure 1 The equipment shown acquires full weld seam experimental data, constructs a finite element laser welding temperature field evolution model, and obtains full weld seam temperature field simulation data;

[0110] S42: Using MPI parallel computing technology, construct a temperature field reconstruction data set at the corresponding moment for each observation window using the data processing methods described in S1 and S2, and realize three-dimensional reconstruction of the temperature field using the method described in step 3;

[0111] S43: The reconstructed temperature field is used as the initial condition of the laser welding temperature field evolution model. The historical temperature data of the embedded thermocouple is used to evaluate the prediction results of each time step. The model prediction deviation is corrected by dynamically adjusting model parameters such as laser absorption coefficient and thermal conductivity.

[0112] S44: Finally, the reconstructed temperature field evolution data between each observation window is obtained;

[0113] The S5 extracts the prediction results of each parallel branch through MPIGather and outputs them as a dat file, merges the prediction results between each observation window together, and realizes the efficient calculation of the three-dimensional temperature field evolution behavior of the entire weld; extracts model results such as penetration depth and penetration width to realize the prediction of the entire weld formation of laser welding.

[0114] like Figure 2 As shown, an embodiment of the present invention provides a laser welding three-dimensional temperature field reconstruction and full weld formation prediction system based on multi-source data drive, which specifically includes:

[0115] The data acquisition module uses high-energy synchrotron radiation, high-speed cameras, infrared cameras, thermocouples and other experimental methods to obtain multi-source experimental data;

[0116] The three-dimensional numerical reconstruction module adopts a parallel solution approach and uses MPI parallel computing technology to achieve efficient three-dimensional reconstruction of the welding pool morphology and temperature field at different observation times;

[0117] The prediction module introduces the reconstruction results as initial conditions into the numerical model, predicts the evolution behavior of the molten pool temperature field at subsequent moments, and integrates the results of different parallel processes to obtain the formation prediction results of the entire weld.

[0118] The experimental equipment and observation directions involved in this embodiment are as follows Figure 1 The overall flow chart of the method is shown in Figure 3 shown.

[0119] like Figure 1The experimental equipment and observation directions shown in the figure are used in this embodiment of the present invention to capture images of the keyhole and the weld pool in the XY and XZ directions of the laser welding molten pool. A high-speed camera is used to follow the movement of the laser welding head to capture images of the weld pool and keyhole opening in the YZ directions. These images are processed to obtain corresponding binary images, which are then converted into voxel matrices. All voxels in the voxel space that do not meet the projection constraints are removed, ultimately restoring the three-dimensional structure of the keyhole and weld pool's outer contours. By removing the keyhole structure from the outer contour of the weld pool, the three-dimensional structure of the weld pool (i.e., the three-dimensional voxel spatial distribution) is obtained.

[0120] In an embodiment of the present invention, an infrared camera follows the movement of the laser welding head to capture the temperature distribution on the upper surface of the molten pool, and makes appropriate assumptions and combines the three-dimensional keyhole morphology to restore the keyhole to avoid temperature; an embedded thermocouple is used to supplement the real temperature information inside the molten pool during laser welding, and the temperature data of an appropriate number of grids inside the molten pool of the laser welding numerical model are randomly selected to jointly constitute the internal temperature data of the welding molten pool; some boundary temperature grid points are uniformly selected to form a temperature field reconstruction data set with the three-dimensional internal temperature data of the welding molten pool. The number of data points is selected according to the size of the temperature gradient. In an embodiment of the present invention, a total of 200 data points are selected, including 50 internal temperature data (12 from embedded thermocouples and 38 from numerical models) and 150 boundary temperature data (50 data points on the upper surface of the molten pool, 80 data points on the keyhole wall, and 20 data points on the solid-liquid boundary of the molten pool are uniformly selected).

[0121] The embodiment of the present invention adopts the radial basis function interpolation method to realize the three-dimensional reconstruction of the laser welding molten pool. First, a 200×200 coefficient matrix composed of known data points is constructed to solve the weight coefficient of the radial basis interpolation function. The radial basis interpolation function is then used to calculate the temperature distribution of the three-dimensional molten pool to realize the reconstruction of the three-dimensional temperature field. The three-dimensional structure of the welding molten pool can be represented by the voxel value in the grid. The voxel value of 1 indicates that it is inside the molten pool, and the voxel value of 0 indicates that it is outside the molten pool, thereby limiting the interpolation result to the inside of the molten pool. The relevant explanations of the formulas and variables involved in the above process have been explained in detail in step three. The remaining boundary data points are used to evaluate the model prediction error. The radial basis function is iteratively optimized by adjusting the basis function parameters to obtain a three-dimensional temperature field reconstruction model with high prediction accuracy. The schematic diagram of the three-dimensional temperature field reconstruction in this method is shown as follows. Figure 4 shown.

[0122] This embodiment of the present invention employs MPI parallel computing technology to construct a temperature field reconstruction dataset for each observation window using the data processing methods described in steps one and two. Three-dimensional reconstruction of the temperature field is then achieved using the method described in step three. The reconstructed temperature field is used as the initial condition for the laser welding temperature field evolution model. The prediction results for each time step are evaluated using historical temperature data from embedded thermocouples. Model prediction deviations are corrected by dynamically adjusting model parameters such as the laser absorption coefficient and thermal conductivity. Ultimately, reconstructed temperature field evolution data for each observation window is obtained.

[0123] Finally, the embodiment of the present invention extracts the prediction results of each parallel branch through MPIGather and outputs them as a dat file, merging the prediction results between each observation window together to achieve efficient calculation of the three-dimensional temperature field evolution behavior of the entire weld. Extract model results such as penetration depth and penetration width to achieve full weld formation prediction of laser welding. The flowchart of parallel solution in this method is as follows Figure 5 shown.

[0124] Figure 6 A comparison chart of prediction results from a specific example is provided. The results of a 3D temperature field reconstruction driven by multi-source data are shown in the figure. This was used as the initial condition in the numerical model to obtain the weld formation results for a specific section of laser welding. Comparison with actual welding results demonstrates that the proposed 3D temperature field reconstruction and weld formation prediction methods are effective and have demonstrated practical utility.

[0125] The internal forming condition after laser welding is unknown, and numerical simulation provides a non-destructive prediction method. The present invention improves the prediction accuracy of this non-destructive prediction method by integrating multi-source experimental data. A specific application scenario is provided: the forming prediction of the full weld of titanium alloy laser welding. Figure 1 The titanium alloy laser welding experimental device is set up as shown, and multi-source data of the titanium alloy molten pool during the welding process is obtained. The technical solution of the present invention is used to realize the multi-time 3D reconstruction of the welding molten pool. This is used as the initial condition of the numerical model to achieve a more accurate prediction of the full weld formation. The parallel part uses the 3D molten pool reconstruction results at multiple times to predict the weld formation at each time in parallel, realizing efficient prediction of the full weld formation. The specific prediction results are shown in the attached figure. Figure 6 As shown in the figure, by comparing with the actual welding results, it is found that the weld width and weld depth match well, proving that this technical solution has certain application value.

[0126] First, as attached Figure 4 、 5, 6, the results of the three-dimensional reconstruction of the laser welding molten pool of the present invention are shown in the figure. It can be seen from the reconstruction results that it is feasible to use multi-source data to realize the prediction of the full weld formation of laser welding, and the reconstruction results are consistent with many literatures and actual welding effects. Secondly, it is also feasible to bring the three-dimensional reconstruction results into the numerical model as initial conditions. The solution of the numerical model is essentially the numerical solution of a group of partial differential equations. The role of the initial conditions is to give the algorithm a starting point and ensure that the calculation results are limited to a reasonable range to ensure its convergence. In summary, the two important components of this technical solution: the three-dimensional reconstruction of the laser welding molten pool and the parallel prediction of the full weld formation results are both technically feasible, and both can achieve the technical effects described in the embodiments of the present invention.

[0127] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware portion can be implemented using dedicated logic; the software portion can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those skilled in the art will appreciate that the above-mentioned devices and methods can be implemented using computer-executable instructions and / or contained in processor control code, for example, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, or can be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0128] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.

Claims

1. A method for laser welding three-dimensional temperature field reconstruction and full weld formation prediction based on multi-source data, characterized in that: The following steps are involved: S1: Acquire multi-angle synchrotron radiation images and high-speed images of the upper surface during laser welding, and reconstruct the 3D structure of the weld pool based on image processing and voxel carving; S2: Combine infrared camera temperature measurement with embedded thermocouple measurement to construct the boundary temperature distribution and internal temperature data of the molten pool; S3: 3D modeling of the temperature reconstruction dataset is performed based on the radial basis function interpolation method, and the interpolation calculation is restricted to the voxel space inside the melt pool; S4: Parallel computing is used to reconstruct the temperature field at different times during the welding process in multiple observation windows, and a numerical model is introduced for prediction and correction; S5: Integrate the prediction results between each observation window to achieve continuous reconstruction of the three-dimensional temperature evolution behavior of the weld area, and then output the characteristic parameters of the entire weld formation.

2. The method according to claim 1, characterized in that The voxel carving step of reconstructing the three-dimensional structure in S1 includes converting the orthogonal projection image into three sets of binary matrices, and performing a direction-by-direction projection screening operation on the initial voxel space to eliminate voxels that do not meet the projection constraints to obtain the keyhole and molten pool contour structure.

3. The method according to claim 1, characterized in that The molten pool boundary temperature distribution in S2 includes the molten pool surface temperature and the keyhole Gaussian temperature model. The internal temperature is measured by multi-point thermocouples and combined with the numerical model grid point data to form a three-dimensional internal temperature data set.

4. The method according to claim 1, wherein The radial basis function interpolation formula in S3 is constructed based on the three-dimensional space distance function, and its shape parameters are determined by error feedback optimization iteration. Interpolation is only performed in the area where the voxel value is 1, ensuring that the temperature field is confined to the inside of the molten pool volume.

5. The method according to claim 1, wherein In the S4, MPI parallel computing is used to construct temperature reconstruction data sets in multiple cross-sectional observation windows, and the temperature reconstruction results are used as the initial conditions of the heat conduction model in each window, and a dynamic parameter adjustment mechanism is introduced to improve the model accuracy.

6. The method according to claim 1, wherein The reconstructed three-dimensional temperature field in S5 is merged and output as a structured data file through MPIGather. Subsequently, key parameters such as penetration depth and penetration width are extracted based on the continuous temperature field to achieve full weld formation prediction.

7. A laser welding three-dimensional temperature field reconstruction and full weld formation prediction system based on multi-source data drive, characterized by: include: Data acquisition module, used to collect multi-angle images, infrared temperature field and embedded temperature data during laser welding; 3D reconstruction module, used to generate the voxel space of the melt pool structure based on the voxel carving method and define the temperature interpolation area; Temperature modeling module, which constructs an interpolation dataset containing boundary and internal data and reconstructs the three-dimensional temperature field based on the radial basis function method; A prediction module that performs parallel simulation and parameter correction to predict the temperature evolution trend during welding; The forming output module splices and integrates all segmented data to output the temperature field and structural characteristics of the entire weld.

8. The system according to claim 7, characterized in that The data acquisition module includes a high-speed camera, an infrared camera and multiple sets of thermocouples, which are respectively used to obtain the surface image of the molten pool, the surface temperature distribution and the real temperature values ​​at different depths.

9. The system according to claim 7, wherein: The temperature modeling module is provided with a basis function adjustment submodule for adjusting the shape parameters of the interpolation function based on error feedback to improve prediction accuracy.

10. The system according to claim 7, wherein: The forming output module includes an MPIGather data merging mechanism and a structured data export interface, which is used to output the full weld forming prediction results and its three-dimensional temperature field evolution map.

Images