Thermal deformation intelligent compensation and control method and system in metal part welding process

Abstract

The invention provides a thermal deformation intelligent compensation and control method and system in a metal part welding process, and relates to the technical field of metal welding, and the thermal deformation intelligent compensation and control method comprises the following steps: obtaining a three-dimensional model and process parameters, constructing a thermal-mechanical coupling analysis model, and collecting real-time temperature and strain data; thermal deformation is predicted through a dynamic hierarchical graph structure and a bidirectional propagation neural network, optimal welding process parameters are calculated through a recursive optimizer integrating a gradient strategy and an evolution strategy based on thermal deformation deviation, accurate prediction and intelligent compensation of thermal deformation in the welding process can be achieved, and the welding quality and precision are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of metal welding, and particularly to an intelligent compensation and control method and system for thermal deformation during the welding process of metal parts. Background Art

[0002] Metal welding is an indispensable joining process in modern industrial manufacturing, and is widely used in fields such as aerospace, automotive manufacturing, and shipbuilding industry. During the welding process, due to uneven heating and cooling caused by high-temperature heat sources, metal parts often produce thermal deformation, which will seriously affect the geometric accuracy and assembly quality of products, and increase the workload and cost of subsequent trimming; Traditional control of thermal deformation in metal welding mainly relies on empirical rules and simplified physical models. In recent years, with the development of computer technology and finite element analysis methods, current thermal deformation prediction and control technologies mainly include simulation calculations based on finite element analysis, empirical process parameter adjustment, and real-time monitoring and feedback control methods; However, the existing technologies still have problems such as long simulation calculation time, difficulty in accurately handling the nonlinear heat conduction and thermal stress coupling processes of metal parts with complex structures, difficulty in effectively integrating multi-scale and multi-dimensional data features, insufficient response to dynamic changes during the welding process, resulting in low prediction accuracy, lack of intelligence and self-adaptability, often relying on manual experience for adjustment, difficult to achieve precise thermal deformation compensation, and unable to meet the requirements of high-precision manufacturing; Therefore, there is an urgent need for a solution to solve the problems existing in the existing technologies. Summary of the Invention

[0003] Embodiments of the present invention provide an intelligent compensation and control method and system for thermal deformation during the welding process of metal parts, which can at least solve some of the problems existing in the existing technologies.

[0004] In the first aspect of the embodiments of the present invention, an intelligent compensation and control method for thermal deformation during the welding process of metal parts is provided, including: Obtain the three-dimensional model data and welding process parameters of the metal part; Based on the three-dimensional model data, construct a thermal-mechanical coupling analysis model, divide the metal part into multiple hexahedral mesh units and establish a nonlinear heat conduction equation set, model the welding heat source as a double-ellipsoid heat source distribution function, construct a coupling transfer matrix of the temperature field and the stress field, and use an explicit integration algorithm to solve the nonlinear heat conduction equation set to obtain the transient temperature field distribution of the metal part; Collect the real-time temperature data and real-time strain data during the welding process of the metal part; Construct the hexahedral mesh elements into a dynamic hierarchical graph structure, calculate the thermal conduction relationship between mesh elements based on the transient temperature field distribution, the real-time temperature data, the real-time strain data, and the welding process parameters, fuse features of different scales using an adaptive attention mechanism, extract spatio-temporal features through a bidirectional propagation neural network, and generate the thermal deformation prediction result of the metal part; Compare the thermal deformation prediction result with a preset thermal deformation threshold to determine the thermal deformation deviation. Based on the thermal deformation deviation, take the welding process parameters as optimization variables, and calculate a compensation scheme using a recursive optimizer that combines the integrated gradient strategy and the evolutionary strategy. Iteratively solve to obtain the optimal welding process parameters and send them to the welding component for execution.

[0005] In an alternative embodiment, Construct a thermo-mechanical coupling analysis model based on the three-dimensional model data, divide the metal part into multiple hexahedral mesh elements, and establish a non-linear heat conduction equation set including: Obtain the three-dimensional model data of the metal part, identify and extract the key geometric features of the weld area and the heat affected zone after eliminating duplicate faces and dangling edges, establish a feature topology relationship graph, establish a stress analysis equation set based on the feature topology relationship graph, and perform an associated mapping between the stress analysis equation set and the temperature field distribution equation set to construct a thermo-mechanical coupling analysis model; Calculate the surface curvature distribution of the three-dimensional model data based on the thermo-mechanical coupling analysis model, determine the curvature value according to the surface curvature distribution, compare the curvature value with a preset curvature threshold to determine the mesh encryption area, determine the minimum mesh size and the maximum mesh size according to the mesh encryption area, establish a mesh size calculation function, perform mesh division on the metal part, and generate initial hexahedral mesh elements; Calculate the determinant value, mesh angle, and side length ratio of the initial hexahedral mesh elements, compare them with preset mesh quality parameters, adjust the node positions and reconstruct the boundaries of the hexahedral mesh elements that do not meet the preset mesh quality parameters, and perform segmentation processing on irregular hexahedral mesh elements to obtain optimized hexahedral mesh elements and establish a non-linear heat conduction equation set including temperature-dependent specific heat capacity coefficients and temperature-dependent thermal conductivity coefficients.

[0006] In an alternative embodiment, Model the welding heat source as a double-ellipsoid heat source distribution function, construct a coupling transfer matrix between the temperature field and the stress field, and use an explicit integration algorithm to solve the non-linear heat conduction equation set to obtain the transient temperature field distribution of the metal part including: Establish a front-ellipsoid heat source distribution function and a rear-ellipsoid heat source distribution function according to the welding voltage, welding current, thermal efficiency, ellipsoid major semi-axis, ellipsoid minor semi-axis, ellipsoid depth semi-axis, and heat distribution coefficient to obtain the initial heat source density distribution; Calculate the temperature distribution based on the initial heat source density distribution, obtain the temperature-dependent coefficient of thermal expansion and calculate the thermal strain, and superimpose the thermal strain with the elastic strain and plastic strain to obtain the total strain; Construct the heat conduction stiffness matrix, heat capacity matrix, heat load vector, structural stiffness matrix, structural damping matrix and mechanical load vector based on the total strain, combine to construct the coupled transfer matrix of the temperature field and stress field and the corresponding nonlinear heat conduction equations, and determine the explicit integration initial time step based on the element characteristic length, elastic wave speed and system characteristic frequency; At the explicit integration initial time step, calculate the velocity increment and displacement increment of the nonlinear heat conduction equations through the central difference scheme, determine the stress increment and substitute it into the nonlinear heat conduction equations for explicit integration calculation to obtain the initial value of the temperature field, calculate the temperature error based on the initial value of the temperature field, determine the adaptive time step and substitute it into the nonlinear heat conduction equations for explicit integration iterative calculation to obtain the transient temperature field distribution of the metal part.

[0007] In an alternative embodiment, At the explicit integration initial time step, calculating the velocity increment and displacement increment of the nonlinear heat conduction equations through the central difference scheme, determining the stress increment and substituting it into the nonlinear heat conduction equations for explicit integration calculation to obtain the initial value of the temperature field includes: At the explicit integration initial time step, add the product of the velocity increment at the previous moment and the time step and acceleration to calculate the velocity increment at the current moment, and add the product of the displacement increment at the previous moment and the time step and the velocity increment at the current moment to calculate the displacement increment at the current moment; Multiply the coefficient of thermal expansion by the temperature increment and the unit tensor to obtain the thermal strain increment, determine whether plastic deformation occurs according to the current stress state and calculate the plastic strain increment, subtract the thermal strain increment and plastic strain increment from the total strain increment to obtain the elastic strain increment, and calculate the stress increment through the elastic modulus and elastic strain increment; Subtract the partial derivative of the internal energy with respect to time from the divergence of the heat flux density and the heat source term and take the absolute value to calculate the energy balance error and correction coefficient, multiply the exponential function of the absolute value of the temperature gradient by the linear function of the absolute value of the gradient of the stress increment to calculate the adaptive weight function, multiply the correction coefficient by the energy balance error and the temperature gradient direction and take the negative value to calculate the temperature correction amount, and calculate the corrected temperature field based on the adaptive weight function and the temperature correction amount; Substitute the stress increment into the non - linear heat conduction equations for explicit integral calculation to obtain the second temperature field distribution. Compare the norm ratio of the difference between the corrected temperature field and the second temperature field distribution and the norm of the energy balance error with the temperature field convergence tolerance and the energy balance convergence tolerance respectively. Based on the comparison results, perform iterative calculation to obtain the initial value of the temperature field.

[0008] In an alternative embodiment, Construct the hexahedral mesh elements into a dynamic hierarchical graph structure. Calculate the thermal - mechanical conduction relationship between mesh elements based on the transient temperature field distribution, the real - time temperature data, the real - time strain data, and the welding process parameters. Use an adaptive attention mechanism to fuse features at different scales, extract spatio - temporal features through a bidirectional propagation neural network, and generate the thermal deformation prediction result of the metal part, including: Construct an octree space partition structure based on the central point coordinates of the hexahedral mesh elements. Calculate the topological adjacency relationship between mesh elements, assign spatial coordinates and level indices to each mesh node, generate a dynamic hierarchical graph structure, extract the geometric, topological, and physical features of each mesh node to form a node feature vector, and calculate the product of the Euclidean distance square exponential function of the spatial position of the mesh nodes and the level difference exponential function of the mesh nodes to obtain the connection weight between mesh nodes; Input the transient temperature field distribution, real - time temperature data, and real - time strain data into the conduction calculation model. Calculate the temperature difference and strain difference between adjacent mesh nodes and perform exponential mapping respectively to obtain the temperature influence factor and the strain influence factor. Combine the material thermal conductivity to determine the thermal conduction coefficient between adjacent mesh nodes, and establish the thermal - mechanical conduction relationship between mesh elements based on the thermal conduction coefficient and the current welding process parameters; Input features at different scales into the adaptive attention mechanism. Calculate the scale attention weight through the dot - product operation between the mesh scale hierarchical features and the global feature matrix, perform weighted summation on the feature - transformed mesh scale hierarchical features to obtain the fused features, and input the fused features into the bidirectional propagation neural network. Use the gated recurrent unit for forward propagation to obtain the forward hidden state, perform backward propagation to obtain the backward hidden state, and splice the forward hidden state and the backward hidden state to extract spatio - temporal features; Input the spatio - temporal features and the constraint conditions into the prediction function. Combine the topological thermal - mechanical field self - correction mechanism to calculate the deformation field, estimate the prediction uncertainty based on the deformation field and the spatio - temporal features, and generate the thermal deformation prediction result of the metal part.

[0009] In an alternative embodiment, Input the spatio - temporal features and the constraint conditions into the prediction function. Combine the topological thermal - mechanical field self - correction mechanism to calculate the deformation field, estimate the prediction uncertainty based on the deformation field and the spatio - temporal features, and generate the thermal deformation prediction result of the metal part, including: Input the spatio-temporal characteristics of the metal part and pre-set constraint conditions into a prediction function, where the constraint conditions include a thermo-mechanical constitutive equation that establishes the correlation between the stress tensor and the strain tensor through an elastic constant tensor and a coefficient of thermal expansion; Construct a topological thermo-mechanical field self-correction mechanism based on persistent homology theory. Calculate the Betti number sequence of the initial deformation field output by the prediction function to form a deformation field topological skeleton. Calculate the persistence diagram of the initial deformation field according to the Betti number sequence, and calibrate the key deformation region of the metal part according to the mutation points of the persistence diagram; For the key deformation region, construct a distance metric function to calculate the characteristic evolution trajectory of the deformation field topological skeleton. Project the initial deformation field onto a manifold space that satisfies the constraint conditions to obtain a projected deformation field. Iteratively optimize the initial deformation field by minimizing the distance between the initial deformation field and the projected deformation field and the evolution trajectory distance of the deformation field topological skeleton to obtain a corrected deformation field; Calculate the persistent homology entropy according to the corrected deformation field. Weight the persistent homology entropy and the corrected deformation field to obtain an uncertainty evaluation value. Input the corrected deformation field, the uncertainty evaluation value, and the spatio-temporal characteristics into a preset basis function, and generate a thermal deformation prediction result of the metal part by calculating the weighted sum of the basis function.

[0010] In an alternative embodiment, Compare the thermal deformation prediction result with a preset thermal deformation threshold to determine the thermal deformation deviation. Based on the thermal deformation deviation, take the welding process parameters as optimization variables, and combine a recursive optimizer of an integrated gradient strategy and an evolutionary strategy to calculate a compensation scheme. Iteratively solve to obtain the optimal welding process parameters and send them to the welding assembly for execution, including: Compare the thermal deformation prediction result with a preset thermal deformation threshold to obtain the thermal deformation deviation, and perform a time integral on the thermal deformation deviation to calculate the accumulated thermal deformation deviation; Take the welding process parameters as optimization variables, calculate the gradient of the thermal deformation deviation with respect to the welding process parameters and construct a gradient strategy, combine Gaussian noise to construct an evolutionary strategy, and integrate the gradient strategy and the evolutionary strategy to construct a recursive optimizer; Establish an optimization objective function containing the thermal deformation deviation based on the recursive optimizer. Perform constraint processing on the value range and change rate of the welding process parameters based on the gradient of the thermal deformation deviation. Take the welding process parameters that meet the convergence conditions as the optimal welding process parameters and send them to the welding assembly to perform welding operations.

[0011] In the second aspect of the embodiments of the present invention, a thermal deformation intelligent compensation and control system during the welding process of a metal part is provided, including: A first unit for obtaining three-dimensional model data of a metal part and welding process parameters; A second unit for constructing a thermal-mechanical coupling analysis model based on the three-dimensional model data, dividing the metal part into a plurality of hexahedral mesh elements and establishing a non-linear heat conduction equation set, modeling the welding heat source as a double-ellipsoid heat source distribution function, constructing a coupling transfer matrix of the temperature field and the stress field, and using an explicit integration algorithm to solve the non-linear heat conduction equation set to obtain the transient temperature field distribution of the metal part; A third unit for collecting real-time temperature data and real-time strain data during the welding process of the metal part; A fourth unit for constructing the hexahedral mesh elements into a dynamic hierarchical graph structure, calculating the thermal-mechanical conduction relationship between the mesh elements based on the transient temperature field distribution, the real-time temperature data, the real-time strain data, and the welding process parameters, fusing features of different scales using an adaptive attention mechanism, extracting spatio-temporal features through a bidirectional propagation neural network, and generating a thermal deformation prediction result of the metal part; A fifth unit for comparing the thermal deformation prediction result with a preset thermal deformation threshold to determine a thermal deformation deviation, based on the thermal deformation deviation, taking the welding process parameters as optimization variables, calculating a compensation scheme using a recursive optimizer combining an integrated gradient strategy and an evolutionary strategy, iteratively solving to obtain optimal welding process parameters and sending them to a welding component for execution.

[0012] In a third aspect of an embodiment of the present invention, there is provided an electronic device, including: A processor and a memory for storing processor-executable instructions, wherein the processor is configured to call the instructions stored in the memory to execute the method described above.

[0013] In a fourth aspect of an embodiment of the present invention, there is provided a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0014] In the present invention, by constructing a thermo-mechanical coupling analysis model and combining the temperature and strain data collected in real time, the thermal deformation behavior of metal parts during the welding process can be accurately predicted, realizing the precise prediction and control of welding deformation, effectively reducing the welding deformation error, improving the welding precision, constructing the hexahedral mesh elements into a dynamic hierarchical graph structure, and using the adaptive attention mechanism to fuse different-scale features, extracting spatio-temporal features through a bidirectional propagation neural network, improving the accuracy and real-time performance of thermal deformation prediction, enabling the system to adapt to complex and variable welding working conditions, calculating the compensation scheme using a recursive optimizer integrating the integrated gradient strategy and the evolutionary strategy, realizing the intelligent optimization and adjustment of welding process parameters, being able to automatically generate the optimal welding parameters according to the thermal deformation prediction results, greatly improving the welding automation level and production efficiency, and reducing manual intervention and material waste. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is a schematic flow chart of the method for intelligent compensation and control of thermal deformation during the welding process of metal parts in an embodiment of the present invention; Figure 2 is a simulation diagram of the development of the plastic zone of the method for intelligent compensation and control of thermal deformation during the welding process of metal parts in an embodiment of the present invention; Figure 3 is a comparison diagram of the topological skeleton evolution trajectories of the method for intelligent compensation and control of thermal deformation during the welding process of metal parts in an embodiment of the present invention; Figure 4 is a flow chart for solving the optimal welding parameters of the method for intelligent compensation and control of thermal deformation during the welding process of metal parts in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, rather than all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0017] The technical solutions of the present invention will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0018] Figure 1 is a schematic flow chart of the method for intelligent compensation and control of thermal deformation during the welding process of metal parts in an embodiment of the present invention, as Figure 1 shown, the method includes: Obtain the three-dimensional model data of the metal part and the welding process parameters; Construct a thermo-mechanical coupling analysis model based on the 3D model data, divide the metal part into multiple hexahedral mesh elements and establish a non-linear heat conduction equation set, model the welding heat source as a double ellipsoidal heat source distribution function, construct a coupling transfer matrix between the temperature field and the stress field, and use an explicit integration algorithm to solve the non-linear heat conduction equation set to obtain the transient temperature field distribution of the metal part; Collect the real-time temperature data and real-time strain data during the welding process of the metal part; Construct the hexahedral mesh elements into a dynamic hierarchical graph structure, calculate the thermo-mechanical conduction relationship between the mesh elements based on the transient temperature field distribution, the real-time temperature data, the real-time strain data and the welding process parameters, use an adaptive attention mechanism to fuse features at different scales, extract spatio-temporal features through a bidirectional propagation neural network, and generate the thermal deformation prediction result of the metal part; Compare the thermal deformation prediction result with a preset thermal deformation threshold to determine the thermal deformation deviation. Based on the thermal deformation deviation, take the welding process parameters as optimization variables, and calculate a compensation scheme using a recursive optimizer that combines the integrated gradient strategy and the evolutionary strategy. Iteratively solve to obtain the optimal welding process parameters and send them to the welding assembly for execution.

[0019] In an alternative implementation, Constructing a thermo-mechanical coupling analysis model based on the 3D model data and dividing the metal part into multiple hexahedral mesh elements and establishing a non-linear heat conduction equation set includes: Obtain the 3D model data of the metal part, identify and extract the key geometric features of the weld area and the heat affected zone after eliminating duplicate faces and dangling edges, establish a feature topology relationship graph, establish a stress analysis equation set based on the feature topology relationship graph, and perform an associative mapping between the stress analysis equation set and the temperature field distribution equation set to construct a thermo-mechanical coupling analysis model; Calculate the surface curvature distribution of the 3D model data based on the thermo-mechanical coupling analysis model, determine the curvature value according to the surface curvature distribution, compare the curvature value with a preset curvature threshold to determine the mesh encryption area, determine the minimum mesh size and the maximum mesh size according to the mesh encryption area, establish a mesh size calculation function, perform mesh division on the metal part, and generate initial hexahedral mesh elements; Calculate the determinant value, mesh angle and side length ratio of the initial hexahedral mesh elements, compare them with preset mesh quality parameters, adjust the node positions and reconstruct the boundaries of the hexahedral mesh elements that do not meet the preset mesh quality parameters, and perform segmentation processing on irregular hexahedral mesh elements to obtain optimized hexahedral mesh elements and establish a non-linear heat conduction equation set including temperature-dependent specific heat capacity coefficients and temperature-dependent thermal conductivity coefficients.

[0020] Import the 3D model data of the metal part. Using the spatial topology search algorithm, traverse all the face elements in the model in sequence. By comparing the normal vectors and boundary vertex coordinates of the face elements, identify the duplicate face elements. For the identified duplicate face elements, retain one of them and delete the other duplicates. Use the connectivity analysis method to detect the boundary elements. If an edge is only connected to one face element, it is determined as a hanging edge and deleted.

[0021] Adopt the region growing algorithm to identify the weld region. Starting from the weld center line, expand outward based on the curvature continuity and geometric feature similarity criteria until reaching the weld boundary. On the basis of determining the weld region, use the temperature gradient prediction method to determine the range of the heat affected zone, and mark the region with drastic temperature change as the heat affected zone. Extract the key geometric features of the weld and the heat affected zone, including feature points, feature lines and feature surfaces.

[0022] When constructing the feature topology relationship graph, use the graph theory method to map the geometric features to the nodes of the graph, and map the spatial association relationship between features to the edges of the graph. Traverse all features through the depth first search algorithm to establish a complete topological connection relationship. Based on the topological connection relationship graph, use the stress transfer path analysis method to identify the main stress paths and stress concentration regions, and establish a balance equation set describing the stress distribution.

[0023] At the same time, establish the temperature field distribution equation set. Adopt the thermal-mechanical coupling algorithm to associate and map the stress field and the temperature field through material properties. Specifically, through the iterative solution method, alternately update the temperature field and the stress field in each iterative step until the two fields reach the equilibrium state, so as to obtain a complete thermo-mechanical coupling analysis model.

[0024] Use the differential geometry algorithm to calculate the curvature distribution on the model surface. First, construct a local coordinate system of the surface and calculate the principal curvature and Gaussian curvature. Adopt the adaptive threshold segmentation method to compare the curvature values with the dynamically determined threshold to identify the high curvature regions that need mesh refinement. Combine the overall size of the model and determine the minimum and maximum mesh sizes through the proportional relationship.

[0025] Establish a mesh size calculation function considering the curvature distribution. This function uses the distance weighted interpolation algorithm to smoothly transition to a smaller mesh size in the high curvature region and gradually transition to a larger mesh size in the low curvature region. Use the octree meshing algorithm for the initial mesh division, and determine whether to continue subdividing according to the mesh size function in each meshing step.

[0026] The quality of the initial mesh is evaluated and the quality parameters of each mesh unit are calculated. The mesh smoothing algorithm is used to optimize the mesh units that do not meet the quality requirements, and the mesh shape is improved through iterative optimization of the node positions. For severely deformed mesh units, the mesh reconstruction algorithm is used to regenerate the local mesh. For irregular hexahedral meshes, the template matching method is used to split them into multiple regular mesh units.

[0027] On the optimized grid, the finite element interpolation method is used to construct the temperature-dependent material property distribution function, and the specific heat capacity and thermal conductivity are expressed as functions of temperature. Based on the temperature-dependent material properties, the nonlinear heat conduction equations are constructed, and the matrix assembly algorithm is used to generate the overall stiffness matrix and load vector.

[0028] For example, a T-shaped welded joint is formed by vertically welding two plates. After importing the 3D model of the T-shaped joint, the duplicate faces at the overlap of the two plates and the dangling edges at the edges are deleted. The geometric features of the vertical weld area are determined by feature recognition, and the heat-affected zone is determined by expanding outward from the weld. The spatial relationship between the weld, heat-affected zone, plate and other features is organized into a topological relationship diagram.

[0029] Based on the topological relationship diagram, the stress transfer path is analyzed, and the stress analysis equation group is established. At the same time, the temperature field distribution equation group is established. The two equation groups are associated through the thermal expansion coefficient to build a thermal-mechanical coupling analysis model. The curvature distribution of the T-joint surface is calculated, and it is found that the curvature values ​​at the weld and the heat-affected zone are large, and these areas are marked as mesh encryption areas.

[0030] The minimum grid size is set to 2 mm, the maximum grid size is set to 10 mm, and a grid size calculation function is established. The T-joint is initially meshed, with denser grids used at the weld and heat-affected zone, and sparser grids used in areas far from the weld. After evaluating the mesh quality, it was found that there were severely deformed grid units at the weld corners, and the mesh quality was improved by optimizing the node position and reconstructing the mesh. A group of heat conduction equations was established on the optimized mesh to complete the meshing and equation construction work.

[0031] In this embodiment, the key geometric features of the weld area and the heat-affected zone can be accurately identified through spatial topological analysis and feature extraction methods, and duplicate faces and dangling edges in the model can be eliminated, thereby improving the accuracy of subsequent analysis. Curvature analysis and adaptive meshing strategies are used to automatically encrypt the mesh in areas with large stress and temperature gradients, and appropriately enlarge the mesh size in areas with gentle changes, thereby ensuring both calculation accuracy and improving calculation efficiency. Through mesh quality evaluation and optimization algorithms, the mesh quality is significantly improved, the numerical calculation error caused by mesh distortion is reduced, and the accuracy of thermal deformation analysis results is improved.

[0032] In an optional embodiment, Model the welding heat source as a double-ellipsoid heat source distribution function, construct the coupled transfer matrix of the temperature field and the stress field, and use the explicit integration algorithm to solve the non-linear heat conduction equations to obtain the transient temperature field distribution of the metal part, including: Establish the front-ellipsoid heat source distribution function and the rear-ellipsoid heat source distribution function according to the welding voltage, welding current, thermal efficiency, major semi-axis of the ellipsoid, minor semi-axis of the ellipsoid, depth semi-axis of the ellipsoid, and heat distribution coefficient, and obtain the initial heat source density distribution; Calculate the temperature distribution based on the initial heat source density distribution, obtain the temperature-dependent coefficient of thermal expansion and calculate the thermal strain, and superimpose the thermal strain with the elastic strain and plastic strain to obtain the total strain; Construct the heat conduction stiffness matrix, heat capacity matrix, heat load vector, structural stiffness matrix, structural damping matrix, and mechanical load vector based on the total strain, combine to construct the coupled transfer matrix of the temperature field and the stress field and the corresponding non-linear heat conduction equations, and determine the explicit integration initial time step based on the element characteristic length, elastic wave speed, and system characteristic frequency; At the explicit integration initial time step, calculate the velocity increment and displacement increment of the non-linear heat conduction equations through the central difference scheme, determine the stress increment and substitute it into the non-linear heat conduction equations for explicit integration calculation to obtain the initial value of the temperature field, calculate the temperature error based on the initial value of the temperature field, determine the adaptive time step and substitute it into the non-linear heat conduction equations for explicit integration iterative calculation to obtain the transient temperature field distribution of the metal part.

[0033] Construct a double-ellipsoid heat source distribution model through welding process parameters, and the front and rear ellipsoid heat sources are described by different spatial distribution functions. Determine the total heat input power according to the welding voltage and current, and calculate the effective heat input in combination with the thermal efficiency. Define the heat source spatial distribution shape based on the major semi-axis, minor semi-axis, and depth semi-axis of the ellipsoid, and distribute the total heat to the front and rear ellipsoid regions through the heat distribution coefficient, and establish the front-ellipsoid heat source distribution function and the rear-ellipsoid heat source distribution function respectively. The two distribution functions are superimposed to obtain the complete initial heat source density distribution.

[0034] Adopt the transient heat conduction analysis method, use the initial heat source density distribution as the input, and calculate the temperature distribution during welding. Obtain the coefficient of thermal expansion at different temperatures from the material property database, and multiply the temperature field change amount by the coefficient of thermal expansion to obtain the thermal strain. At the same time, calculate the elastic strain generated by the material under the action of temperature and load, and calculate the plastic strain when the stress exceeds the yield strength. Superimpose the three strains to obtain the total strain.

[0035] Based on the finite element discretization method, a thermo-mechanical coupling matrix is constructed using the total strain. A heat conduction stiffness matrix and a heat capacity matrix describing the heat conduction characteristics are established, and a heat load vector reflecting the heat source input is constructed. A structural stiffness matrix, a structural damping matrix describing the structural deformation characteristics, and a mechanical load vector reflecting the mechanical constraints are established. The matrices are combined into a coupling transfer matrix of the temperature field and the stress field through a coupling algorithm, and a corresponding non-linear heat conduction equation set is established.

[0036] The element characteristic length is calculated using the mesh size analysis method. The elastic wave speed is determined by combining the material elastic modulus and density. The system characteristic frequency is obtained through modal analysis. Based on these parameters, the explicit integration initial time step is determined, and this time step needs to meet the calculation stability requirements. Within the determined time step, the non-linear heat conduction equation set is discretized using the central difference scheme, and the velocity increment and displacement increment at each time step are calculated.

[0037] The stress increment is calculated based on the velocity increment and displacement increment, and the stress increment is substituted into the non-linear heat conduction equation set for explicit integration calculation to obtain the initial value of the temperature field. The temperature error is calculated by comparing with the measured temperature data, and the adaptive time step is determined based on the temperature error and the temperature gradient. An iterative calculation method is used to solve the non-linear heat conduction equation set within each adaptive time step until the temperature field and the displacement field converge, and the transient temperature field distribution of the metal part during the entire welding process is obtained.

[0038] Exemplarily, taking the butt welding of a stainless steel plate as an example, the welding voltage is 25 volts, the current is 200 amperes, and the heat efficiency is 0.8. According to the welding process and material properties, the major axis of the front ellipsoidal heat source is set to 8 mm, the minor axis is 6 mm, and the depth axis is 5 mm. The corresponding parameters of the rear ellipsoid are 12 mm, 8 mm, and 7 mm respectively, and the heat distribution coefficient is 0.4 for the front ellipsoid and 0.6 for the rear ellipsoid.

[0039] The heat source density distribution in the weld area is calculated through the double-ellipsoid heat source model. The heat source density in the front ellipsoid area is relatively large and concentrated, while the heat source density in the rear ellipsoid area is relatively small but has a larger distribution range. Based on this heat source distribution, the temperature field is calculated, and the temperature at the weld center can reach up to 1800 degrees Celsius, and the temperature drops rapidly as the distance increases.

[0040] Combined with the temperature-related material properties of stainless steel, the maximum thermal strain in the weld area is calculated to be approximately 0.02, and the thermal strain in the heat-affected zone is between 0.005 and 0.015. After superimposing the elastic strain and plastic strain, the total strain in the weld area can reach 0.035. Based on these strain values, a coupling matrix is constructed. The characteristic length of the grid near the weld is taken as 0.5 mm, and the calculated stable time step is 0.0001 seconds.

[0041] The initial temperature field distribution is calculated by explicit integration, and the maximum error compared with the measured temperature is within 5%. The adaptive time step algorithm is adopted. In the region where the temperature changes violently, the time step is automatically reduced to 0.00005 seconds, and in the region where the temperature changes gently, the time step is increased to 0.0002 seconds. After about 2000 steps of iterative calculation, the complete transient temperature field distribution nephogram is obtained.

[0042] In this embodiment, by constructing an accurate double-ellipsoid heat source model and a thermal-mechanical coupling analysis method, high-precision prediction of the temperature field and stress field during the welding process is realized. By establishing a full-field strain analysis model including thermal strain, elastic strain, and plastic strain, the nonlinear mechanical behavior of the material under high-temperature conditions is accurately reflected. The adopted adaptive time step strategy dynamically adjusts the calculation step according to the severity of the temperature gradient and stress changes, significantly improving the calculation efficiency while ensuring the calculation accuracy, providing a reliable theoretical basis for welding process optimization and quality control.

[0043] In an alternative embodiment, At the explicit integration initial time step, the velocity increment and displacement increment of the nonlinear heat conduction equations are calculated by the central difference scheme, the stress increment is determined and substituted into the nonlinear heat conduction equations for explicit integration calculation, and the initial temperature field values obtained include: At the explicit integration initial time step, the current velocity increment is calculated by adding the product of the previous velocity increment and the time step and the acceleration, and the current displacement increment is calculated by adding the product of the previous displacement increment and the time step and the current velocity increment; The thermal strain increment is obtained by multiplying the coefficient of thermal expansion, the temperature increment, and the unit tensor. Whether plastic deformation occurs is determined according to the current stress state and the plastic strain increment is calculated. The elastic strain increment is obtained by subtracting the thermal strain increment and the plastic strain increment from the total strain increment. The stress increment is calculated by the elastic modulus and the elastic strain increment; The absolute value of the difference between the partial derivative of the internal energy with respect to time and the divergence of the heat flux density and the heat source term is calculated to obtain the energy balance error and the correction coefficient. The exponential function of the absolute value of the temperature gradient and the linear function of the absolute value of the gradient of the stress increment are multiplied to obtain the adaptive weight function. The negative value of the product of the correction coefficient, the energy balance error, and the temperature gradient direction is calculated to obtain the temperature correction amount. The corrected temperature field is calculated based on the adaptive weight function and the temperature correction amount; The stress increment is substituted into the nonlinear heat conduction equations for explicit integration calculation to obtain the second temperature field distribution. The norm ratio of the difference between the corrected temperature field and the second temperature field distribution and the norm of the energy balance error are respectively compared with the temperature field convergence tolerance and the energy balance convergence tolerance, and iterative calculation is performed based on the comparison results to obtain the initial temperature field values.

[0044] In the initial time step of explicit integration, a recursive algorithm is used to perform dynamic response analysis. The acceleration influence matrix is ​​constructed, the velocity state at the previous moment is combined with the acceleration influence, and the velocity increment at the current moment is calculated through the time domain integrator. A similar recursive strategy is used to establish a displacement prediction model, the displacement state at the previous moment is coupled with the velocity influence, and the displacement increment at the current moment is obtained through time step mapping.

[0045] The strain analysis adopts a multi-field coupling decomposition strategy. A thermal strain calculation model is established to perform tensor mapping of the thermal expansion characteristics of the material and the temperature field changes to obtain the thermal strain increment field in three-dimensional space. A plasticity criterion evaluator is constructed to judge the stress state of each calculation point through the yield surface evolution algorithm. When the stress exceeds the yield surface, the plastic flow calculator is started and the iterative projection method is used to solve the plastic strain increment. The strain decomposition algorithm performs differential operations on the total strain increment, the thermal strain increment and the plastic strain increment to obtain the elastic strain increment, and then calculates the stress increment through the material constitutive model.

[0046] The energy balance correction adopts an adaptive multi-scale algorithm. An internal energy derivative calculator is constructed, and the internal energy change rate is calculated using the time discretization method. The difference operation is performed with the heat flux density divergence operator and the heat source term to obtain the energy balance error. An adaptive correction function is constructed based on the error distribution. The function adopts the polynomial basis function expansion method, and the coefficients are determined by least squares optimization. At the same time, a gradient weight analyzer is established, and the temperature gradient uses an exponential mapping function for feature enhancement, and the stress gradient uses a piecewise linear function for feature extraction.

[0047] The temperature field correction process adopts an iterative optimization strategy. The energy balance correction function and the temperature gradient direction field are tensor-producted to construct a temperature correction operator. The adaptive weight function is used to modulate the spatial distribution of the correction value. The weight function is automatically adjusted through local feature analysis, and the weight is increased in areas with drastic gradient changes to ensure the effectiveness of the correction. The corrected temperature field is smoothed by the relaxation iteration method to ensure the continuity of the temperature distribution.

[0048] The convergence analysis adopts a multi-criteria judgment method. The stress increment is substituted into the nonlinear equation group to construct a residual calculator, and the second temperature field is obtained by an explicit integration algorithm. A field difference evaluator is established, and the difference between the corrected temperature field and the second temperature field is compared by the norm calculation method. At the same time, an energy balance evaluator is constructed to calculate the norm of the energy balance error. A dual threshold judgment criterion is used to compare with the temperature field convergence tolerance and the energy balance convergence tolerance respectively. If any indicator exceeds the tolerance range, the adaptive time step adjuster is started, and the calculation parameters are updated and the iteration loop is re-entered until all convergence indicators meet the requirements, and the initial value of the temperature field is finally determined.

[0049] Exemplarily, taking the stainless steel plate welding process as an example, the initial time step is set to 0.0001 seconds. Within this time step, the velocity increment at a certain calculation point at the previous moment is 0.002 millimeters per second. Considering the influence of acceleration, the velocity increment at the current moment is calculated to be 0.0025 millimeters per second. Combining the displacement increment of 0.0015 millimeters at the previous moment, the displacement increment at the current moment is calculated to be 0.002 millimeters.

[0050] The temperature at this calculation point rises from 800 degrees Celsius to 820 degrees Celsius. The corresponding coefficient of thermal expansion is 0.000012 per degree Celsius. The thermal strain increment is calculated to be 0.00024. Checking the stress state, it is found that the yield strength of 300 MPa has been exceeded, and the plastic strain increment is calculated to be 0.00015. Subtracting the above two components from the total strain increment of 0.0005, the elastic strain increment of 0.00011 is obtained, and the stress increment is calculated to be 22 MPa.

[0051] The energy balance analysis shows that the energy balance error at this point is 0.5 joules per cubic millimeter. Based on this, the correction coefficient is determined to be 0.8. The temperature gradient at this point is 50 degrees Celsius per millimeter, and the stress gradient is 15 MPa per millimeter. The value of the adaptive weight function is calculated to be 1.2. Combining these parameters, the temperature correction amount is calculated to be 2 degrees Celsius, and the temperature field is corrected.

[0052] Substituting the calculation results into the equations for explicit integration, the second temperature field distribution is obtained. By comparison, the ratio of the difference norm between the corrected temperature field and the second temperature field is 0.002, and the energy balance error norm is 0.003. Both of these values are less than the preset convergence tolerance of 0.005. Therefore, the initial value of the temperature field at this point is determined to be 818 degrees Celsius.

[0053] In this embodiment, by constructing a recursive dynamic response analysis algorithm, high-precision calculations of velocity increment and displacement increment are achieved. By adopting a multi-field coupling decomposition strategy, the total strain is decomposed into thermal strain, plastic strain, and elastic strain, accurately capturing the nonlinear mechanical behavior of the material under high-temperature conditions. Through an adaptive multi-scale algorithm for energy balance correction, a weight function based on local characteristics is established, achieving precise correction of the temperature field; In the prior art, the calculation of the welding temperature field usually adopts a simple explicit integration method, lacking an adaptive adjustment mechanism in the selection of time steps, with low calculation efficiency. The temperature field and stress field are treated separately, ignoring the multi-field coupling effect, resulting in insufficient calculation accuracy. The prior art mostly adopts a globally unified correction coefficient, which is difficult to accurately reflect the temperature and stress change characteristics of local regions, affecting the accuracy of calculation results; In this embodiment, through the adaptive time step control and multi-criterion convergence judgment method, the number of iterations is significantly reduced, and the calculation speed is improved. The multi-field coupling analysis and adaptive correction strategy make the calculation results of the temperature field and stress field closer to the actual physical process. By establishing a complete error control mechanism, the convergence and reliability of the numerical solution are ensured, providing reliable theoretical support for welding process optimization and quality control.

[0054] Figure 2 This is the simulation diagram of the plastic zone development in the intelligent compensation and control method for thermal deformation during the welding process of metal parts in the embodiment of the present invention, showing the evolution process of the plastic zone during the welding of a stainless steel plate over time. The figure shows a cross-sectional view of the weld, with concentric circles representing the boundaries of the plastic zone at three time points: the initial stage (0.5 s), the middle stage (2.5 s), and the later stage (4.5 s) of welding. Different colors represent different prediction methods or experimental measurement values. The radius of the plastic zone predicted by this technical solution (red line) at the initial stage of welding is 3.25 mm, with an error of only 2.11% compared to the experimental measurement value (purple line) of 3.32 mm; at the middle stage of welding, the predicted value is 5.87 mm, with an error of 2.49% compared to the experimental value of 6.02 mm; at the later stage of welding, the predicted value is 7.43 mm, with an error of 1.98% compared to the experimental value of 7.58 mm. The predicted plastic zone radii of the traditional finite element method (blue line) at these three time points are 2.95 mm, 5.36 mm, and 6.84 mm respectively, with errors compared to the experimental values reaching 11.14%, 10.96%, and 9.76% respectively, which are significantly larger than the experimental observation results. The predicted values of the Johnson-Cook model (green line) are 3.08 mm, 5.54 mm, and 7.12 mm respectively, with errors of 7.23%, 7.97%, and 6.07% respectively. Although the accuracy is higher than that of the traditional finite element method, it is still lower than this technical solution.

[0055] It can be clearly observed from the figure that as the welding time extends, the radius of the plastic zone continuously expands, and the expansion rate gradually decreases, reflecting the law of plastic deformation of the material at high temperatures. This technical solution successfully captures the non-linear plastic behavior of the material at high temperatures by accurately calculating the coupling effect of the thermal strain increment and the plastic strain increment. During the entire welding process (0.5 - 5.0 s), the average prediction error is only 2.26%, reducing the errors by 78.8% and 68.9% compared to the traditional finite element method and the Johnson-Cook model respectively, achieving a high-precision simulation of the plastic zone expansion process.

[0056] In an alternative embodiment, Construct the hexahedral mesh elements into a dynamic hierarchical graph structure, calculate the thermal conduction relationship between mesh elements based on the transient temperature field distribution, the real-time temperature data, the real-time strain data, and the welding process parameters, fuse features of different scales using an adaptive attention mechanism, extract spatio-temporal features through a bidirectional propagation neural network, and generate the thermal deformation prediction result of the metal part, including: Construct an octree space partitioning structure based on the central point coordinates of the hexahedral mesh elements, calculate the topological adjacency relationship between mesh elements, assign spatial coordinates and level indices to each mesh node, generate a dynamic hierarchical graph structure, extract the geometric, topological, and physical features of each mesh node to form a node feature vector, and calculate the product of the Euclidean distance squared exponential function of the mesh node spatial positions and the level difference exponential function of the mesh nodes to obtain the connection weights between mesh nodes; Input the transient temperature field distribution, real-time temperature data, and real-time strain data into the conduction calculation model, calculate the temperature difference and strain difference between adjacent mesh nodes and perform exponential mapping respectively to obtain the temperature influence factor and the strain influence factor, determine the thermal conductivity between adjacent mesh nodes in combination with the material thermal conductivity, and establish the thermal conduction relationship between mesh elements based on the thermal conductivity and the current welding process parameters; Input features of different scales into the adaptive attention mechanism, calculate the scale attention weights through the dot product operation between the mesh scale level features and the global feature matrix, perform weighted summation on the feature-transformed mesh scale level features to obtain the fused features and input them into the bidirectional propagation neural network, use the gated recurrent unit for forward propagation to obtain the forward hidden state, perform backward propagation to obtain the backward hidden state, concatenate the forward hidden state and the backward hidden state, and extract spatio-temporal features; Input the spatio-temporal features and the constraint conditions into the prediction function, calculate the deformation field in combination with the topological thermal field self-correction mechanism, estimate the prediction uncertainty based on the deformation field and the spatio-temporal features, and generate the thermal deformation prediction result of the metal part.

[0057] Construct the grid space structure. Starting from the central point of the hexahedral mesh element, use a recursive partitioning strategy to establish an octree space partitioning structure. By traversing adjacent elements, establish a topological connection relationship table between grid elements and record the adjacent element information of each element. Assign three-dimensional spatial coordinates to each node in the grid structure, and at the same time establish a level index system to construct a dynamic hierarchical graph structure with parent-child relationships. Extract the geometric, topological, and physical features of each grid node, and combine these features to form a node feature vector. Use a spatial distance calculation method to multiply the square exponential function of the Euclidean distance between nodes by the exponential function of the node level difference to obtain the connection weight matrix between grid nodes.

[0058] Construct the thermal conduction relationship. Input the temperature field distribution data, temperature monitoring data, and strain monitoring data collected in real time into the conduction calculation model. Calculate the temperature gradient and strain gradient between adjacent grid nodes, and convert them into temperature influence factors and strain influence factors respectively through the exponential mapping function. Combine the thermal conductivity characteristics of the material to establish a calculation model for the thermal conductivity coefficient between adjacent grid nodes. Based on the current welding process parameters and thermal conductivity coefficient, construct a thermal conduction relationship network between grid cells.

[0059] The adaptive feature extraction mechanism inputs feature data of different scales into the adaptive attention module, and calculates the attention weights through the correlation analysis of grid features and global features. Perform weighted fusion on the grid scale hierarchical features after feature transformation to generate a fused feature vector. Input the fused feature into the bidirectional propagation network structure. In the forward propagation process, gradually extract the temporal features through the gated recurrent unit to obtain the forward hidden state. In the reverse propagation process, adopt a gating mechanism to extract features to obtain the reverse hidden state. Concatenate the forward and reverse hidden states to extract the complete spatio-temporal feature information.

[0060] Perform thermal deformation prediction and uncertainty analysis. Input the extracted spatio-temporal features and process constraint conditions into the prediction function module together. Through the topological thermal field self-correction mechanism, establish a deformation field prediction model to realize the prediction of the deformation state of metal parts. Based on the predicted deformation field and spatio-temporal features, construct an uncertainty estimation model to evaluate the reliability of the prediction results. Through comprehensive analysis, output the thermal deformation prediction results of metal parts.

[0061] Exemplarily, take the steel plate welding process as an example to divide the welding area into grids. Establish a basic hexahedron grid according to the plate size, and encrypt the grid in the weld area to form a multi-level grid structure. Divide the welding area into spatial units of different levels through the octree algorithm. Smaller grid cells are used to describe the area near the weld, and larger grid cells are used to describe the far-field area. Assign spatial position information and level identifiers to each grid node to construct a complete grid topology structure.

[0062] Collect the temperature field distribution data during the welding process, arrange temperature sensors and strain sensors at key positions to monitor the temperature and strain changes in real time. Based on the monitoring data, calculate the temperature difference and strain difference between adjacent grid nodes, and obtain the temperature influence and strain influence characteristics of the local area through feature transformation. Combine welding process parameters such as welding current, voltage, and welding speed to establish the heat transfer and force transfer relationships between grid cells.

[0063] The grid features are input into the adaptive attention module to evaluate the importance of features at different scales. The attention mechanism focuses on the regions with large temperature gradients and significant strain changes, and obtains the comprehensive feature representation through feature fusion. The feature information is processed in the bidirectional propagation network to extract the deformation features in the time dimension and space dimension, and capture the evolution law of thermal deformation.

[0064] Based on the extracted features and constraints, the deformation field distribution during the welding process is predicted. The prediction results are corrected through the self-correction mechanism, and at the same time, the prediction uncertainty is evaluated to give a reliable thermal deformation prediction result, providing a basis for welding process optimization.

[0065] In this embodiment, the recursive dynamic response analysis method ensures the continuity and stability of the calculation of velocity increment and displacement increment, effectively avoiding the oscillation phenomenon in the numerical calculation process. The strain decomposition algorithm can not only accurately capture the evolution law of thermal strain, but also timely identify the plastic deformation behavior of the material, ensuring the accuracy of stress calculation. According to the distribution characteristics of temperature gradient and stress gradient, the correction parameters are automatically adjusted to make the temperature field calculation more in line with the actual characteristics of the physical process, and can accurately reflect the dynamic evolution law of temperature field and stress field during the welding process, providing a reliable theoretical basis for welding process optimization and quality control, and having important engineering application value.

[0066] In an alternative embodiment, Input the spatio-temporal features and constraints into the prediction function, calculate the deformation field in combination with the topological thermal field self-correction mechanism, estimate the prediction uncertainty based on the deformation field and the spatio-temporal features, and the generated thermal deformation prediction results of the metal part include: Input the spatio-temporal features of the metal part and the preset constraints into the prediction function, where the constraints include the thermo-mechanical coupling constitutive equation, and the thermo-mechanical coupling constitutive equation establishes the correlation between the stress tensor and the strain tensor through the elastic constant tensor and the coefficient of thermal expansion; Construct a topological thermal field self-correction mechanism based on the persistent homology theory, calculate the Betti number sequence of the initial deformation field output by the prediction function to form the topological skeleton of the deformation field, calculate the persistence diagram of the initial deformation field according to the Betti number sequence, and calibrate the key deformation regions of the metal part according to the mutation points of the persistence diagram; For the key deformation regions, construct a distance metric function to calculate the characteristic evolution trajectory of the topological skeleton of the deformation field, project the initial deformation field onto the manifold space that satisfies the constraints to obtain the projected deformation field, and iteratively optimize the initial deformation field by minimizing the distance between the initial deformation field and the projected deformation field and the distance of the evolution trajectory of the topological skeleton of the deformation field to obtain the corrected deformation field; Calculate the persistent homology entropy according to the corrected deformation field, weight the persistent homology entropy and the corrected deformation field to obtain an uncertainty evaluation value, input the corrected deformation field, the uncertainty evaluation value, and the spatio-temporal features into a preset basis function, and generate a thermal deformation prediction result of the metal part by calculating the weighted sum of the basis function.

[0067] The prediction function construction process first preprocesses multi-dimensional spatio-temporal features. Resample the spatio-temporal features through an adaptive sampling strategy to ensure the uniformity of feature distribution. Use the sliding window method to segment the time series features, and enhance the expression ability of the features within each window through non-linear transformation. The processing of constraint conditions adopts a multi-level decomposition strategy, which transforms the thermo-mechanical coupling constitutive equation into a hierarchical constraint system. Integrate the influence of elastic constants and thermal expansion coefficients into the constraint space through tensor mapping to construct a complete stress-strain mapping relationship.

[0068] The persistent homology analysis adopts a multi-scale feature extraction method, constructs a simple complex sequence of the initial deformation field, and extracts topological features of each dimension through boundary operator operations. When calculating the Betti number sequence, a progressive feature filtering algorithm is used to screen out significant topological features layer by layer. For features of each dimension, construct a persistence metric index to record the life cycle of the features. Identify key points in the persistence diagram through feature clustering methods, and the key points correspond to important changes in the topological structure of the deformation field.

[0069] The topological skeleton construction process adopts a hierarchical processing strategy. Extract the main structural features at the coarsest scale to form the backbone of the skeleton. Gradually increase the precision and supplement the secondary structural features through a local refinement algorithm. In this process, use a dynamic threshold method to filter out noise features to ensure the stability of the skeleton structure. At the same time, establish a feature transfer mechanism to ensure the coherence between features at different scales.

[0070] The calibration of the key deformation area adopts a multi-criterion judgment method. Analyze the local features of the persistence diagram to construct a mutation point recognizer. The recognizer comprehensively considers factors such as feature persistence time, change amplitude, and spatial distribution, and uses a weighted voting mechanism to determine the key area. To improve the recognition accuracy, introduce spatial correlation analysis and consider the mutual influence of adjacent areas.

[0071] The construction of the distance metric function adopts an adaptive weight strategy. Establish a local feature extractor to capture the local geometric features and physical features of the deformation field. Dynamically adjust the weights of different features in distance calculation through feature importance analysis. At the same time, introduce time series correlation analysis to ensure the continuity of the feature evolution trajectory.

[0072] The deformed field projection adopts an iterative optimization strategy. A manifold space that meets the constraint conditions is constructed, and this space is formed through the implicit expression of the constraint conditions. The progressive projection algorithm is used to gradually project the initial deformed field into the target space. During the projection process, the projection direction is optimized by the gradient descent method to ensure convergence to the optimal solution.

[0073] The calculation of the characteristic evolution trajectory adopts the dynamic programming method. By constructing a state transition matrix, the state changes of the topological skeleton during the evolution process are recorded. The backtracking algorithm is used to extract the optimal evolution path, which not only meets the physical constraints but also maintains the continuity of topological features.

[0074] The calculation of the persistent homology entropy adopts the multi-scale analysis method. The corrected deformed field is decomposed at multiple scales, and the local entropy values are calculated at each scale. The global entropy value is obtained through scale-weighted combination, and this value reflects the overall complexity of the deformed field. At the same time, an entropy value dynamic update mechanism is established to track the changes of the deformed field in real time.

[0075] The generation of the prediction result adopts the adaptive basis function expansion method. A basis function library is constructed, which contains various types of basis functions. Through feature correlation analysis, the most suitable combination of basis functions for the current problem is selected. The adaptive weight allocation strategy is adopted to determine the contribution of each basis function to the final result. At the same time, a reliability evaluation mechanism for the prediction result is established, a confidence interval is configured for each predicted value, and the corresponding thermal deformation prediction result is determined based on the confidence interval.

[0076] Exemplarily, take the welding process of a steel bracket as an example. Spatiotemporal characteristic data such as the temperature field distribution and stress field distribution of the bracket during the welding process are collected. At the same time, the constraint conditions composed of the elastic modulus, thermal expansion coefficient, etc. of the material are imported to establish a thermo-mechanical coupling relationship model.

[0077] Topological analysis is performed on the initially predicted deformed field, and the Betti number sequence reflecting the characteristics of different dimensions is calculated. The topological skeleton of the bracket deformed field is constructed through the sequence, and this skeleton reflects the main structural characteristics of the deformed field. The generated persistent diagram shows significant mutation points at the weld joints and heat-affected zones of the bracket, and these regions are marked as key deformation regions.

[0078] Within the key deformation regions, a distance metric function considering geometric and physical characteristics is constructed to track the evolution process of the topological characteristics of the deformed field. The initial deformed field is projected into a constraint space considering material strength limitations and welding process requirements. Through the iterative optimization algorithm, the deformed field is adjusted to gradually meet the constraint conditions while maintaining the continuous evolution of topological characteristics, and finally the corrected deformed field is obtained.

[0079] Calculate the persistent homology entropy of the corrected deformation field, which reflects the complexity of the stent deformation process. Combine the entropy value with the deformation field data through weighting to obtain the uncertainty evaluation value of the prediction result. Finally, input all the calculation results into the preset basis function, and obtain the final thermal deformation prediction result of the stent through weighted combination, including the deformation amount at each key position and its reliability evaluation result.

[0080] In this embodiment, by constructing a topological thermal field self-correction mechanism based on persistent homology theory, the accurate extraction and analysis of the topological features of the deformation field are realized. Using the Betti number sequence and the persistence diagram, the topological skeleton of the deformation field is established, providing a theoretical basis for the accurate identification of key deformation regions. By constructing a special distance metric function and projection optimization strategy, it is ensured that the prediction result of the deformation field not only satisfies the physical constraints but also maintains the continuity of topological features. In the prior art, the welding deformation prediction method mainly relies on simple numerical simulations and empirical models, lacking in-depth analysis of the topological features of the deformation field, simplifying the thermo-mechanical coupling effect, making it difficult to accurately describe the deformation evolution process under complex working conditions, being relatively rough in the identification and treatment of key deformation regions, and often ignoring the influence of local deformation features on the overall prediction accuracy. The analysis method based on persistent homology theory in this embodiment improves the accuracy of deformation field feature extraction, enhances the adaptability of the prediction model to complex deformation patterns. The topological thermal field self-correction mechanism realizes the accurate identification and treatment of key deformation regions, improving the accuracy of local deformation prediction. By calculating and weighted combination of persistent homology entropy, a systematic uncertainty evaluation system is established, providing a quantitative basis for the reliability of the prediction result, realizing high-precision prediction and reliability evaluation of the deformation field during the welding process, and providing strong technical support for welding process optimization and quality control.

[0081] Figure 3 This is a comparison diagram of the topological skeleton evolution trajectories of the intelligent thermal deformation compensation and control method during the welding process of metal parts in the embodiment of the present invention. Through a regional display method, it clearly presents the comparison of the topological skeletons and evolution trajectories predicted by different methods during the welding thermal deformation of the steel stent. The top dotted contour represents the original shape of the stent, including the horizontal bottom plate and two vertical columns, and the two circular marked points represent the left and right welding point positions respectively.

[0082] The first region shows the topological skeleton constructed by this technical solution (solid line, diamond mark). According to the experimental data, the maximum deformation amount near the left welding point of this solution reaches 7.82 mm, and the maximum deformation amount near the right welding point reaches 8.15 mm, and the deviation from the actual measurement value does not exceed 3%. Through the analysis of persistent homology theory, the Betti numbers identified by this technical solution are = 23.7 and = 15.3, indicating that the topological skeleton has rich structural features. The persistent homology entropy H = 2.58 and the evolution trajectory distance D = 0.127, reflecting that the deformation field has rich topological information and low topological structure variability. Particularly prominent are the obvious concave deformation of the bracket bottom plate between the two welding points and the outward inclination deformation of the two columns, which is highly consistent with the actual thermal deformation law. The areas marked by the two dashed circles in the figure are the key deformation areas. The characteristic Betti numbers of the left key deformation area (located at x = 300, y = 220) are = 14.3, = 8.7. The characteristic Betti numbers of the right key deformation area (located at x = 580, y = 230) are = 11.5, = 6.2. The deformation gradient at the left key deformation area is the largest, reaching 0.87 mm / mm, while the deformation gradient of the right key deformation area is 0.63 mm / mm.

[0083] The second area shows the topological skeleton predicted by the finite element method (ABAQUS, dashed line, square marker). The deformation amount predicted by the finite element method is significantly smaller. The maximum deformation amount near the left welding point is only 5.23 mm, and near the right welding point is 6.47 mm. The deviation compared with the actual measurement values reaches 32% and 25.7%. The Betti numbers of the finite element method = 8.5, = 5.2, and the persistent homology entropy H = 0.93, indicating that it is difficult to capture the fine structural features of the deformation field. It can be clearly seen from the figure that the deformation of the bottom plate predicted by the finite element method is almost flat, lacking the description of the local deformation characteristics in the middle area (at x = 300 - 450), which does not match the observed wavy deformation. The deformation prediction of the column is also relatively conservative and does not show enough outward inclination trend.

[0084] The third area shows the topological skeleton predicted by the traditional machine learning method (SVR, dotted line, triangle marker). There are obvious deviations in the prediction of the deformation trend. Especially in the middle area of the bracket bottom plate (at x = 300 - 450), it shows a convex shape, which is completely opposite to the actual concave deformation. In the deformation prediction of the left column, the traditional machine learning method shows an almost vertical deformation line, which does not match the observed outward inclination deformation. The deformation prediction accuracy of this method in the heat affected zone is relatively low, and the average error reaches 7.88%, mainly because it lacks due consideration of the welding heat affected mechanism and only fits based on data characteristics.

[0085] The fourth area shows the topological skeleton predicted by the deep neural network method (CNN, dashes, cross marks). Although this method can better describe the overall deformation trend, showing the concave bottom plate and the outward inclination of the column, it is still inferior to the present technical solution in capturing the fine features of the key deformation area. In particular, the deformation characteristics at the connection between the right column and the bottom plate are not described enough, and the predicted deformation is 7.43mm, which is 8.8% different from the actual measured value of 8.15mm. The deformation curve of the deep neural network in the middle of the bottom plate is wavy, but the peak position is offset by about 2.5mm from the actual measurement result, and the peak amplitude is too large, which indicates that it does not understand the deformation mechanism caused by the thermal gradient distribution enough.

[0086] By clearly displaying the prediction results of each method by region, it can be clearly seen that the topological thermal field self-correction mechanism constructed by this technical solution based on the theory of persistent homology can most accurately capture the spatiotemporal evolution characteristics of thermal deformation, especially at the boundary of the heat-affected zone and the non-uniform deformation area, where the predicted morphology is closest to the actual deformation. Experimental data show that in the area with the steepest temperature gradient (within 5-10mm around the weld), the average deformation prediction error of this technical solution is 3.12%, while the errors of the finite element method, traditional machine learning method and deep neural network are 6.24%, 7.88% and 4.83% respectively. High-precision prediction has important engineering application value for the welding deformation control of precision metal components.

[0087] In an optional embodiment, The thermal deformation prediction result is compared with the preset thermal deformation threshold to determine the thermal deformation deviation. Based on the thermal deformation deviation, the welding process parameters are used as optimization variables, and a recursive optimizer combining the integrated gradient strategy and the evolution strategy is used to calculate the compensation scheme. The optimal welding process parameters are iteratively solved and sent to the welding assembly for execution, including: The thermal deformation prediction result is compared with a preset thermal deformation threshold to obtain a thermal deformation deviation, and the thermal deformation deviation is time-integrated to calculate the accumulated thermal deformation deviation; Taking the welding process parameters as optimization variables, calculating the gradient of the thermal deformation deviation to the welding process parameters and constructing a gradient strategy, combining Gaussian noise to construct an evolution strategy, and integrating the gradient strategy and the evolution strategy to construct a recursive optimizer; Based on the recursive optimizer, an optimization objective function including the thermal deformation deviation is established, and based on the gradient of the thermal deformation deviation, the value range and change rate of the welding process parameters are constrained. The welding process parameters that meet the convergence conditions are used as the optimal welding process parameters and sent to the welding assembly to perform the welding operation.

[0088] The calculation of thermal deformation deviation adopts a hierarchical comparison strategy. The predicted results of thermal deformation are divided according to different spatial regions and time periods, and are compared layer by layer with the preset thermal deformation threshold. The comparison process adopts an adaptive weight mechanism, focusing on the deviation in the deformation-sensitive area. The cumulative effect of thermal deformation deviation is calculated by a time integration algorithm, and a variable step-size integration strategy is adopted. A smaller integration step-size is used in the area with severe deformation to improve the calculation accuracy.

[0089] The construction of the gradient strategy adopts a multi-level analysis method. The welding process parameters are used as optimization variables to construct a parameter sensitivity analysis model. The gradient of thermal deformation deviation with respect to each process parameter is calculated by the forward difference method, and an adaptive step-size control strategy is adopted to improve the stability of gradient calculation. During the gradient calculation process, a local smoothing mechanism is introduced to reduce the influence of numerical fluctuations.

[0090] The construction of the evolutionary strategy adopts a random perturbation method. A random perturbation with specific distribution characteristics is generated by a Gaussian noise generator, and the perturbation amount is determined by an adaptive adjustment mechanism. According to the convergence situation of the optimization process, the intensity and distribution characteristics of the noise are dynamically adjusted. At the same time, a perturbation effect evaluation mechanism is established to screen effective perturbation directions.

[0091] The construction of the recursive optimizer adopts a dual optimization strategy. The gradient strategy and the evolutionary strategy are fused, and the contribution ratio of the two strategies is determined by a weight allocation mechanism. The optimizer adopts an adaptive learning rate adjustment mechanism to dynamically adjust the parameter update step-size according to the optimization process. At the same time, an optimization history record is established to guide the subsequent optimization direction.

[0092] The construction of the optimization objective function adopts a multi-objective trade-off strategy. The thermal deformation deviation is used as the main optimization objective, and the feasibility constraints of the process parameters are considered at the same time. By constructing a composite weight function, the importance between different objectives is balanced. During the optimization process, a dynamic weight adjustment mechanism is adopted to adjust the weights of each objective according to the optimization process.

[0093] The constraint handling adopts a multi-level constraint system. The value range of the welding process parameters is hierarchically constrained, including physical feasibility constraints and process realizability constraints. By constructing a constraint violation degree evaluation function, it is quantified whether the parameters meet the constraint conditions. The constraint on the parameter change rate adopts a sliding window strategy to ensure the smoothness of parameter changes.

[0094] The convergence judgment adopts a multi-criterion evaluation mechanism. Multiple convergence indicators are set, including the relative change of the objective function value, the norm of the parameter update amount, the degree of constraint satisfaction, etc. By constructing a comprehensive evaluation function, each indicator is weighted and combined. When all indicators meet the preset threshold, the optimal welding process parameters are determined.

[0095] Exemplarily, taking the butt welding of steel plates as an example, the thermal deformation data during the welding process is collected. Through comparative analysis, it is found that the deformation amount within 50 mm on both sides of the weld exceeds the preset threshold. The time integral algorithm is used to calculate the cumulative deformation deviation, and key attention is paid to the time period when the deformation amount grows rapidly during the welding process.

[0096] Taking process parameters such as welding current, voltage, and welding speed as optimization variables, calculate the influence degree of these parameters on the deformation deviation. Through gradient calculation, it is found that the welding speed has the most significant influence on the deformation. Based on this, a gradient optimization strategy dominated by the welding speed is constructed.

[0097] During the optimization process, Gaussian noise is introduced to randomly perturb the process parameters to avoid the optimization falling into a local optimal solution. Combining the gradient strategy and the evolutionary strategy, a recursive optimizer is constructed. The optimization goal is set to minimize the thermal deformation deviation while ensuring that the process parameters change smoothly within the feasible range.

[0098] Through the iterative optimization process, when the relative change of the objective function value is less than the preset threshold and the process parameters meet all constraint conditions, the optimal combination of welding process parameters is obtained. This set of parameters can not only effectively control the welding deformation but also meet the feasibility requirements of process implementation.

[0099] In this embodiment, the adaptive weight mechanism adopted in the deformation-sensitive area improves the accuracy of deviation calculation. The adaptive step size control strategy and the local smoothing mechanism improve the stability and reliability of gradient calculation. Through parameter sensitivity analysis, the process parameters that play a key role in deformation control are accurately identified. The random perturbation mechanism introduced by Gaussian noise effectively avoids the optimization falling into a local optimal solution. The adaptive learning rate adjustment mechanism ensures the stable convergence of the optimization process. While ensuring the calculation efficiency, it realizes the precise optimization of welding process parameters and provides effective technical support for welding deformation control.

[0100] Figure 4 This is the flowchart for solving the optimal welding parameters of the intelligent thermal deformation compensation and control method for metal part welding in the embodiment of the present invention.

[0101] In the second aspect of the embodiment of the present invention, there is provided an intelligent thermal deformation compensation and control system for metal part welding, and the system includes: A first unit for obtaining the three-dimensional model data of the metal part and the welding process parameters; A second unit for constructing a thermo-mechanical coupling analysis model based on the three-dimensional model data, dividing the metal part into multiple hexahedron mesh units and establishing a non-linear heat conduction equation set, modeling the welding heat source as a double ellipsoid heat source distribution function, constructing a coupling transfer matrix of the temperature field and the stress field, and using an explicit integration algorithm to solve the non-linear heat conduction equation set to obtain the transient temperature field distribution of the metal part. A third unit for collecting real-time temperature data and real-time strain data during the welding process of the metal part; A fourth unit for constructing the hexahedral mesh elements into a dynamic hierarchical graph structure, calculating the thermal conduction relationship between mesh elements based on the transient temperature field distribution, the real-time temperature data, the real-time strain data, and the welding process parameters, fusing features of different scales using an adaptive attention mechanism, extracting spatio-temporal features through a bidirectional propagation neural network, and generating a thermal deformation prediction result of the metal part; A fifth unit for comparing the thermal deformation prediction result with a preset thermal deformation threshold to determine a thermal deformation deviation, based on the thermal deformation deviation, taking the welding process parameters as optimization variables, calculating a compensation scheme using a recursive optimizer that combines the integrated gradient strategy and the evolutionary strategy, and iteratively solving to obtain optimal welding process parameters and sending them to the welding assembly for execution.

[0102] In a third aspect of the embodiments of the present invention, an electronic device is provided, including: A processor and a memory for storing processor-executable instructions, wherein the processor is configured to call the instructions stored in the memory to execute the method described above.

[0103] In a fourth aspect of the embodiments of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0104] The present invention can be a method, an apparatus, a system, and / or a computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions for performing various aspects of the present invention loaded thereon.

[0105] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. Intelligent compensation and control method for thermal deformation during metal welding, characterized in that: include: Obtain 3D model data and welding process parameters of metal parts; A thermal-mechanical coupling analysis model is constructed based on the three-dimensional model data, the metal part is divided into a plurality of hexahedral grid units and a nonlinear heat conduction equation group is established, the welding heat source is modeled as a double ellipsoid heat source distribution function, a coupling transfer matrix of the temperature field and the stress field is constructed, and an explicit integral algorithm is used to solve the nonlinear heat conduction equation group to obtain the transient temperature field distribution of the metal part; Collecting real-time temperature data and real-time strain data during the welding process of the metal parts; The hexahedral grid unit is constructed into a dynamic hierarchical graph structure, the thermal conduction relationship between the grid units is calculated based on the transient temperature field distribution, the real-time temperature data, the real-time strain data and the welding process parameters, the adaptive attention mechanism is used to fuse the features of different scales, the spatiotemporal features are extracted through a bidirectional propagation neural network, and the thermal deformation prediction result of the metal part is generated; The thermal deformation prediction result is compared with the preset thermal deformation threshold to determine the thermal deformation deviation. Based on the thermal deformation deviation, the welding process parameters are used as optimization variables, and the compensation scheme is calculated by a recursive optimizer combining the integrated gradient strategy and the evolution strategy. The optimal welding process parameters are iteratively solved and sent to the welding assembly for execution.

2. The method according to claim 1, characterized in that: Constructing a thermal-mechanical coupling analysis model based on the three-dimensional model data, dividing the metal part into a plurality of hexahedral grid units and establishing a nonlinear heat conduction equation group includes: Acquire the three-dimensional model data of the metal part, identify and extract the key geometric features of the weld area and the heat-affected zone after eliminating duplicate faces and dangling edges, establish a feature topological relationship diagram, establish a stress analysis equation group based on the feature topological relationship diagram, associate and map the stress analysis equation group with the temperature field distribution equation group, and construct a thermal-mechanical coupling analysis model; Calculating the surface curvature distribution of the three-dimensional model data based on the thermal-mechanical coupling analysis model, determining a curvature value according to the surface curvature distribution, comparing the curvature value with a preset curvature threshold to determine a mesh encryption area, determining a minimum mesh size and a maximum mesh size according to the mesh encryption area, establishing a mesh size calculation function, meshing the metal part, and generating an initial hexahedral mesh unit; The determinant value, grid angle and side length ratio of the initial hexahedral grid unit are calculated and compared with the preset grid quality parameters, the node positions of the hexahedral grid units that do not meet the preset grid quality parameters are adjusted and the boundaries are reconstructed, the irregular hexahedral grid units are segmented, and the optimized hexahedral grid units are obtained, and a nonlinear heat conduction equation group including the temperature-dependent specific heat capacity coefficient and the temperature-dependent thermal conductivity coefficient is established.

3. The method according to claim 1, characterized in that The welding heat source is modeled as a double ellipsoid heat source distribution function, the coupled transfer matrix of the temperature field and the stress field is constructed, and the nonlinear heat conduction equations are solved by an explicit integration algorithm to obtain the transient temperature field distribution of the metal part, including: According to welding voltage, welding current, thermal efficiency, ellipsoid major semi-axis, ellipsoid width semi-axis, ellipsoid depth semi-axis and heat distribution coefficient, the front ellipsoid heat source distribution function and the rear ellipsoid heat source distribution function are established to obtain the initial heat source density distribution; Calculating the temperature distribution according to the initial heat source density distribution, obtaining the temperature-related thermal expansion coefficient and calculating the thermal strain, and superimposing the thermal strain with the elastic strain and the plastic strain to obtain the total strain; Based on the total strain, a heat conduction stiffness matrix, a heat capacity matrix, a heat load vector, a structural stiffness matrix, a structural damping matrix and a mechanical load vector are constructed, and a coupled transfer matrix of a temperature field and a stress field and a corresponding nonlinear heat conduction equation group are constructed in combination, and an initial time step of explicit integration is determined based on a unit characteristic length, an elastic wave velocity and a system characteristic frequency; At the initial time step of the explicit integration, the velocity increment and the displacement increment of the nonlinear heat conduction equation group are calculated by the central difference format, the stress increment is determined and substituted into the nonlinear heat conduction equation group for explicit integration calculation to obtain the initial value of the temperature field, the temperature error is calculated based on the initial value of the temperature field, the adaptive time step is determined and substituted into the nonlinear heat conduction equation group for explicit integration iterative calculation to obtain the transient temperature field distribution of the metal part.

4. The method according to claim 3, characterized in that At the initial time step of the explicit integration, the velocity increment and displacement increment of the nonlinear heat conduction equations are calculated by the central difference format, the stress increment is determined and substituted into the nonlinear heat conduction equations for explicit integration calculation, and the initial value of the temperature field is obtained, including: At the initial time step of explicit integration, the velocity increment at the previous moment is added to the product of the time step and the acceleration to obtain the velocity increment at the current moment, and the displacement increment at the previous moment is added to the product of the time step and the velocity increment at the current moment to obtain the displacement increment at the current moment; The thermal expansion coefficient is multiplied by the temperature increment and the unit tensor to obtain the thermal strain increment. Whether plastic deformation occurs is determined according to the current stress state and the plastic strain increment is calculated. The thermal strain increment and the plastic strain increment are subtracted from the total strain increment to obtain the elastic strain increment. The stress increment is calculated by the elastic modulus and the elastic strain increment. The partial derivative of internal energy with respect to time is subtracted from the heat flux density divergence and the heat source term and the absolute value is taken to calculate the energy balance error and the correction coefficient, the exponential function of the absolute value of the temperature gradient is multiplied by the linear function of the absolute value of the gradient of the stress increment to calculate the adaptive weight function, the correction coefficient is multiplied by the energy balance error and the temperature gradient direction and the negative value is taken to calculate the temperature correction amount, and the corrected temperature field is calculated based on the adaptive weight function and the temperature correction amount; The stress increment is substituted into the nonlinear heat conduction equation group for explicit integral calculation to obtain the second temperature field distribution, the norm ratio of the difference between the corrected temperature field and the second temperature field distribution and the norm of the energy balance error are compared with the temperature field convergence tolerance and the energy balance convergence tolerance respectively, and an iterative calculation is performed based on the comparison result to obtain the initial value of the temperature field.

5. The method according to claim 1, characterized in that The hexahedral grid unit is constructed into a dynamic hierarchical graph structure, the thermal conduction relationship between the grid units is calculated based on the transient temperature field distribution, the real-time temperature data, the real-time strain data and the welding process parameters, the adaptive attention mechanism is used to fuse the features of different scales, and the spatiotemporal features are extracted through a bidirectional propagation neural network to generate the thermal deformation prediction results of the metal part, including: Based on the coordinates of the center points of the hexahedral grid units, an octree space partitioning structure is constructed, the topological adjacency relationship between the grid units is calculated, the spatial coordinates and hierarchical index are assigned to each grid node, a dynamic hierarchical graph structure is generated, the geometric features, topological features and physical features of each grid node are extracted to form a node feature vector, and the product of the Euclidean distance square exponential function of the grid node spatial position and the grid node hierarchical difference exponential function is calculated to obtain the connection weight between the grid nodes; Input the transient temperature field distribution, real-time temperature data and real-time strain data into the conduction calculation model, calculate the temperature difference and strain difference between adjacent grid nodes, and perform exponential mapping to obtain the temperature influence factor and the strain influence factor respectively, determine the heat conduction coefficient between adjacent grid nodes in combination with the thermal conductivity of the material, and establish the thermal conduction relationship between grid units based on the heat conduction coefficient and the current welding process parameters; Input different scale features into the adaptive attention mechanism, calculate the scale attention weight by the dot product operation of the grid scale level features and the global feature matrix, perform weighted summation on the grid scale level features after feature transformation, obtain fusion features and input them into the bidirectional propagation neural network, use the gated recurrent unit to perform forward propagation to obtain the forward hidden state, perform back propagation to obtain the reverse hidden state, splice the forward hidden state and the reverse hidden state, and extract spatiotemporal features; The spatiotemporal characteristics and constraints are input into a prediction function, the deformation field is calculated in combination with a topological thermal field self-correction mechanism, prediction uncertainty is estimated based on the deformation field and the spatiotemporal characteristics, and a thermal deformation prediction result of a metal part is generated.

6. The method according to claim 5, characterized in that Inputting the spatiotemporal characteristics and constraints into the prediction function, calculating the deformation field in combination with the topological thermal field self-correction mechanism, estimating the prediction uncertainty based on the deformation field and the spatiotemporal characteristics, and generating the thermal deformation prediction result of the metal part includes: Inputting the spatiotemporal characteristics of the metal part and pre-set constraints into a prediction function, wherein the constraints include a thermomechanical coupling constitutive equation, and the thermomechanical coupling constitutive equation establishes a correlation relationship between a stress tensor and a strain tensor through an elastic constant tensor and a thermal expansion coefficient; A topological thermal field self-correction mechanism is constructed based on the persistence homology theory, the Betti number sequence of the initial deformation field output by the prediction function is calculated to form a topological skeleton of the deformation field, the persistence map of the initial deformation field is calculated according to the Betti number sequence, and the key deformation area of ​​the metal part is calibrated according to the mutation point of the persistence map; For the key deformation area, a distance metric function is constructed to calculate the characteristic evolution trajectory of the deformation field topological skeleton, the initial deformation field is projected to the manifold space that satisfies the constraint condition to obtain the projected deformation field, and the initial deformation field is iteratively optimized by minimizing the distance between the initial deformation field and the projected deformation field and the evolution trajectory distance of the deformation field topological skeleton to obtain the corrected deformation field; The continuous coherent entropy is calculated according to the corrected deformation field, the continuous coherent entropy and the corrected deformation field are weighted to obtain an uncertainty assessment value, the corrected deformation field, the uncertainty assessment value and the spatiotemporal characteristics are input into a preset basis function, and the thermal deformation prediction result of the metal part is generated by calculating the weighted sum of the basis functions.

7. The method according to claim 1, characterized in that The thermal deformation prediction result is compared with the preset thermal deformation threshold to determine the thermal deformation deviation. Based on the thermal deformation deviation, the welding process parameters are used as optimization variables, and a recursive optimizer combining the integrated gradient strategy and the evolution strategy is used to calculate the compensation scheme. The optimal welding process parameters are iteratively solved and sent to the welding assembly for execution, including: The thermal deformation prediction result is compared with a preset thermal deformation threshold to obtain a thermal deformation deviation, and the thermal deformation deviation is time-integrated to calculate the accumulated thermal deformation deviation; Taking the welding process parameters as optimization variables, calculating the gradient of the thermal deformation deviation to the welding process parameters and constructing a gradient strategy, combining Gaussian noise to construct an evolution strategy, and integrating the gradient strategy and the evolution strategy to construct a recursive optimizer; Based on the recursive optimizer, an optimization objective function including the thermal deformation deviation is established, and based on the gradient of the thermal deformation deviation, the value range and change rate of the welding process parameters are constrained. The welding process parameters that meet the convergence conditions are used as the optimal welding process parameters and sent to the welding assembly to perform the welding operation.

8. An intelligent compensation and control system for thermal deformation in metal welding, used to implement the method described in any one of claims 1 to 7, characterized in that: include: The first unit is used to obtain the three-dimensional model data and welding process parameters of the metal parts; The second unit is used to construct a thermal-mechanical coupling analysis model based on the three-dimensional model data, divide the metal part into a plurality of hexahedral grid units and establish a nonlinear heat conduction equation group, model the welding heat source as a double ellipsoid heat source distribution function, construct a coupling transfer matrix of the temperature field and the stress field, and solve the nonlinear heat conduction equation group using an explicit integral algorithm to obtain the transient temperature field distribution of the metal part; The third unit is used to collect real-time temperature data and real-time strain data during the welding process of the metal parts; A fourth unit is used to construct the hexahedral grid unit into a dynamic hierarchical graph structure, calculate the thermal conduction relationship between the grid units based on the transient temperature field distribution, the real-time temperature data, the real-time strain data and the welding process parameters, fuse the features of different scales using an adaptive attention mechanism, extract the spatiotemporal features through a bidirectional propagation neural network, and generate a thermal deformation prediction result of the metal part; The fifth unit is used to compare the thermal deformation prediction result with the preset thermal deformation threshold to determine the thermal deformation deviation. Based on the thermal deformation deviation, the welding process parameters are used as optimization variables, and the compensation scheme is calculated by the recursive optimizer combining the integrated gradient strategy and the evolution strategy. The optimal welding process parameters are iteratively solved and sent to the welding assembly for execution.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images