A high-precision processing method and system for multi-specification copper tubes

Abstract

The present invention discloses a high-precision processing method and system for multi-specification copper tubes, specifically related to the technical field of copper tube processing analysis. Raw material parameters, process timing data, and detection data during the processing of multi-specification copper tubes are collected. Through multi-source data acquisition and physical enhancement generation technology, combined with the B-spline interpolation algorithm and conditional adversarial generation network, an enhanced data set is constructed to solve the problems of sparse measured data and insufficient coverage of extreme working conditions. A grain evolution-stress coupling simulation model is established to quantify the influence of grain boundary migration on stress, and the coordinates and correction coefficients of high-stress regions are marked. A graph convolutional network is used to achieve cross-scale stress prediction, and dynamic process parameters are generated through a multi-objective optimization algorithm. Finally, real-time control is driven by digital twins, solving the technical problems of low deformation control accuracy and high defect rate caused by uneven residual stress distribution in traditional copper tube processing.

Description

Technical Field

[0001] The present invention relates to the technical field of copper tube processing analysis, and more specifically, to a high-precision processing method and system for multi-specification copper tubes. Background Art

[0002] As the core link of precision processing of copper tubes, the cold drawing process causes plastic deformation of the tubes through die drawing. In the fields of new energy, microelectronics, etc., the high-precision requirements for multi-specification copper tubes are becoming increasingly stringent. Especially for thin-walled and large diameter-thickness ratio copper tubes, the non-linear effect of the residual stress distribution after cold drawing is significant, directly affecting the qualification rate of subsequent processing (such as flaring, bending) and service life.

[0003] The problems faced by the existing cold drawing process are as follows:

[0004] Limitations of experience-driven methods: Traditional process design highly relies on empirical formulas, fails to effectively integrate the internal relationships of multi-source heterogeneous data such as process parameters, material properties, and equipment status, and lacks the cross-scale modeling ability of the microstructure evolution (grain boundary migration, dislocation density change) and macroscopic stress transfer mechanism of copper tubes, resulting in frequent quality problems such as residual stress concentration and dimensional tolerance.

[0005] Lag in dynamic control: In industrial production, the real-time regulation of process parameters mainly relies on the PID control strategy based on fixed rules, and it is difficult to establish a real-time coupling relationship model between geometric specification parameters and the dynamic stress field. Under abnormal working conditions such as sudden changes in cold drawing speed and temperature fluctuations, the control system responds laggingly, and the virtual data generation method has the defect of violating the stress balance condition, resulting in the distortion of the prediction model. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art, the present invention provides a high-precision processing method and system for multi-specification copper tubes, which realizes the intelligent closed-loop control of processing quality through multi-modal data fusion and physical constraint modeling to solve the problems proposed in the above background art.

[0007] To achieve the above object, the present invention provides the following technical solution: A high-precision processing method for multi-specification copper tubes, comprising:

[0008] Step 1: Multi-source data collection and physical enhancement generation: Collect raw material parameters (grain size distribution data, impurity content, geometric dimensions), process timing data (cold drawing speed timing, heat treatment temperature curve, cooling rate curve), and inspection data (actual residual stress data, such as X-ray diffraction residual stress distribution on the surface and cross-section, ultrasonic flaw detection defect coordinates) during the processing of multi-specification copper tubes to obtain real sample data; Process through the B-spline interpolation algorithm and conditional adversarial generation network. Improve the spatial matching accuracy through the B-spline interpolation algorithm; Expand the dataset through finite element simulation, and obtain virtual sample data through the conditional adversarial generation network. The virtual sample data and the aligned real sample data constitute the enhanced dataset; The virtual sample data needs to satisfy the axial force balance and Hollomon hardening model constraints; The loss function of the conditional adversarial generation network includes adversarial loss, axial force balance term and Hollomon hardening model constraints to ensure that the generated data conforms to mechanical laws; The axial force balance term means that in the z-axis direction, the vector sum of all normal stresses (tensile stress or compressive stress) acting on cross-section A is zero, indicating that the object has no net external force in the axial direction and is in a static equilibrium state; The Hollomon hardening model refers to the relationship that during the plastic deformation process of copper tubes, the stress increases with the strain in a power function

[0009] Step 2: Input the enhanced dataset, enhance the dataset through grain evolution-stress coupling simulation modeling and physical constraint-driven process parameter correction, output the physical verification dataset, and mark the coordinates of high-stress regions and correction coefficients

[0010] Step 3: Input the physical verification dataset and the defect-stress mapping table, and build a stress prediction model through the XGBoost regression model

[0011] Step 4: Cross-modal causal enhancement-based hybrid proxy optimization: Input the stress prediction model and processing target constraints; Generate the optimal parameter combination that meets the processing target constraints based on the multi-objective optimization algorithm (NSGA-II), and dynamically adjust the process parameters in combination with real-time stress data

[0012] Step 5: Digital twin-driven real-time control: Input the optimal parameter combination and real-time sensor data, combine fuzzy logic and PID control, dynamically adjust the process parameters, and output real-time process adjustment instructions, process adjustment log information, and processing quality reports

[0013] Preferably, the B-spline interpolation algorithm with local curvature weight factors is used to perform three-dimensional interpolation on the ultrasonic defect coordinates. When mapping the ultrasonic defect coordinates to the X-ray stress grid, the spatial offset error between defect positioning and the stress field is eliminated, and the spatial matching accuracy is improved. When using the conditional adversarial generation network to generate virtual sample data, extreme working conditions of the copper tube are covered, such as cold drawing speeds of 1.0 - 3.0 m / s, temperatures of 450 - 600 °C, and stress data under extreme working conditions of copper tubes with wall thicknesses less than 1 mm and diameter-to-thickness ratios greater than 30.

[0014] Preferably, an enhanced data set is used for grain evolution-stress coupling simulation modeling to quantify the dynamic effect of grain boundary migration on residual stress and generate a physical verification data set containing data on the influence of grain boundary migration;

[0015] The data on the influence of grain boundary migration includes the dynamic changes in grain boundary position, grain size, and dislocation density, as well as the corresponding residual stress distribution;

[0016] The cellular automaton model is used for simulation modeling. The state of the grain unit is defined as the orientation angle θ and the dislocation density ρ. Based on the Moore neighborhood rule, the grain boundary migration rate is calculated, and the grain orientation field is dynamically updated; the grain boundary migration rate reflects the influence of grain dynamic evolution on residual stress;

[0017] According to the grain size d and the dislocation density ρ, the anisotropic yield strength is calculated through the following formula, and the simulated residual stress distribution is obtained based on the anisotropic yield strength:

[0018] ;

[0019] where is the matrix strength of the copper tube, G represents the shear modulus, k is the grain boundary strengthening coefficient calibrated through nanoindentation experiments, δ is the dislocation strengthening coefficient obtained based on the Taylor relationship of the dislocation density ρ; usually ; k, , G, and b (Burgers vector) are material constants; combining the contributions of dislocation hardening and grain refinement to strength helps simulate the microscopic behavior of the material during copper tube processing;

[0020] The generated simulated residual stress distribution is compared with the actual residual stress data (such as the X-ray residual stress distribution) in the enhanced data set to verify the simulation accuracy; if differences are found, the abnormal parts in the enhanced data set are identified and corrected.

[0021] Preferably, based on physical constraint-driven process parameter correction: the circumferential tensile stress area ratio in the simulated residual stress distribution is calculated in real time. If it exceeds a preset value, such as 40%, a sequential quadratic programming optimization is triggered to adjust the cold drawing speed v and the temperature T, and the objective function is:

[0022] ;

[0023] Among them, is the proportion of the circumferential tensile stress area, representing the abnormal area ratio of the circumferential tensile stress of the copper tube; is the deviation of the cold drawing speed from the initial speed, is the deviation of the temperature from the initial temperature;

[0024] Based on the optimization results of the objective function, a defect-stress mapping table for annotating the coordinates of high-stress areas and correction factors is as follows:

[0025] The objective function reduces the proportion of the circumferential tensile stress area by optimizing the cold drawing speed and temperature; the areas that still exceed the threshold in the simulated residual stress distribution after optimization are identified as high-stress areas, and their coordinates and the corresponding speed and temperature correction factors are recorded to generate a defect-stress mapping table.

[0026] Preferably, building a stress prediction model includes:

[0027] Converting the copper tube geometry into a graph structure, where the nodes are discrete grid points, and the features include position coordinates, real-time hardness, and local cold drawing speed; the edges are defined as the connections between adjacent nodes, and the edge weights are calculated from the stress gradient and grain boundary orientation difference between adjacent nodes;

[0028] In the present invention, the edge weight is ; among them, represents the circumferential stress gradient, is the grain orientation angle, used to describe the grain orientation radian; represents the grain boundary orientation difference, reflecting the difference in grain orientation radian between adjacent grains; Explanation: The edge weight coefficient is set based on the contribution ratio of the grain boundary orientation difference to stress transfer. For example, the contribution ratio of the grain boundary orientation difference to stress transfer is 30%-50%, so 0.5 is taken to balance the influence of the gradient and the orientation difference;

[0029] Implementing cross-scale coupling modeling for characterizing grain boundary migration and dislocation density evolution by combining a graph convolutional network (GAT+IGC): calculating the influence of grain boundary slip on the stress of adjacent nodes, the grain boundary orientation difference and stress gradient in the edge weight are directly related to the grain boundary migration behavior, and the node hardness feature is related to the dislocation density evolution; regularly updating the graph structure to reflect the change in edge weight caused by grain coarsening; using incremental graph convolution (IGC) to manage the computational efficiency, and using multi-head graph convolution (GAT) to calculate the influence of grain boundary slip on the stress of adjacent nodes;

[0030] Training an XGBoost regression model, with node features and edge weights as inputs, to predict the stress value at the nodes; using a physical verification data set to ensure axial force balance verification; during the training process, the loss function includes the mean square error and the axial force balance penalty term, and the hyperparameters of the stress prediction model are optimized by the Bayesian method;

[0031] The post - processing model predicts to ensure that the proportion of the circumferential tensile stress area does not exceed 20%, and the region growing algorithm is used to correct the over - standard area.

[0032] Preferably, the training process of the stress prediction model includes:

[0033] Pre - training is performed using finite - element simulation data verified for axial force balance to ensure the axial force balance term , where is a predetermined tolerance;

[0034] Fine - tuning is performed using experimental data. The loss function of the stress prediction model includes the mean square error and the axial force balance penalty term , and the hyperparameters are adjusted through Bayesian optimization;

[0035] The post - processing uses the region growing algorithm to ensure that the proportion of the circumferential tensile stress area does not exceed 20% and corrects the over - standard area.

[0036] Preferably, the process of obtaining the optimal parameter combination includes:

[0037] Construct a cross - modal causal model, input the prediction results of the stress prediction model and the processing target constraints (maximum deformation < 0.05mm, circumferential stress, axial stress), and identify the causal effect of process parameters on the stress distribution through causal inference methods to generate a causal effect diagram;

[0038] Based on the causal effect diagram, set the goal to minimize the stress concentration and deformation, and use the NSGA - II algorithm to generate a Pareto - optimal parameter set;

[0039] Combined with real - time stress data, calculate the real - time stress concentration and trigger process parameter adjustment;

[0040] Output the optimal parameter combination and the switching strategy.

[0041] Preferably, it includes the stress prediction model performance verification step: calculate the stress prediction model adaptation index through the processing quality report and the process adjustment log information. If the stress prediction model adaptation index is lower than the threshold, trigger an alarm and update the stress prediction model.

[0042] Preferably, the method for obtaining the stress prediction model adaptation index is:

[0043] Step S11, calculate the processing quality fluctuation factor :

[0044] ;

[0045] where represents the actual deformation of the copper tube, Represents the standard deviation of the reference deformation amount (statistical value of historical data), Represents the reference stress uniformity index, and S represents the actual stress uniformity index; an exponential function is introduced to enhance the sensitivity of S. When S is lower than it significantly improves the fluctuation factor, more accurately reflecting the deterioration of the stress distribution;

[0046] Step S12, Process adjustment stability factor :

[0047] ;

[0048] Among them, Represents the actual adjustment frequency of the process, Represents the reasonable adjustment frequency threshold, Represents the actual adjustment amplitude, Represents the upper limit of the allowable adjustment amplitude; an inverse proportional function is used to suppress the excessive adjustment amplitude, avoiding system oscillations caused by frequent large adjustments;

[0049] Step S13, Stress prediction model adaptation index :

[0050] ;

[0051] Among them, Represents the time decay coefficient, Represents the current time, Represents the last update time of the model; it is notified to add a time decay term to make the adaptation index of the long-unupdated model decrease rapidly, forcing periodic verification; the square root operation smooths the influence of adjustment stability and avoids interference from extreme values.

[0052] To achieve the above object, the present invention provides the following technical solution: A high-precision processing system for multi-specification copper tubes, including:

[0053] Data acquisition and enhancement module: Collect raw material parameters (grain size, impurities, dimensions), process timing data (cold drawing speed, temperature curve, cooling rate), and detection data (residual stress, defect coordinates) during the processing of multi-specification copper tubes. Improve the spatial matching accuracy through B-spline interpolation, and use a conditional adversarial generation network to generate virtual sample data covering extreme working conditions. The virtual sample data needs to satisfy the axial force balance and Hollomon hardening model constraints, and output an enhanced data set;

[0054] Simulation verification module: Based on the enhanced data set, simulate the dynamic changes of grain boundary migration and dislocation density through grain evolution-stress coupling, calculate the anisotropic yield strength, and generate a physical verification data set; Mark the coordinates of high-stress regions and correction coefficients, and output the physical verification data set and defect-stress mapping table;

[0055] Prediction Modeling Module: Input physical verification data and defect-stress mapping table, train the XGBoost regression model to predict the local stress distribution, combine the axial force balance verification to optimize the model parameters, and output the stress prediction model and the heat map of the stress transfer path;

[0056] Optimization and Adjustment Module: Input the stress prediction model and processing target constraints, identify the causal effects of parameters through the cross-modal causal model, use the NSGA-II algorithm to generate the Pareto optimal parameter set, combine the real-time stress data to dynamically adjust the cold drawing speed and temperature, and output the optimal parameter combination and the corresponding dynamic adjustment strategy.

[0057] Real-time Control Module: Based on the optimal parameters and real-time sensor data, dynamically adjust the process parameters through the fuzzy PID controller (such as triggering the gradient slow cooling mode when the cold drawing speed changes suddenly), and output real-time adjustment instructions, process logs, and processing quality reports.

[0058] Technical Effects and Advantages of the Present Invention:

[0059] (1) By constructing a multi-modal association system of raw materials - process - detection data, using B-spline interpolation to align the spatio-temporal dimension data, and combining the graph convolutional network (GAT+IGC) to explicitly represent the grain boundary migration (edge weight = grain boundary orientation difference + stress gradient) and the evolution of dislocation density (node hardness characteristics), the present invention realizes the cross-scale coupling modeling of microstructure - macro stress, obtains a more accurate stress prediction model, and solves the existing problems of residual stress concentration and dimensional tolerance.

[0060] (2) By establishing a digital twin-driven fuzzy PID closed loop, dynamically analyzing the coupling relationship between geometric specifications and stress field based on the stress prediction model (update frequency 10Hz), and generating the Pareto parameter set through hybrid agent optimization, the present invention realizes the optimization of the processing technology; when the cold drawing speed changes suddenly, the real-time optimization of process parameters is realized through the dynamic path strategy (such as switching the mode when the stress gradient exceeds the threshold), and the problems of prediction model distortion and response lag are solved. Description of the Drawings

[0061] Figure 1 It is a flowchart of the high-precision processing method for multi-specification copper tubes of the present invention. Detailed Embodiments

[0062] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be fully communicated to those skilled in the art.

[0063] Meanwhile, it should be understood that, for the convenience of description, the dimensions of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0064] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way a limitation on the present application, its application, or its use.

[0065] Technologies, methods, and devices known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, such technologies, methods, and devices should be regarded as part of the specification.

[0066] Example 1, referring to the flowchart of the high-precision processing method for multi-specification copper tubes as Figure 1 shown, the present invention provides a high-precision processing method for multi-specification copper tubes as Figure 1 shown, including:

[0067] Step 1: Multi-source data acquisition and physical enhancement generation: Collect raw material parameters, process timing data, and inspection data (actual residual stress data, such as X-ray diffraction residual stress distribution on the surface and cross-section, ultrasonic flaw detection defect coordinates) to obtain real sample data; Process through the B-spline interpolation algorithm and conditional adversarial generation network. Improve the spatial matching accuracy through the B-spline interpolation algorithm; Expand the data set through finite element simulation, and obtain virtual sample data through the conditional adversarial generation network. The virtual sample data and the aligned real sample data constitute an enhanced data set; The virtual sample data needs to satisfy the axial force balance and Hollomon hardening model constraints; The loss function of the conditional adversarial generation network includes adversarial loss, axial force balance term, and Hollomon hardening model constraints to ensure that the generated data conforms to mechanical laws;

[0068] Explanation: The raw material parameters include chemical composition (such as copper content, impurity element percentage, for example, Cu≥99.9%, Zn≤0.05%), initial grain size (measured by an optical microscope, unit is micron, typical range 10 - 50μm), initial hardness (measured by a Vickers hardness tester, unit HV, typical range 50 - 100HV). These raw material parameters affect the processing performance and stress distribution of copper tubes; The process timing data is collected by time series sensors (such as encoders, thermocouples) on the processing equipment, including cold drawing speed (unit m / min, recorded for each pass), processing temperature (unit °C, recorded for ambient and die temperatures), lubricant usage (unit mL / min), and tensile deformation amount, cooling rate curve; Use an ultrasonic flaw detector to obtain ultrasonic flaw detection defect coordinates, and record the three-dimensional coordinates and defect types of the defects;

[0069] Explanation: The B-spline interpolation algorithm is a 3rd-order B-spline, and the implementation process includes:

[0070] Calculate the centroid of two sets of coordinates and translate to make the centroids coincide;

[0071] Use the Iterative Closest Point (ICP) algorithm to optimize the rotation matrix with an error convergence threshold of 0.01 mm;

[0072] Project the interpolated defect coordinates onto the stress grid (grid resolution 1 mm × 1 mm × 1 mm) to ensure spatial consistency;

[0073] Step 2: Input the enhanced dataset, enhance the dataset through grain evolution-stress coupling simulation modeling and physical constraint-driven process parameter correction, output the physically verified dataset, and label the coordinates of high-stress regions and correction coefficients;

[0074] Step 3: Input the physically verified dataset and the defect-stress mapping table, and build a stress prediction model through the XGBoost regression model;

[0075] Step 4: Cross-modal causal enhanced hybrid surrogate optimization: Input the stress prediction model and processing target constraints; generate the optimal parameter combination that meets the processing target constraints based on the multi-objective optimization algorithm (NSGA-II), and dynamically adjust the process parameters in combination with real-time stress data;

[0076] Step 5: Digital twin-driven real-time control: Input the optimal parameter combination and real-time sensor data, dynamically adjust the process parameters through the combination of fuzzy logic and PID control, and output real-time process adjustment instructions, process adjustment log information, and processing quality reports (deformation distribution, stress uniformity index).

[0077] In a possible embodiment, in Step 5, establish a digital twin-driven fuzzy PID closed-loop control system, input the optimal parameter combination and real-time sensor data, dynamically analyze the coupling relationship between the geometric specifications and the stress field based on the stress prediction model; generate the Pareto parameter set through hybrid surrogate optimization to achieve the optimization of the processing technology; when the cold drawing speed changes suddenly, adopt a dynamic path strategy.

[0078] In the embodiments of the present invention, it needs to be further explained that the B-spline interpolation algorithm with a local curvature weight factor is used for three-dimensional interpolation of ultrasonic defect coordinates. When mapping the ultrasonic defect coordinates to the X-ray stress grid, the spatial offset error between defect positioning and the stress field is eliminated, and the spatial matching accuracy is improved; when using the conditional adversarial generation network to generate virtual sample data, the extreme working conditions of copper tubes are covered, such as the stress data under the extreme working conditions of copper tubes with a cold drawing speed of 1.0 - 3.0 m / s, a temperature of 450 - 600 °C, and a wall thickness less than 1 mm and a diameter-to-thickness ratio greater than 30.

[0079] In a possible embodiment, the conditional adversarial generation network can be replaced by a variational autoencoder, and an axial force balance term and a Hollomon hardening model constraint are added to the variational autoencoder loss function to ensure that the generated data conforms to the mechanical laws.

[0080] In the embodiments of the present invention, it needs to be further explained that an enhanced data set is used for grain evolution-stress coupling simulation modeling to quantify the dynamic influence of grain boundary migration on residual stress and generate a physical verification data set containing data on the influence of grain boundary migration;

[0081] The data on the influence of grain boundary migration includes the dynamic changes in grain boundary position, grain size, and dislocation density, as well as the corresponding residual stress distribution;

[0082] A cellular automaton model is used for simulation modeling. The state of the grain unit is defined as the orientation angle θ and the dislocation density ρ. The grain boundary migration rate is calculated based on the Moore neighborhood rule to dynamically update the grain orientation field; the grain boundary migration rate reflects the influence of grain dynamic evolution on residual stress.

[0083] According to the grain size d and the dislocation density ρ, the anisotropic yield strength is calculated by the following formula, and the simulated residual stress distribution is obtained based on the anisotropic yield strength:

[0084] ;

[0085] where is the matrix strength of the copper tube, G represents the shear modulus, k is the grain boundary strengthening coefficient calibrated through nanoindentation experiments, δ is the dislocation strengthening coefficient obtained based on the Taylor relationship of the dislocation density ρ; usually ; k, , G, and b (Burgers vector) are material constants; combining the contributions of dislocation hardening and grain refinement to strength helps simulate the microbehavior of the material during copper tube processing.

[0086] The specific steps are as follows: 30 measurement points are selected on the surface of the copper tube in a 1 cm × 1 cm area, the loading rate is 0.05 mN / s, the indentation depth is 200 nm, the measurement points cover the grain boundary and intragranular regions, the k value is calculated from the difference in hardness near the grain boundary and the intragranular hardness, and the average value is taken to obtain the grain boundary strengthening coefficient;

[0087] The generated simulated residual stress distribution is compared with the actual residual stress data (such as X-ray residual stress distribution) in the enhanced data set to verify the simulation accuracy; if differences are found, the abnormal parts in the enhanced data set are identified and corrected.

[0088] In a possible embodiment, the cellular automaton can be replaced by a phase field model, where phase field variables (such as grain orientation field) and evolution equations (such as Allen-Cahn equation) need to be defined, which are applicable to continuum simulation.

[0089] In the embodiment of the present invention, what needs to be further explained is the correction of process parameters driven by physical constraints: the circumferential tensile stress area ratio in the simulated residual stress distribution is calculated in real time. If it exceeds a preset value, for example, 40%, the cold drawing speed v and temperature T are adjusted by sequential quadratic programming optimization. The objective function is:

[0090] ;

[0091] where, is the circumferential tensile stress area ratio, representing the abnormal area ratio of the circumferential tensile stress of the copper tube; is the deviation of the cold drawing speed from the initial speed, is the deviation of the temperature from the initial temperature;

[0092] Based on the optimization results of the objective function, a defect-stress mapping table annotating the coordinates of high-stress areas and correction coefficients is generated, specifically:

[0093] The objective function reduces the circumferential tensile stress area ratio by optimizing the cold drawing speed and temperature; the areas in the simulated residual stress distribution that still exceed the threshold (the threshold is set to 20% in the embodiment of the present invention) after optimization are identified as high-stress areas, and their coordinates and corresponding speed and temperature correction coefficients are recorded to generate a defect-stress mapping table.

[0094] In a possible embodiment, sequential quadratic programming can be replaced by particle swarm optimization (PSO), and the fitness function is defined as the above objective function, with the population size ≥ 50 and the number of iterations ≥ 200.

[0095] Explanation: The objective function tries to minimize the circumferential tensile stress area ratio. At the same time, it tries to keep the changes in speed and temperature small, but the change in speed is penalized less than the change in temperature because its weight is 0.1 while the weight of temperature is 1; the purpose of the objective function is to find an optimal cold drawing speed and temperature that can reduce the tensile stress problem without causing too much change in parameters, especially the temperature change; it is more inclined to solve the problem of large circumferential tensile stress area ratio by adjusting the speed.

[0096] In the embodiment of the present invention, what needs to be further explained is that building a stress prediction model includes:

[0097] The copper tube geometry is transformed into a graph structure, where the nodes are discrete grid points, and the features include position coordinates, real-time hardness, and local cold drawing speed; the edges are defined as the connections between adjacent nodes, and the edge weights are calculated from the stress gradients and grain boundary orientation differences between adjacent nodes;

[0098] In a possible embodiment, the edge weight is ; where represents the circumferential stress gradient, is the grain orientation angle, used to describe the grain orientation in radians; represents the grain boundary misorientation, reflecting the difference in grain orientation in radians between adjacent grains; Explanation: The edge weight coefficient is set based on the contribution ratio of the grain boundary misorientation to stress transfer. For example, if the contribution ratio of the grain boundary misorientation to stress transfer is 30%-50%, then 0.5 is taken to balance the effects of the gradient and misorientation;

[0099] Implement cross-scale coupling modeling of grain boundary migration and dislocation density evolution by combining the graph convolutional network (GAT+IGC): Calculate the influence of grain boundary slip on the stress of adjacent nodes. The grain boundary misorientation and stress gradient in the edge weight are directly related to the grain boundary migration behavior, and the node hardness feature is related to the dislocation density evolution; Regularly update the graph structure to reflect the change in edge weight caused by grain coarsening; Use incremental graph convolution (IGC) to manage the computational efficiency, and use multi-head graph convolution (GAT) to calculate the influence of grain boundary slip on the stress of adjacent nodes;

[0100] Train the XGBoost regression model, with node features and edge weight as inputs, to predict the stress value at the node; Use the physical verification dataset to ensure axial force balance verification; During the training process, the loss function includes the mean square error and the axial force balance penalty term, and the hyperparameters of the stress prediction model are optimized by the Bayesian method;

[0101] Post-process the model prediction to ensure that the proportion of the circumferential tensile stress area does not exceed 20%, and use the region growing algorithm to correct the over-standard area.

[0102] In the embodiment of the present invention, it needs to be further explained that the training process of the stress prediction model includes:

[0103] Use the finite element simulation data verified by axial force balance for pre-training to ensure the axial force balance term where is the predetermined tolerance;

[0104] Use experimental data for fine-tuning. The loss function of the stress prediction model includes the mean square error and the axial force balance penalty term , is the adjustment parameter, and the hyperparameters are adjusted by Bayesian optimization;

[0105] Post-processing uses the region growing algorithm to ensure that the proportion of the circumferential tensile stress area does not exceed 20% and correct the over-standard area;

[0106] In the region growing algorithm, the threshold setting and iteration rule are specifically as follows:

[0107] Seed point selection: The grid point where the maximum circumferential tensile stress is located, with coordinates (x max , y max ) is used as the initial seed point. If there are multiple maximum value points, the geometric center point is selected;

[0108] Neighborhood expansion condition: Traverse the Moore neighborhood (8-neighborhood) of the seed point. If the circumferential tensile stress value σ θ > 0.9σ threshold (σ threshold is the preset threshold, for example, 200 MPa), then this node is included in the area to be corrected;

[0109] Iteration termination condition: When the number of newly added nodes is less than 1% of the total number of nodes (or the absolute quantity ≤ 5) in three consecutive iterations, the expansion is terminated to avoid overfitting;

[0110] Correction strategy: For the nodes in the over-standard area, adjust the cold drawing speed and temperature according to the correction coefficients marked in the defect-stress mapping table. Specifically:

[0111] If σ θ ∈[0.9σ threshold , σ threshold , the cold drawing speed is reduced by 5%;

[0112] If σ θ > σ threshold σ, the cold drawing speed is reduced by 10% and the temperature is increased by 20 °C to relieve local stress concentration.

[0113] Technical basis: The setting of the threshold 0.9σ threshold is based on the finite element simulation results and can cover 90% of the potential over-standard risk area;

[0114] The iteration termination condition is verified through engineering tests, balancing the correction efficiency and accuracy;

[0115] The correlation between the correction coefficient and the process parameters (speed, temperature) is calibrated through multi-objective optimization experiments, and the proportion of the circumferential tensile stress area does not exceed 20%.

[0116] In the embodiments of the present invention, it needs to be further explained that the process of obtaining the optimal parameter combination includes:

[0117] Construct a cross-modal causal model, input the prediction results of the stress prediction model and the processing target constraints (maximum deformation < 0.05 mm, circumferential stress, axial stress), and identify the causal effects of process parameters on the stress distribution through causal inference methods to generate a causal effect diagram;

[0118] Based on the causal effect diagram, set the goal to minimize the stress concentration and deformation, and use the NSGA-II algorithm to generate a Pareto optimal parameter set;

[0119] Combined with real-time stress data, the real-time stress concentration is calculated to trigger process parameter adjustments;

[0120] Output the optimal parameter combination and switching strategy.

[0121] Summary: Through multi-source data enhancement and microstructure evolution modeling, the problem of process parameters relying on experience and being difficult to cover extreme working conditions in traditional copper tube processing is solved; specifically, the following technical means are included: the B-spline interpolation algorithm with local curvature weight factor is used to eliminate defect positioning errors, and the conditional adversarial generation network is combined to generate virtual data covering extreme working conditions, and mechanical constraints are used to ensure that the generated data conforms to actual laws; based on cellular automata, the grain boundary migration and dislocation density evolution are dynamically simulated, and the anisotropic yield strength formula is used to quantify the grain refinement and dislocation strengthening effects. The process parameter optimization is triggered by the threshold of the circumferential tensile stress area ratio, and the cold drawing speed is adjusted preferentially to reduce the influence of temperature fluctuations, so as to achieve high-precision parameter optimization for thin-walled tubes (wall thickness less than 1 mm) and under extreme working conditions; the fuzzy PID controller is combined to adjust the cold drawing speed and temperature in real time, and the causal effect of process parameters on stress distribution is extracted based on the cross-modal causal model, and the Pareto optimal parameter set is generated by a multi-objective optimization algorithm.

[0122] Embodiment 2: The difference between the embodiment of the present invention and embodiment 1 is that the method further includes a stress prediction model performance verification step: calculating the stress prediction model adaptation index through the processing quality report and the process adjustment log information, and if the stress prediction model adaptation index is lower than the threshold, triggering an early warning and updating the stress prediction model; specifically including:

[0123] Extract control instruction data from process adjustment logs, analyze instruction delays and execution deviations, and eliminate processing quality data corresponding to abnormal control instructions;

[0124] Extract the standard deviation of deformation distribution and the average value of stress uniformity index from the processing quality report, and extract the average value of adjustment frequency and adjustment amplitude per unit time based on the process adjustment log after eliminating abnormal data;

[0125] The processing quality fluctuation factor reflecting the degree of fluctuation of processing quality is calculated; the process adjustment stability factor reflecting the stability and effectiveness of process adjustment is calculated, and the stress prediction model adaptability index is calculated through the nonlinear combination of the processing quality fluctuation factor and the process adjustment stability factor.

[0126] It needs to be further explained in the embodiment of the present invention that the stress prediction model adaptation index is obtained in the following manner:

[0127] Step S11: Calculate the processing quality fluctuation factor :

[0128] ;

[0129] Among them, represents the actual deformation of the copper tube, represents the standard deviation of the reference deformation (statistical value of historical data), represents the reference stress uniformity index, and S represents the actual stress uniformity index; an exponential function is introduced to enhance the sensitivity of S. When S is lower than , the fluctuation factor is significantly increased, more accurately reflecting the deterioration of the stress distribution;

[0130] Step S12, process adjustment stability factor :

[0131] ;

[0132] Among them, represents the actual adjustment frequency of the process, represents the reasonable adjustment frequency threshold, represents the actual adjustment amplitude, represents the upper limit of the allowable adjustment amplitude; an inverse proportional function is used to suppress the excessive adjustment amplitude, avoiding system oscillation caused by frequent large adjustments;

[0133] Step S13, stress prediction model adaptation index :

[0134] ;

[0135] Among them, represents the time decay coefficient, represents the current time, represents the last update time of the model; it is notified to add a time decay term to make the adaptation index of the long-unupdated model decrease rapidly, forcing periodic verification; the square root operation smooths the influence of adjustment stability, avoiding interference from extreme values.

[0136] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A high-precision processing method for copper tubes of multiple specifications, characterized in that: include: Collect raw material parameters, process timing data and test data during the processing of copper tubes of multiple specifications. The test data includes actual residual stress data and ultrasonic flaw detection defect coordinates to obtain real sample data. Through B-spline interpolation algorithm and conditional adversarial generative network processing, virtual sample data is obtained through conditional adversarial generative network. The real sample data and virtual sample data constitute an enhanced data set. The virtual sample data must meet the constraints of axial force balance and Hollomon hardening model. Input the enhanced data set, modify the enhanced data set through grain evolution-stress coupling simulation modeling and physical constraint-driven process parameter correction, output the physical verification data set, and mark the coordinates of the high stress area and the correction coefficient; quantify the dynamic effect of grain boundary migration on residual stress, and generate a physical verification data set containing grain boundary migration effect data; including the dynamic change of grain boundary position, grain size d and dislocation density, and the corresponding residual stress distribution; use the cellular automaton model for simulation modeling, define the grain unit state as orientation angle θ and dislocation density ρ, calculate the grain boundary migration rate based on the Moore neighborhood rule, and dynamically update the grain orientation field; the grain boundary migration rate reflects the influence of the dynamic evolution of the grain on the residual stress; calculate the anisotropic yield strength through the following formula, and obtain the simulated residual stress distribution based on the anisotropic yield strength: ； in, is the matrix strength of the copper tube, G represents the shear modulus, k is the grain boundary strengthening coefficient, which is calibrated by nanoindentation experiment, δ is the dislocation strengthening coefficient, which is obtained based on the Taylor relationship of dislocation density ρ, and b is the Burgers vector of the crystal; Compare the generated simulated residual stress distribution with the actual residual stress data in the enhanced data set to verify the simulation accuracy; if differences are found, identify and correct the abnormal parts in the enhanced data set; Input the physical verification data set and defect-stress mapping table, and build a stress prediction model through the XGBoost regression model; Input stress prediction model and processing target constraints; generate the optimal parameter combination that meets the processing target constraints based on multi-objective optimization algorithm, and dynamically adjust process parameters based on real-time stress data; Input the optimal parameter combination and real-time sensor data, combine fuzzy logic with PID control, dynamically adjust the process parameters, and output real-time process adjustment instructions, process adjustment log information and processing quality reports.

2. A high-precision processing method for multi-specification copper tubes according to claim 1, characterized in that: The B-spline interpolation algorithm with local curvature weight factor is used to perform three-dimensional interpolation of ultrasonic defect coordinates. When the ultrasonic defect coordinates are mapped to the X-ray stress grid, the spatial offset error between the defect location and the stress field is eliminated. When the conditional adversarial generative network is used to generate virtual sample data, the extreme working conditions of the copper pipe are covered.

3. The high-precision processing method for multi-specification copper tubes according to claim 2 is characterized in that: Process parameter correction driven by physical constraints: Real-time calculation of the proportion of circumferential tensile stress area in the simulated residual stress distribution. If it exceeds the preset value, the sequence quadratic programming is triggered to optimize and adjust the cold drawing speed v and temperature T. The objective function is: ； in, is the proportion of circumferential tensile stress area, indicating the proportion of abnormal area of ​​circumferential tensile stress of copper tube; is the deviation between the cold drawing speed and the initial speed, is the deviation of temperature from the initial temperature; The defect-stress mapping table with the coordinates of the high stress area and the correction coefficient marked based on the objective function optimization results is as follows: The objective function reduces the proportion of circumferential tensile stress area by optimizing the cold drawing speed and temperature; The areas in the simulated residual stress distribution after optimization that still exceed the threshold are identified as high-stress areas, and the coordinates and corresponding velocity and temperature correction coefficients are recorded to generate a defect-stress mapping table.

4. The high-precision processing method for multi-specification copper tubes according to claim 2 is characterized in that: Building a stress prediction model includes: The copper tube geometry is converted into a graph structure, where nodes are discrete grid points, and features include position coordinates, real-time hardness, and local cold drawing speed; edges are defined as connections between adjacent nodes, and edge weights are calculated based on adjacent node stress gradients and grain boundary orientation differences; The edge weight is ;in, represents the circumferential stress gradient, is the grain orientation angle, which is used to describe the grain orientation arc; It indicates the grain boundary orientation difference, reflecting the difference in the orientation curvature of adjacent grains; By combining graph convolutional networks, we can achieve cross-scale coupled modeling to characterize grain boundary migration and dislocation density evolution: calculate the effect of grain boundary slip on the stress of adjacent nodes, directly correlate the grain boundary misorientation in edge weights with stress gradients, and correlate node hardness characteristics with dislocation density evolution; The graph structure is updated regularly to reflect the changes in edge weights caused by grain coarsening; Train an XGBoost regression model to predict stress values ​​at nodes using node features and edge weights as input; use a physical verification dataset to ensure axial force balance verification; During the training process, the loss function includes the mean square error and the axial force balance penalty term, and the hyperparameters of the stress prediction model are optimized by the Bayesian method.

5. The high-precision processing method for multi-specification copper tubes according to claim 3 is characterized in that: The training process of the stress prediction model includes: Pre-training using finite element simulation data verified by axial force balance to ensure the axial force balance term ,in The axial force balance term represents all normal stresses acting on section A in the z-axis direction. The vector sum of the axial force balance term is zero, indicating that the object has no net external force in the axial direction and is in a state of static equilibrium. Fine-tuning using experimental data, the loss function of the stress prediction model includes mean square error and axial force balance penalty terms , To regulate the parameters, hyperparameters were adjusted through Bayesian optimization; The post-processing uses the region growing algorithm to correct the areas exceeding the standard.

6. The high-precision processing method for multi-specification copper tubes according to claim 1 is characterized in that: The process of obtaining the optimal parameter combination includes: Construct a cross-modal causal model, input the prediction results of the stress prediction model and the processing target constraints, identify the causal effect of process parameters on stress distribution through causal inference methods, and generate a causal effect diagram; Based on the cause-effect diagram, the goal is set to minimize stress concentration and deformation, and the NSGA-II algorithm is used to generate the Pareto optimal parameter set; Combined with real-time stress data, the real-time stress concentration is calculated to trigger process parameter adjustments; Output the optimal parameter combination and switching strategy.

7. The high-precision processing method for multi-specification copper tubes according to claim 1 is characterized in that: The method includes the following steps of verifying the performance of the stress prediction model: calculating the adaptation index of the stress prediction model through the processing quality report and the process adjustment log information; if the adaptation index of the stress prediction model is lower than the threshold, triggering an early warning and updating the stress prediction model.

8. The high-precision processing method for copper tubes of multiple specifications according to claim 7 is characterized in that: The stress prediction model adaptation index is obtained as follows: Step S11: Calculate the processing quality fluctuation factor : ； in, Indicates the actual deformation of the copper tube. represents the standard deviation of the reference deformation, represents the benchmark stress uniformity index, and S represents the actual stress uniformity index. The exponential function is introduced to enhance the sensitivity of S. When S is lower than The fluctuation factor is significantly improved, which more accurately reflects the deterioration of stress distribution; Step S12: Process adjustment stability factor : ； in, Indicates the actual adjustment frequency of the process, Indicates reasonable adjustment of the frequency threshold. Indicates the actual adjustment range. Indicates the upper limit of the adjustment range allowed; an inverse proportional function is used to suppress excessive adjustment range to avoid system shock caused by frequent and large adjustments; Step S13: Stress prediction model adaptation index : ； in, represents the time attenuation coefficient, Indicates the current time. Indicates the last update time of the model; notifies the addition of a time decay term to adapt the model that has not been updated for a long time to an accelerated exponential decline and enforces periodic verification; square root operations smooth out stability effects to avoid interference from extreme values.

9. A high-precision processing system for copper tubes of various specifications, applied to the method described in any one of claims 1 to 8, characterized in that: include: Data acquisition and enhancement module: collects raw material parameters, process timing data and inspection data of copper tubes of various specifications during processing, improves spatial matching accuracy through B-spline interpolation, and uses conditional adversarial generative networks to generate virtual sample data covering extreme working conditions. The virtual sample data must meet the constraints of axial force balance and Hollomon hardening model, and output enhanced data sets; Simulation verification module: Based on the enhanced data set, the dynamic change of grain boundary migration and dislocation density is simulated through grain evolution-stress coupling simulation, the anisotropic yield strength is calculated, and the physical verification data set is generated; the coordinates and correction coefficients of the high stress area are marked, and the physical verification data set and defect-stress mapping table are output; Predictive modeling module: input physical verification data and defect-stress mapping table, train XGBoost regression model to predict local stress distribution, combine axial force balance to verify and optimize model parameters, and output stress prediction model and stress transfer path thermal map; The optimization and adjustment module inputs the stress prediction model and processing target constraints, identifies the parameter causal effect through the cross-modal causal model, generates the Pareto optimal parameter set using the NSGA-II algorithm, dynamically adjusts the cold drawing speed and temperature based on the real-time stress data, and outputs the optimal parameter combination and the corresponding dynamic adjustment strategy; Real-time control module: Based on the optimal parameters and real-time sensor data, the process parameters are dynamically adjusted through the fuzzy PID controller, and real-time adjustment instructions, process logs and processing quality reports are output.

Images