Stamping part forming precision dynamic monitoring system based on digital twinning

Abstract

The invention discloses a stamping part forming precision dynamic monitoring system based on digital twinning, and particularly relates to the field of general monitoring and adjusting systems, and the stamping part forming precision dynamic monitoring system comprises a digital twinning modeling module, a twinning simulation prediction module, an intelligent function coupling decision module and a physical execution closed loop feedback module; the digital twinning modeling module collects stamping related data through a multi-source sensor, and constructs and dynamically updates a geometric, physical and behavior three-level digital twinning body after preprocessing; the twinborn body simulation prediction module is based on three-stage twinborn bodies and is combined with parameter initialization, double-source simulation, result fusion and threshold decision to realize pre-judgment of forming precision; the intelligent function coupling decision-making module generates a targeted regulation and control instruction through a basic regulation and control function and a dynamic coupling mechanism according to the precision out-of-tolerance signal and the physical data; and the physical execution closed-loop feedback module completes instruction execution, state perception and model correction through iterative loop, realizes dynamic monitoring of the forming precision of the stamping part, and improves the forming precision stability and the production efficiency of the stamping part.

Description

Technical Field

[0001] This invention relates to the field of general monitoring and control systems, and more specifically, to a dynamic monitoring system for the forming accuracy of stamped parts based on digital twins. Background Technology

[0002] With the increasing demands for dimensional accuracy and performance stability in industries such as automotive and electronics, digital twin technology and real-time monitoring technology are gradually merging, becoming a core means to achieve high-precision manufacturing status perception. Currently, digital twin monitoring systems mostly revolve around geometric modeling and simple physical simulation to build their basic framework. Analysis modules often rely on linear judgments of single parameters, and physical feedback mechanisms primarily use partial data display or simple comparisons. Their core logic is to simulate physical processes using digital models, combining fixed thresholds or human experience to judge the forming accuracy status, thus only achieving basic monitoring functions such as data acquisition and status display. Simultaneously, monitoring and analysis technologies are evolving from traditional single-parameter threshold judgments towards composite analysis combining data-driven approaches and physical mechanisms, attempting to improve the accuracy of precision status recognition through algorithm optimization. However, in practical applications, some shortcomings still exist, such as: 1. Single dimension of digital twins: Existing digital twin monitoring focuses on single-dimensional modeling, or only achieves geometric shape replication, or only conducts simple physical field simulation. It has not formed a progressive construction system of geometry, physics and behavior, and cannot fully reproduce the morphological characteristics, physical mechanism and dynamic behavior law of physical entities. As a result, the monitoring results cannot fully reflect the actual production status, and the guidance value for accuracy risk warning is limited.

[0003] 2. Insufficient adaptability of analysis logic: The monitoring and analysis module lacks a systematic multi-dimensional analysis function construction logic, and relies heavily on single parameters or fixed empirical formulas to judge the accuracy status. It does not quantify the independent impact of multiple factors such as stress, temperature, and material parameters on accuracy, making it difficult to adapt to the monitoring needs of multi-variable interaction in complex manufacturing scenarios, and easily leading to misjudgment of accuracy risks.

[0004] 3. Lack of multi-factor interaction analysis: The existing monitoring and decision-making mechanism has not established an effective multi-factor dynamic coupling logic. It mostly adopts a linear weighting method to integrate the influence of different parameters, which cannot simulate the nonlinear interaction relationship between multiple physical fields and multiple factors. This leads to a deviation between the accuracy risk identification results and the actual manufacturing status, and the rate of missed or false judgments is high.

[0005] 4. Monitoring closed loop not formed: The physical feedback and model optimization are disconnected. Feedback data is only used for local status display and does not feed back into the twin model parameter optimization and analysis logic iteration. The system has weak adaptability to operating condition fluctuations, is prone to monitoring accuracy drift, and cannot continuously and accurately identify accuracy risks. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, the present invention provides a dynamic monitoring system for the forming accuracy of stamped parts based on digital twins, which solves the problems mentioned in the background art through the following solutions.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a dynamic monitoring system for the forming accuracy of stamped parts based on digital twins, comprising, Digital twin modeling module: Collects multi-source data affecting stamping accuracy through preset sensors, outputs a standardized dataset after preprocessing the collected data, and constructs a three-level digital twin consisting of a geometric twin, a physical twin, and a behavioral twin based on the standardized dataset. At the same time, the twin parameters are dynamically updated according to real-time data. Twin simulation prediction module: Receives the three-level digital twin output by the data-driven digital twin modeling module, completes parameter initialization by combining initial process parameters, material parameters and boundary conditions, obtains mechanism simulation results and data-driven prediction results through dual-source simulation, obtains comprehensive accuracy prediction value by result fusion, and then determines whether to output accuracy out-of-tolerance warning signal through threshold decision. Intelligent Function Coupling Decision Module: Receives the accuracy deviation warning signal and associated physical data output by the twin simulation prediction module, constructs a multi-dimensional basic control function, quantifies the influence of multiple factors on stamping accuracy, and integrates nonlinear interaction relationships through a dynamic coupling mechanism to generate steady-state process control instructions that conform to the physical constraints of the equipment. Physical execution closed-loop feedback module: It collects multi-physics field state data of the physical system again, integrates the accuracy prediction value and the measured data and analyzes the state deviation. After identifying the core factors of the deviation, it generates specific control commands to drive the actuator to act. At the same time, it dynamically compensates for the lag of the actuator. It judges whether the control has converged through closed-loop iteration. If it has not converged, it returns to the data acquisition stage to repeat the control process, so as to realize the dynamic control of the forming accuracy of the stamped parts.

[0008] Preferably, the multi-source data includes: time-series data of stamping force of the target stamping equipment, denoted as... Collect slider displacement data, denoted as... And calculate the slider speed data, denoted as Collect the clearance data at three points on the die cutting edge, and record it as follows: The temperature field data of the mold was collected and recorded as follows: Principal component analysis was used to extract the top three principal components with a cumulative contribution rate ≥95%, and the dimensionality-reduced temperature characteristic parameters were calculated. , , Collect plate thickness distribution data and record it as follows: Collect actual form and position error data of stamped parts, and record them as follows: ; The physical 3D point cloud of the mold is obtained by scanning the surface of the mold with a 3D laser scanner, denoted as ; Tensile tests were performed on each batch of sheet metal using a universal testing machine to collect raw stress-strain curve data. Linear regression was used to fit the plastic stage curve, and the stress value corresponding to the yield point and the natural logarithm ratio of the curve slope were calculated, denoted as yield strength, respectively. Hardening index n.

[0009] Preferably, the geometric twin includes: acquiring surface geometric data of a physical entity through 3D scanning technology to obtain a point cloud model of the entity; preprocessing the acquired point cloud data; constructing a 3D geometric model of the entity based on the preprocessed point cloud through reverse modeling technology; defining the size constraints and relationships of geometric features using parametric modeling; and finally aligning and calibrating the constructed geometric model with the coordinate system of the physical entity.

[0010] Preferably, the physical twin includes: associating the acquired parameters with the corresponding components of the geometric model, constructing multiphysics field control equations based on the laws of conservation of momentum and energy, clarifying the boundaries and initial conditions, and using numerical discretization technology to realize physical field simulation, and verifying and correcting the model by combining physical experimental data.

[0011] Preferably, the behavioral twin includes: integrating real-time sensor data and historical operating data, extracting entity behavioral features after cleaning and standardization; then combining physical mechanisms and data-driven methods to construct a behavioral prediction model, which can output the entity's future performance changes and failure probability status; then driving the model to run through real-time data, comparing the actual behavior of the entity to optimize parameters, and achieving dynamic prediction.

[0012] Preferably, the twin simulation prediction module includes: using the digital twin output by the digital twin modeling module and the collected multi-source data as initialization parameters, and achieving high-precision prediction through dual-source simulation fusion: first, constructing a physical twin based on the thermo-coupling control equation, combining it with an elastoplastic constitutive model, and simulating the evolution of the physical field using the finite element method; then, training a behavioral twin based on historical data, and outputting a precision prediction value based on an LSTM neural network; fusing the dual-source results through a dynamic weighting algorithm, adjusting the weights according to the historical errors of both, and finally generating a comprehensive result.

[0013] Preferably, the construction of the multi-dimensional basic control function includes: constructing a function prototype through the analytical relationship between two dimensions, and combining the entity response calibration function coefficients under different process parameters, and training and optimizing the function boundary conditions through historical production data; finally, based on the data output by the preceding module and the collected multi-source data, constructing a stress-accuracy correlation function, a material-process adaptation function, and an error tracing function.

[0014] Preferably, the dynamic coupling mechanism includes: establishing a multi-function coupling model based on partial differential equations, embedding the constructed multi-dimensional basic control function as a source term into the equation, and quantifying the nonlinear interaction relationship between functions; using numerical discretization technology to spatiotemporally discretize the coupling model, solving for the steady-state control command, and introducing physical system constraints in the process; and then iteratively optimizing the diffusion coefficient and weight coefficient of the coupling model through historical control data and physical execution feedback error to obtain the steady-state control command.

[0015] Preferably, the physical execution closed-loop feedback module includes: collecting physical entity state data again through a deployed high-precision sensor network, converting the control instructions of the decision module into signals recognizable by the actuator, driving the servo motor and temperature control system to perform actions, eliminating the actuator hysteresis effect through a dynamic compensation algorithm; and finally, adjusting the control instructions and twin parameters based on the deviation between the actual response of the physical entity and the target state.

[0016] The technical effects and advantages of this invention are as follows: 1. Comprehensive Monitoring Infrastructure: Breaking through the limitations of single-dimensional modeling in existing digital twin monitoring, a three-level progressive twin system of geometry, physics, and behavior is constructed. Through multi-source data preprocessing and dynamic update mechanisms, the morphological characteristics, physical mechanisms, and dynamic behavior patterns of physical entities are fully reproduced, avoiding the lack of monitoring information caused by merely replicating the form or simply simulating the physical environment. This enables the simulation analysis results to fully match the actual production status, significantly improving the accuracy of stamping production precision monitoring and risk prediction capabilities, and solving the problem of limited simulation guidance value in traditional monitoring.

[0017] 2. Highly Adaptable Monitoring and Analysis: The intelligent function coupling analysis module abandons the reliance on single parameters or fixed empirical formulas and establishes a multi-dimensional basic monitoring and analysis function system. By quantifying the independent influence of multiple factors such as stress, temperature, material parameters, and process parameters on stamping accuracy, it clarifies the weight and logic of each factor, forming a systematic monitoring and analysis logic. This can effectively adapt to the monitoring needs of multi-variable interactions in complex manufacturing scenarios, avoid the problem of insufficient adaptability caused by a single monitoring logic, and improve the scientific nature of accuracy risk identification in complex scenarios.

[0018] 3. Dynamic Coupled Monitoring and Analysis: To address the shortcomings of existing linear weighted fusion of multiple factors in monitoring, this solution employs partial differential equations to construct a multi-function dynamic coupling mechanism. Using key monitoring variables as the core, transient coupling equations are constructed and solved discretized using specialized algorithms. This accurately simulates the nonlinear interactions between multiple physical fields and factors, replacing the simple linear weighting method. This ensures that the generated accuracy-influencing factor analysis results closely match the actual manufacturing state, significantly reducing the bias in accuracy risk identification and resolving the discrepancy between monitoring conclusions and actual needs caused by linear fusion.

[0019] 4. High Adaptive Monitoring Capability: The physical state closed-loop feedback and model optimization module constructs a fully closed-loop system for data acquisition, simulation prediction, deviation analysis, state feedback, and model optimization. The feedback data is not only used to correct local monitoring results, but also feeds back into the optimization of digital twin model parameters and the iterative monitoring and analysis logic, realizing the linkage optimization of each link of the system, effectively responding to fluctuations in operating conditions, avoiding monitoring accuracy drift during long-term operation, greatly enhancing the system's adaptive capability to changes in operating conditions, solving the problems of weak adaptability and easy accuracy drift in existing monitoring schemes, and ensuring long-term stable monitoring effect of stamping part forming accuracy. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall structure of the present invention.

[0021] Figure 2 This is a schematic diagram of the twin simulation prediction module of the present invention.

[0022] Figure 3 This is a schematic diagram of the physical execution closed-loop feedback module of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] refer to Figures 1-3 The illustrated digital twin-based dynamic monitoring system for the forming accuracy of stamped parts includes: Digital twin modeling module: Collects multi-source data affecting stamping accuracy through preset sensors, outputs a standardized dataset after preprocessing the collected data, and constructs a three-level digital twin consisting of a geometric twin, a physical twin, and a behavioral twin based on the standardized dataset. At the same time, the twin parameters are dynamically updated according to real-time data. Twin simulation prediction module: Receives the three-level digital twin output by the data-driven digital twin modeling module, completes parameter initialization by combining initial process parameters, material parameters and boundary conditions, obtains mechanism simulation results and data-driven prediction results through dual-source simulation, obtains comprehensive accuracy prediction value by result fusion, and then determines whether to output accuracy out-of-tolerance warning signal through threshold decision. Intelligent Function Coupling Decision Module: Receives the accuracy deviation warning signal and associated physical data output by the twin simulation prediction module, constructs a multi-dimensional basic control function, quantifies the influence of multiple factors on stamping accuracy, and integrates nonlinear interaction relationships through a dynamic coupling mechanism to generate steady-state process control instructions that conform to the physical constraints of the equipment. Physical execution closed-loop feedback module: It collects multi-physics field state data of the physical system again, integrates the accuracy prediction value and the measured data and analyzes the state deviation. After identifying the core factors of the deviation, it generates specific control commands to drive the actuator to act. At the same time, it dynamically compensates for the lag of the actuator. It judges whether the control has converged through closed-loop iteration. If it has not converged, it returns to the data acquisition stage to repeat the control process, so as to realize the dynamic control of the forming accuracy of the stamped parts.

[0025] The digital twin modeling module includes the acquisition of multi-source data, which includes: The time-series data of the stamping force of the target stamping equipment are collected using a piezoelectric sensor, and denoted as follows: ; The slider displacement data was acquired using a laser interferometer deployed alongside the slider's motion trajectory, and denoted as... The slider velocity data was calculated using the central difference algorithm and denoted as . ; By using eddy current sensors embedded at key locations on the mating surfaces of the punch and die, three-point clearance data of the die cutting edge were collected and denoted as follows. ; The temperature field data of the mold was collected by a 32-point fiber optic grating sensor array deployed at the concave mold cavity and the cutting edge, denoted as . Principal component analysis was used to extract the top three principal components with a cumulative contribution rate ≥95%, and the dimensionality-reduced temperature characteristic parameters were calculated. , , ; The thickness distribution data of the sheet metal was collected using an X-ray thickness gauge at the entrance of the stamping line, and denoted as . ; The actual form and position error data of the stamped parts are collected by a 3D laser scanner at the exit of the stamping line, and are denoted as follows: ; The physical 3D point cloud of the mold is obtained by scanning the surface of the mold with a 3D laser scanner, denoted as Point cloud density ≥100 points / mm², covering key areas of punch and die cutting edges, fillets, and die base; Tensile tests were performed on each batch of sheet metal using a universal testing machine to collect raw stress-strain curve data. Linear regression was used to fit the plastic stage curve, and the stress value corresponding to the yield point and the natural logarithm ratio of the curve slope were calculated, denoted as yield strength, respectively. Hardening index n; It should be further explained that x and y represent two-dimensional spatial coordinates in a two-dimensional Cartesian coordinate system, which are used to accurately locate the specific position on the physical object and ensure that the data corresponds one-to-one with the actual spatial position; t represents time, keeping the data in the same time dimension. It should be further explained that multi-source data preprocessing operations include: Dynamic data noise reduction: Addressing the impact force during data acquisition. and temperature field The data was decomposed into three levels using the db4 wavelet basis. Mechanical vibration interference was removed using an adaptive thresholding method, and a stationary signal was reconstructed. The threshold value was then set. , The standard deviation of noise; Spatiotemporal alignment processing: All sensors are uniformly connected to an industrial real-time clock, with a control timestamp accuracy of 1 microsecond. Using the displacement sampling time of the laser interferometer as a reference, the time axes of all dynamic parameters are aligned using linear interpolation (synchronization error ≤ 5ms). Spatial data position matching is achieved through coordinate system transformation, with the center of the upper surface of the die as the origin (0, 0, 0), the x-axis along the length of the die, the y-axis along the width, and the z-axis perpendicular to the die surface. The software will... , The point cloud coordinates are transformed to the mold coordinate system with a transformation error of ≤0.005mm; Data standardization: All directly collected and indirectly calculated parameters are mapped to the [0, 1] interval for normalization, generating a structured dataset. .

[0026] The construction of a three-level digital twin is completed through a closed loop of four steps: data input, model building, accuracy verification, and dynamic updating, to build a geometric twin, a physical twin, and a behavioral twin. Geometric twin construction: First, the actual shape point cloud of the physical mold is obtained through 3D scanning and calibrated with the design CAD model using the ICP algorithm to eliminate errors. Then, the real-time collected dynamic parameters of mold gap and sheet thickness distribution are integrated into the calibrated model to generate an initial geometric twin. Finally, based on mold wear and sheet batch fluctuation data, the gap and sheet morphology of the twin are periodically updated to ensure long-term morphological consistency. The specific steps are as follows: Minimize physical point cloud Point cloud of CAD model The geometric deviation is addressed by the Iterative Closest Point (ICP) algorithm, which optimizes the rotation matrix. With translation vector The mathematical function to minimize the sum of squared Euclidean distances between two point clouds is: ; Let be the rotation matrix about the x, y, and z axes. is the rotation angle, and is the translation vector, representing the positional offset of the CAD model relative to the scanned point cloud; The calibration process is as follows: Coarse alignment: Based on the reflective markers, ... and Deviation controlled within 1mm; fine iteration: calculation Each point in Find the closest point in the solution. and ,renew Position; stop when the deviation change between two iterations is ≤0.001mm or the number of iterations reaches 50, so that the final geometric deviation is ≤0.1mm, which meets the shape accuracy requirements of the stamping die; It should be further explained that the dynamic integration of this geometric twin is manifested in the integration of mold gaps and the integration of sheet thickness distribution; Its mold clearance integration logic is as follows: real-time clearance values ​​are obtained through Python scripts. Import a geometric twin and dynamically adjust the mating dimensions of the punch and die; when the measured clearance... At that time, among them For design gaps, To account for wear increments, the cutting edge size of the punch digital model is automatically reduced. To maintain the accuracy of the gap simulation, its update frequency is synchronized with the sensor sampling frequency; The integrated logic for the thickness distribution of the sheet material is as follows: In the CAD sheet material model, the thickness is parameterized and assigned... Mapping to each mesh cell generates a digital model of a plate with non-uniform thickness; avoiding the stress distribution simulation deviation caused by traditional uniform thickness models; Calculate the average wear of the die clearance after every 50 stampings are completed. If the average wear is greater than 0.005 mm, a wear compensation factor is added to the geometric twin to offset the die cutting edge outward along the z-axis. The punch cutting edge shifts inward. This ensures that the gap simulation matches the physical wear condition; Physical twin construction: First, the calibrated geometric twin is imported as the geometric basis; then, based on the material tensile test data, the Johnson-Cook constitutive equation is calibrated to describe the material's plastic behavior; next, the mesh, boundary conditions, and contact relationships are set in Abaqus to construct a thermo-mechanical coupling simulation model; finally, the model is solved to output virtual forming error and stress field results, which are compared and optimized with physical measured data. The specific function of its Johnson-Cook constitutive equation is: ;in, Let A be the equivalent plastic stress of the material, B be the yield strength at the reference temperature and strain rate, C be the strain rate sensitivity coefficient, and m be the thermal softening index. For equivalent plastic strain, To normalize the strain rate, Normalized temperature Displace the slider As the slider motion loading curve, , , The temperature field is inverted and used as the initial temperature boundary of the mold, outputting the stress field. strain field Virtual forming error This error is based on the deviation between the extracted key feature points and the CAD model; Behavioral twin: First, input features and output labels are selected from the collected data to construct a training dataset; then, an LSTM neural network structure is designed; next, the loss function is optimized through gradient descent, and the model is trained until the validation set error reaches the target; finally, real-time process parameters are input into the model to output the predicted accuracy value, and the model weights are continuously updated based on the physical feedback error to ensure consistency of the pattern. Construct a sample dataset, inputting seven dimensions of collected data: stamping peak value. Average speed of the slider Dimensional reduction characteristics of mold temperature field , , Yield strength Hardening index n, output label ; The specific design steps for the LSTM model structure are as follows: Input layer: Input data from 7 dimensions, take the average of the process parameters from the first 10 sampling points to reflect the time series trend; LSTM hidden layer 1: 64 neurons, using ReLU activation function, Dropout=0.2, randomly discarding 20% ​​of neurons to prevent overfitting, extracting first-order nonlinear temporal correlations of input features; LSTM hidden layer 2: 64 neurons, using ReLU activation function, Dropout=0.2, to extract higher-order non-linear correlations of features; Fully connected layer: 32 neurons, using the ReLU activation function, integrating the feature vectors output from the hidden layer; Output layer: One neuron, no activation function, outputs standardized accuracy predictions. ; Its loss function is: Where W represents all trainable weight matrices of the model, and N is the sample size. Let i be the model prediction value for the i-th sample. The actual precision error of the i-th sample; It should be further explained that the optimization mechanism of this module uses the Adam optimizer, with an initial learning rate of 0.001 and a learning rate decay strategy: it decays to 0.9 times the original value every 20 rounds; the training rounds are 100, and an early stopping strategy is adopted - if the validation set loss does not decrease for 5 consecutive rounds, then training is stopped. Calculate the prediction error after every 30 stampings are completed. average If the average prediction error is greater than 0.005 mm, an update will be initiated. Twin simulation prediction module: Through a four-step process of parameter initialization, dual-source simulation, result fusion, and threshold decision, the accuracy prediction of the three-level digital twin based on the digital twin modeling module is achieved. The parameters are initialized as input parameters: the digital twin output by the digital twin modeling module, the current process initial parameters, the material parameters for this batch, and the boundary conditions; Initial process parameters: punching force Slider speed Initial temperature of mold Holding time ; Material parameters for this batch: yield strength, hardening index, and plate thickness distribution; Boundary conditions: The spatial position of the mold in the simulation is constrained to match the physical system, ensuring that the boundaries of the virtual simulation match the actual equipment, based on the collected slider displacement. The preset slider motion trajectory describes the movement path of the slider during the stamping process; Dual-source simulation: Based on physical twins and parameter initialization, the thermo-mechanical coupling control equations are solved to simulate the entire process of material deformation from elastic deformation to plastic forming. The specific mathematical functions are as follows: Dynamic conservation equation: , where Pa is the stress tensor, which describes the stress state inside the material, and f is the volume force (N / m³). Let be the material density (kg / m³), and u be the displacement vector (m). Energy conservation equation: ,in Specific heat capacity is the amount of heat required to raise the temperature of a unit mass of material by 1°C; k is thermal conductivity, obtained based on the analysis of the material's thermal conductivity. The heat generated by plastic deformation is the heat source for mechanical dissipation; T is the temperature field. Based on the constitutive equation, the virtual forming error is output. , representing the form and position error obtained from mechanism simulation, and the maximum stress. (MPa): The maximum stress the material withstands during the forming process, and the peak temperature. (°C): The highest temperature during the forming process; Based on the LSTM neural network model, the forming error of the data-driven prediction is obtained. ; Result fusion: Weights are calculated based on the historical errors of the two prediction methods to generate a fused prediction function. ,in The historical average error of physical twins. The historical average error of the behavioral twin. This is a comprehensive accuracy prediction value; Threshold decision: Set a precision threshold: ;like If the output accuracy is normal, the output accuracy error warning signal will be output, and the program will proceed to the next module.

[0027] Intelligent Function Coupling Decision Module: Based on the accuracy deviation warning signal and associated physical data output by the twin simulation prediction module, it quantifies the influence of multiple factors through a basic control function and integrates nonlinear interactions through a dynamic coupling mechanism, ultimately generating targeted process control commands. This provides a precise operational basis for the next module. A detailed analysis follows: Receive the comprehensive accuracy prediction value output by the preceding module Maximum stress, peak temperature Die clearance G, material parameters ( Current process parameters () This process creates a structured dataset and removes outlier data. Construct a stress-accuracy correlation function: ,in The maximum material stress obtained from physical twin simulation. For reference temperature, G represents the material's melting point, and G represents the current mold clearance. Design a reference clearance for the mold. These are the stress term weighting coefficient, temperature term weighting coefficient, gap term weighting coefficient, and stress nonlinearity coefficient, with values ​​of 0.6, 0.3, 0.2, and 1.3, respectively. The stress term weighting coefficient is jointly calibrated through dual-source simulation verification of the twin simulation prediction module and training with historical production data. The goal is to minimize the deviation between the virtual forming error of the twin simulation prediction module and the measured form and position error of the physical execution closed-loop feedback module, and iterative optimization is achieved using the gradient descent algorithm. The stress nonlinear coefficient is obtained by fitting experimental data of various commonly used stamping sheets. This was done by fitting the data of 20 different sheet materials. Curve determined, This represents the measured form and position error; The weighting coefficient of the clearance term was determined through mold wear experiments related to the geometric twin. The weighting coefficients for the temperature term were determined through temperature gradient experiments designed using the digital twin modeling module. In this function, the stress ratio term quantifies the contribution of plastic deformation error, the exponential temperature term quantifies the contribution of hot deformation error, and the linear gap term quantifies the contribution of geometric deviation error. The three terms are added together to output the comprehensive accuracy influence value. Constructing material and process adaptation functions: ,in Based on the yield strength of the reference material, Here, is the coefficient of variation of the plate thickness distribution, and v is the current slider speed. As the reference slider speed, These are the yield strength adaptation coefficient, hardening index adaptation coefficient, thickness uniformity adaptation coefficient, and speed adaptation coefficient, which are obtained through training and optimization using historical production data. Specifically, gradient descent algorithm or regression analysis is used to minimize prediction errors based on forming accuracy data under different material and process parameters. The coefficient values ​​represent the contribution of each parameter: yield strength has the greatest impact with a weight of 0.45, followed by hardening index with a weight of 0.3, while thickness uniformity and slider speed have smaller weights of 0.15 and 0.1, respectively. Multiplying cov(H) by 2 and adding 1 is a data standardization process designed to amplify the impact of variation and ensure that this term is positive, avoiding interference from negative values. Construct the error source function: Each error component is solved using multiple linear regression: , , ,in For force error, For temperature error, Let a, b, and c represent the gap error, and a, b, and c be the weights of force error, temperature error, and gap error, respectively. Based on historical data training and closed-loop validation, the behavioral twin integrates 1000 sets of historical data. The weights are calculated using the standardized coefficient method of multiple linear regression. The force error variance contributes 70%, with a=0.7; the temperature error contributes 20%, with b=0.2; and the gap error contributes 10%, with c=0.1. Output error components. ; Dynamic coupling: The fundamental functions—stress-accuracy correlation function, material-process adaptation function, and error source function—describe the risk level, material requirements, and error sources, respectively. Nonlinear coupling needs to be achieved through partial differential equations (PDEs) to control the variables. Based on this, a transient coupling equation is constructed, the mathematical function of which is: ,in These represent the adjustment amounts for punching force, temperature, and holding time, respectively, and D is the diffusion matrix. Sensitivity of reaction control parameters These are the sensitivity coefficients for adjusting the punching pressure, temperature, and holding time, respectively, with values ​​of 0.9, 0.6, and 0.7. S is the source term, a nonlinear coupling term of the fundamental function, which directly determines the initial amplitude of the control command; its mathematical function is: The coefficients of the source term S are calibrated through iterative optimization of the physical execution closed-loop feedback module. The specific process is as follows: Initial value: Based on the dual-source simulation results, the coefficient range is initially set; every 10 closed-loop controls are completed, the coefficient is adjusted based on the deviation between the control command and the actual error correction amount—if the deviation > 0.005mm, the coefficient is corrected by 0.1 times the deviation rate; if the deviation ≤ 0.005mm, the coefficient remains unchanged; Final value: After 500 closed-loop iterations, the coefficients stabilize at 0.2, 0.3, and 0.1. This range ensures that the control amplitude output by the source term S matches the execution capability of the equipment; the finite element method is used to spatially discretize the PDE, and the explicit Euler method is used to time discretize the PDE. When the difference in the control vector between adjacent iteration steps is less than the threshold value... When the iteration stops, a steady-state control command is obtained. ,Right now: k is the sequence number; examine If the device is within physical constraints, and exceeds those constraints, the boundary threshold is used as the final command; the final output is the steady-state control command. .

[0028] Physical execution closed-loop feedback module: based on an iterative loop from perception to decision-making to execution and finally back to perception; The multiphysics state data of the physical system are acquired again, including: punching force, slider displacement, die clearance, die temperature field, sheet thickness distribution, and actual form and position errors; and the predicted values ​​and measured values ​​are fused together. The mathematical function is as follows: , ,in Let be the predicted value at time t, A be the state transition matrix, and B be the control input matrix. Let be the control variable at time t-1, and be the optimal value at time t. Kalman gain: , Let H be the actual measured value at time t, and let H be the observation matrix, which describes the mapping from the true state to the measured value. R is the prediction error covariance, and R is the observation noise covariance. The quantified physical state assessment results and state deviation signals are output through signal processing and physical state analysis. The specific mathematical function is as follows: ,That These represent the temperature field deviation, displacement field deviation, and stress field deviation, respectively, and their values ​​are the differences between the measured physical quantities and the standard quantities. This is a comprehensive deviation index; Sensitivity analysis is then used to identify the core factors causing the deviation. Based on the attribution results, a suitable control algorithm is selected to generate specific control commands, while simultaneously correcting the parameters of the digital twin model. The sensitivity analysis function is as follows: The PID control instruction generation function is: ;in These are the proportional coefficient, integral coefficient, and differential coefficient, respectively. Their values ​​are determined by first finding the parameters at the critical oscillation of the system, and then calculating them according to empirical formulas. There are no fixed values. The parameters that affect the deviation, This is the sensitivity coefficient. For control instructions; Control command execution and physical state correction: Based on the control command, the command signal is converted and then driven to drive the actuator, and the actuator hysteresis is dynamically compensated and eliminated; Its command signal conversion: converting control commands into signals that the device can recognize: Punch force adjustment: The punch force command is converted into servo motor current. Based on the magnitude of the punch force command and the force-current curve, the magnitude of the servo motor current is adjusted to adjust the punch force. Temperature compensation: Converts temperature commands into heating rod power, and adjusts the heating rod power and temperature based on the magnitude of the temperature command; Holding time: Adjust the original holding time based on the holding time command; Its actuator drive is as follows: punching force: the servo motor adjusts the slider position to change the force output; temperature: the heating rod or cooling water circuit adjusts the power to control the mold temperature; holding time: the timer adjusts the dwell time of the slider at the lower stop point. Its dynamic compensation function is: ; This refers to the actual output value of the temperature actuator or the impact actuator. This is the theoretical output value. This is a time constant, with a value of 3s for temperature, 0.1s for impact force, and 3s for holding time; Closed-loop iteration: Set a convergence criterion, define a deviation threshold and a rate of change threshold, with values ​​of 1 and 5% respectively; then perform a convergence judgment, compare the current comprehensive deviation with the threshold criterion. If the comprehensive deviation and the rate of change do not exceed the threshold, it means that convergence has been achieved and the control is effective. Otherwise, it means that convergence has not been achieved and the data collection in the digital twin modeling module is returned. It can be iterated up to 5 times. Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A stamping forming precision dynamic monitoring system based on digital twinning, characterized in that, Comprise: Digital twin modeling module: collect multi-source data affecting stamping precision through preset sensors, output standardized data set after preprocessing the collected data, and construct three-level digital twin consisting of geometric twin, physical twin and behavior twin based on the standardized data set, while dynamically updating twin parameters according to real-time data; Twin simulation prediction module: receive three-level digital twin output by the data-driven digital twin modeling module, complete parameter initialization combined with process initial parameters, material parameters and boundary conditions, obtain mechanism simulation results and data-driven prediction results through double-source simulation, get comprehensive precision prediction value through result fusion, and judge whether to output precision out-of-tolerance warning signal through threshold decision; Intelligent function coupling decision module: receive precision out-of-tolerance warning signal and associated physical data output by the twin simulation prediction module, construct multi-dimensional basic regulation and control function, quantify the influence of multi-factors on stamping precision, fuse nonlinear interaction relationship through dynamic coupling mechanism, and generate stable process regulation and control instruction conforming to device physical constraints; Physical execution closed-loop feedback module: collect multi-physical field state data of physical system again, fuse precision prediction value and measured data and analyze state deviation, identify core factors of deviation, generate specific regulation and control instruction to drive actuator action, dynamically compensate actuator hysteresis, judge whether regulation and control converges through closed-loop iteration, and if not, return to data collection link to repeat regulation and control process, realize dynamic regulation and control of stamping part forming precision.

2. The stamping forming precision dynamic monitoring system based on digital twinning according to claim 1, characterized in that: The multi-source data comprises: Collect the stamping force time series data of the target stamping equipment, denoted as ; Collect the slider displacement data, denoted as , and calculate the slider speed data, denoted as ; Collect the 3-point gap data of the die concave die edge, denoted as ; Collect the die temperature field data, denoted as , and use principal component analysis to extract the first three principal components with cumulative contribution rate ≥ 95%, and calculate the reduced temperature characteristic parameters: 、 、 ; Collect the plate thickness distribution data, denoted as ; Collect the actual shape and position error data of the stamping part, denoted as ; The physical mold three-dimensional point cloud is obtained by scanning the mold surface by a three-dimensional laser scanner, denoted as ; The stress-strain curve original data is collected by the tensile test of each batch of plate by the universal material testing machine, the linear regression method is used to fit the plastic stage curve, and the stress value corresponding to the yield point and the natural logarithm ratio of the curve slope are calculated, which are respectively recorded as yield strength , hardening index n.

3. The stamping forming precision dynamic monitoring system based on digital twinning according to claim 1, characterized in that: The geometric twin comprises: Surface geometry data of physical entity is collected through three-dimensional scanning technology, point cloud model of the entity is obtained; the collected point cloud data is preprocessed, and based on the preprocessed point cloud, three-dimensional geometric model of the entity is constructed through reverse modeling technology, parameterized modeling method is used to define size constraints and correlation of geometric features, and finally the constructed geometric model is aligned and calibrated with the coordinate system of the physical entity.

4. The stamping forming precision dynamic monitoring system based on digital twinning of claim 1, wherein: The physical twin comprises: based on the parameters associated to the corresponding components of the geometric model, multi-physical field control equation is constructed based on the law of conservation of momentum and energy, boundary and initial conditions are determined, and numerical dispersion technology is used to realize physical field simulation, and the model is verified and corrected combined with entity experimental data.

5. The stamping forming precision dynamic monitoring system based on digital twinning according to claim 1, characterized in that: The behavior twin comprises: Integrate real-time sensor data and historical operation data, extract entity behavior characteristics after cleaning and standardization; then combine physical mechanism and data-driven method to construct behavior prediction model, which can output entity future performance change and fault probability state, and realize dynamic prediction by driving model running with real-time data, comparing entity actual behavior to optimize parameters, and realizing dynamic prediction.

6. The stamping forming precision dynamic monitoring system based on digital twinning according to claim 1, characterized in that: The twin simulation prediction module comprises: The digital twin output by the digital twin modeling module and the collected multi-source data are taken as initialization parameters, and high-precision prediction is realized through double-source simulation fusion: first, a physical twin is constructed based on the thermal coupling control equation, combined with an elastic-plastic constitutive model, and the evolution of the physical field is simulated through the finite element method; then, the behavior twin is trained based on historical data, and an LSTM neural network is used to output the precision prediction value; the double-source results are fused through a dynamic weight algorithm, and the weight is adjusted according to the historical errors of the two, and finally the comprehensive result is generated.

7. The stamping forming precision dynamic monitoring system based on digital twinning according to claim 1, characterized in that: The multi-dimensional basic regulation function is constructed, including: The function prototype is constructed through the analytical relationship between the two dimensions, and the entity response calibration function coefficient under different process parameters is combined to train and optimize the function boundary conditions through historical production data; finally, based on the data output by the previous module and the collected multi-source data, the stress and precision correlation function, the material and process adaptation function, and the error traceability function are constructed.

8. The stamping forming precision dynamic monitoring system based on digital twinning of claim 1, wherein: The dynamic coupling mechanism includes: A multi-function coupling model is established based on partial differential equations, the multi-dimensional basic regulation function constructed is embedded as a source term into the equation, and the nonlinear interaction relationship between functions is quantified; numerical discretization technology is used to discretize the coupling model in space and time, and the steady-state regulation instruction is solved, and physical system constraints are introduced in the process; then, the diffusion coefficient and weight coefficient of the coupling model are iteratively optimized through historical regulation data and physical execution feedback error, and the steady-state regulation instruction is obtained.

9. The stamping forming precision dynamic monitoring system based on digital twinning of claim 1, wherein: The physical execution closed-loop feedback module includes: The multi-physical field state data of the physical system are acquired again, including stamping force, slider displacement, die gap, die temperature field, plate thickness distribution, and actual shape and position error, and the prediction value and measured value of the twin simulation prediction module are fused, and then the regulation instruction of the decision module is converted into a signal recognizable by the actuator to drive the servo motor and temperature control system to execute actions, and the actuator hysteresis effect is eliminated through a dynamic compensation algorithm; finally, according to the deviation between the actual response of the physical entity and the target state, the regulation instruction and the twin parameters are adjusted and fed back.

Images