Closed loop stamping process based on global pressure perception and real-time dynamic compensation

Abstract

The present application relates to the technical field of automobile parts stamping forming, and particularly relates to a closed-loop stamping forming process based on global pressure sensing and real-time dynamic compensation. The process method realizes real-time and high-resolution monitoring of the pressure distribution on the contact surface between the die and the plate in the whole forming cycle, so that the originally black-box forming process becomes transparent and visible. The operator can master the stress state of each microscopic area in real time, and provides accurate data basis for subsequent dynamic intervention.

Description

Technical Field

[0001] This invention relates to the field of stamping technology for automotive parts, and particularly to a closed-loop stamping process based on full-domain pressure sensing and real-time dynamic compensation. Background Technology

[0002] Traditional stamping forming process is a typical open-loop, pre-defined process: placing sheet metal → upper die descending for forming → upper die ascending → ejection and demolding. This process has inherent defects: its preset, fixed pressure and stroke parameters cannot respond to real-time operating conditions such as fluctuations in material properties and changes in friction conditions, making the forming process like a black box. Part quality (such as springback, wrinkling, and cracking) is only discovered after demolding, relying on post-demolding mold repair or parameter adjustment, resulting in low efficiency and poor consistency.

[0003] Specifically, the existing technology mainly has the following technical problems:

[0004] 1. In traditional stamping processes, the pressure distribution on the contact surface between the die and the sheet metal cannot be monitored in real time. Operators can only indirectly judge whether the process is normal based on the quality of the final product. Once a defect occurs, it is impossible to accurately pinpoint the time and location of the problem, and it is also difficult to trace the cause.

[0005] 2. Production disturbances such as batch differences in sheet metal (e.g., fluctuations in yield strength, thickness tolerance), changes in lubrication conditions, and drift in mold temperature can all cause the actual molding process to deviate from the preset parameters. Due to the lack of a real-time feedback mechanism, these disturbances cannot be compensated for during the molding process, resulting in fluctuations in product quality.

[0006] 3. In traditional processes, defects in parts (such as cracking, wrinkling, and springback) can only be detected through inspection after demolding, which is a passive quality control method. Once a problem is discovered, it is necessary to stop the machine to adjust the mold or process parameters, resulting in scrap losses and reduced production efficiency.

[0007] 4. Traditional process optimization relies on trial and error, requiring multiple mold trials and modifications to find suitable parameter combinations, which is time-consuming, costly, and the optimization results are difficult to reuse across batches.

[0008] Therefore, there is an urgent need in the existing technology for an intelligent stamping process that can upgrade open-loop, pre-defined processes to closed-loop, adaptive processes, so as to achieve real-time perception, dynamic compensation and continuous optimization of the forming process. Summary of the Invention

[0009] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0010] The closed-loop stamping forming process based on global pressure sensing and real-time dynamic compensation is characterized by the following steps:

[0011] S1. For the target part, through numerical simulation process data, a dynamic pressure range map of each micro-region on the contact surface between the mold and the sheet metal is predefined to be reached during the entire molding cycle. The map is associated with the pressure target value of the molding stage and the spatial position.

[0012] S2. During the entire forming cycle from the start of the punch press and the descent of the upper die to the end of the descent of the upper die, the following sub-steps are executed simultaneously:

[0013] S2.1. At a sampling frequency of not less than 1kHz, normal contact pressure data of multiple discrete points on the contact surface between the mold and the sheet metal are collected in real time to generate a real-time pressure distribution field.

[0014] S2.2. The real-time pressure distribution field generated in step S2.1 is compared with the pressure distribution map of the corresponding molding stage defined in step S1 in millisecond-level real-time, and the pressure over-limit area and pressure under-limit area on the current contact surface are identified in real time through calculation.

[0015] S2.3. Based on the comparison results of step S2.2, under the continuous action of the main forming force, a reverse local compensation force is applied to the identified pressure over-limit area to reduce the real-time normal contact pressure value collected in step S2.1 at the pressure over-limit area, so that it falls back to the pressure range defined in step S1.

[0016] At the same time, a positive local compensation force is applied to the identified pressure-deficient area to increase the real-time normal contact pressure value collected in step S2.1 at the pressure-deficient area, so that it rises to within the pressure range defined in step S1.

[0017] The amplitude of the local compensation force is no greater than 10% of the main forming force, and the application of the local compensation force is dynamic, continuous and synchronous with the main forming process;

[0018] S3. After a single molding cycle is completed, archive the pressure-time data, compensation force application log, and final part quality inspection results for the entire molding cycle. Use the data to iteratively optimize the pressure distribution map in step S1 using a machine learning algorithm, so that the map adapts to material batch differences and achieves continuous self-optimization of the process.

[0019] Furthermore, in step S1, the numerical simulation adopts the finite element analysis method to simulate the contact process between the sheet and the mold cavity during the entire molding cycle, extracts the normal contact pressure data of each molding stage and each micro-region, and establishes the dynamic pressure range map in the form of a two-dimensional pressure cloud map or a three-dimensional pressure surface.

[0020] Furthermore, in step S2.1, the normal contact pressure data of the multiple discrete points are collected in real time by a thin-film pressure sensor array or a piezoresistive pressure sensing unit embedded in the surface of the mold cavity, and the density of the sensors or sensing units on the surface of the mold cavity is not less than 1 unit / cm².

[0021] Furthermore, in step S2.2, the millisecond-level real-time comparison is performed by a real-time computing unit deployed locally or at the edge of the stamping equipment. The real-time computing unit compares the real-time pressure distribution field with the pressure distribution map pixel by pixel, calculates the pressure deviation value at each corresponding position, and marks the position as a pressure over-limit area or a pressure under-limit area when the pressure deviation value exceeds a preset threshold.

[0022] Furthermore, in step S2.3, the reverse local compensation force and the positive local compensation force are applied independently by a miniature hydraulic cylinder, piezoelectric ceramic actuator or electromagnetic actuator embedded in the mold cavity. Each actuator corresponds one-to-one with the sensor or sensing unit that collects pressure data in step S2.1, forming a closed-loop control circuit.

[0023] Furthermore, the output force resolution of the micro hydraulic cylinder, piezoelectric ceramic actuator, or electromagnetic actuator is 0.1% to 1% of the rated pressure, the response time is no more than 5ms, and the amplitude of the local compensation force is no more than 10% of the main forming force.

[0024] Furthermore, in step S2.3, the application of the local compensation force adopts a proportional-integral-derivative control algorithm or a model predictive control algorithm, using the pressure deviation value calculated in step S2.2 as input and the output force of each actuator as output, and updating the output force command once in each control cycle, so that the real-time pressure distribution field approximates the pressure distribution spectrum.

[0025] Furthermore, in step S3, the machine learning algorithm is a neural network, support vector regression, or random forest regression algorithm. The pressure-time data and compensation force application log are used as input features, and the final part quality inspection results are used as labels to train a process model for predicting the optimal pressure distribution pattern.

[0026] Furthermore, in step S2, the closed-loop control frequency within the molding cycle is not less than 100Hz, that is, a complete control cycle of acquisition-comparison-compensation is completed every 10ms.

[0027] Furthermore, in step S3, after the pressure-time data, the compensation force application log, and the final part quality inspection results are correlated, a digital process twin file for the part is formed. The digital process twin file is used for quality traceability, process analysis, and continuous improvement.

[0028] The advantages of this invention are:

[0029] 1. This invention, through step S2.1, collects normal contact pressure data at multiple discrete points on the contact surface between the mold and the sheet metal at a sampling frequency of not less than 1 kHz to generate a real-time pressure distribution field. Step S2.2 then performs a millisecond-level real-time comparison between this real-time pressure distribution field and the pressure distribution map defined in step S1, solving the technical problems of uncontrollability and unknowability in traditional stamping processes. This process method achieves real-time, high-resolution monitoring of the pressure distribution on the contact surface between the mold and the sheet metal throughout the entire forming cycle, making the previously black-box forming process transparent and visible. Operators can monitor the stress state of each micro-region in real time, providing a precise data foundation for subsequent dynamic intervention.

[0030] 2. This invention, through step S2.3, applies a reverse local compensation force to areas with excessive pressure and a positive local compensation force to areas with insufficient pressure under continuous main molding force, based on the comparison results. The amplitude of the local compensation force is no greater than 10% of the main molding force. This solves the technical problems of traditional processes being unable to respond to real-time changes in operating conditions and unable to actively suppress defects during molding. This process method moves quality control from pre-inspection to process intervention, actively correcting abnormal pressure distribution. The reverse compensation force alleviates pressure in areas with excessive pressure, while the positive compensation force strengthens pressure in areas with insufficient pressure, thereby actively suppressing defects such as springback, wrinkling, and cracking, significantly improving molding accuracy and batch consistency. Simultaneously, the amplitude of the local compensation force is limited to within 10%, ensuring that it does not excessively interfere with the main molding force and guaranteeing the stability of the overall molding process.

[0031] 3. This invention, through step S3, archives the pressure-time data, compensation force application log, and final part quality inspection results for the entire molding cycle, and utilizes machine learning algorithms to iteratively optimize the pressure distribution map, solving the technical problems of long optimization cycles, high costs, and inability to adapt to batch-to-batch material differences in traditional processes. This process method establishes a complete closed loop from data acquisition and process intervention to model optimization. Each molding generates a complete set of digital process twin data, from which the machine learning algorithm can learn the mapping relationship between pressure distribution and part quality, automatically adjusting the pressure distribution map to adapt to the characteristics of the current batch of materials, reducing the requirements for operator experience and incoming material consistency.

[0032] 4. This invention links pressure-time data, compensation force application logs, and quality inspection results to form a digital process twin archive. Each part has a complete digital process archive, recording all key information throughout its molding cycle. When quality problems occur, the specific time and location of the pressure anomaly can be quickly traced back to accurately pinpoint the root cause of the problem, providing solid data support for continuous improvement.

[0033] In summary, this invention transforms the traditional stamping process, which involves preset parameters and open-loop execution, into a closed-loop intelligent process that integrates real-time perception, decision-making, and execution. This fundamentally solves the industry problem of uncontrollable forming processes and achieves a leap from passive inspection to proactive prevention in quality control, demonstrating significant industrial application value. Detailed Implementation

[0034] The technical solution of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, 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.

[0035] This invention provides a closed-loop stamping process based on global pressure sensing and real-time dynamic compensation. This process aims to solve the technical problems mentioned in the background art, such as uncontrollable process, inability to respond to real-time operating condition changes, passive quality control, and long process optimization cycles in traditional open-loop stamping processes. By combining a high-density pressure sensor array, millisecond-level real-time comparison, and adaptive dynamic local compensation, it achieves integrated closed-loop control of real-time sensing, decision-making, and execution in the stamping process.

[0036] The closed-loop stamping process of this invention consists of three main stages: offline modeling, online closed-loop control, and offline optimization. The online closed-loop control stage is further subdivided into three concurrently executed sub-steps: real-time acquisition of global contact pressure, real-time comparison and analysis of the pressure field, and adaptive dynamic local compensation of the forming force. These three stages and sub-steps together constitute a complete closed-loop system of perception, decision-making, execution, and learning.

[0037] The present invention will be described in detail below with reference to a specific embodiment of the stamping forming of a passenger car door inner panel (the door inner panel is a typical complex curved surface stamping part, which has deep stretching, multiple concave and convex features, is sensitive to pressure distribution, and is prone to wrinkling and cracking defects).

[0038] In this embodiment, the target part is the inner panel of the left front door of a certain vehicle model, made of DC06 deep-drawing steel with a thickness of 0.8mm. First, a stamping simulation model of the part is established in finite element analysis software such as AutoForm or LS-DYNA. The model includes a punch, a die, a blank holder, and a sheet metal mesh, with the mesh size set to 1mm × 1mm to capture pressure changes in the microscopic region.

[0039] Then, the entire forming process is simulated, from the closing of the blank holder ring, the downward movement of the punch to the bottom dead center, the holding pressure stage, and the upward return of the punch. The normal contact pressure values ​​at each node on the contact surface between the mold cavity and the sheet metal are output for each incremental step (time step 0.1ms). These pressure data are then organized according to the forming stage (e.g., blank holder ring closing stage, initial, middle, and near-bottom dead center movement of the punch, holding pressure stage, and return stage) and spatial location (divided by the coordinate grid of the mold cavity surface) to form a series of two-dimensional pressure cloud maps or three-dimensional pressure surfaces.

[0040] Based on simulation results and combined with process data from historical qualified batches (such as the qualified pressure distribution recorded when producing the same part in the past), a dynamic optimal pressure range is defined for each micro-region at each molding stage. This range is not a single value, but an interval (for example, the ideal pressure range for a certain region during the middle of the punch's downward movement is 15MPa to 25MPa). The graph is stored in the form of a data table or matrix, where each element corresponds to a spatial coordinate (i,j) and a molding stage t, with values ​​representing the lower and upper pressure limits.

[0041] Specifically, in this embodiment, the mold cavity surface is divided into 5000 micro-regions (approximately one region per square centimeter for a 70cm x 70cm part). Each region has a defined pressure range for 10 forming stages (blank ring closure, punch descent of 10%, 20%, ..., 90%, and pressure holding). These data form a 5000 x 10 pressure range matrix, serving as a standard template for subsequent real-time comparison.

[0042] The pressure distribution map is stored on a local server or edge computing node on the stamping production line, saved as a binary file or database. At the start of each stamping cycle, the control system automatically loads the map corresponding to the current part and prepares to compare it with the real-time acquired data.

[0043] This embodiment features a specially modified stamping die, embedding a thin-film pressure sensor array on the surface of the concave mold cavity (the side in contact with the sheet metal). The sensors utilize piezoresistive sensing units on a polyimide substrate, each with a sensing area of ​​2mm × 2mm, and a density of 1 unit / cm² (i.e., one sensing unit per square centimeter). For the door inner panel mold in this embodiment (cavity area approximately 5000cm²), approximately 5000 sensing units are arranged. The sensor array is connected to a multi-channel high-speed data acquisition card via flexible printed circuitry. The data acquisition card uses the NI PXIe-6363 series, featuring 32 synchronous analog input channels and a sampling rate of up to 2MS / s. Using time-division multiplexing technology, the 5000 sensing units are grouped into 32 groups, sampling alternately. The actual sampling frequency is set to 2kHz (collecting full-area pressure data every 0.5ms), meeting the requirement of not less than 1kHz. Miniature piezoelectric ceramic actuators are embedded on the surface of the convex mold cavity corresponding to the position of each pressure sensing unit. The actuator uses a multilayer piezoelectric ceramic with a diameter of 5mm and a thickness of 3mm, a maximum output force of 500N, an output force resolution of 0.5% of the rated pressure (i.e., 2.5N), and a response time of less than 1ms. Each actuator is connected to the control computer via an independent drive circuit, and each actuator can independently apply a local compensation force in either the forward (ejection force) or reverse (retraction force) direction. The actuators and sensing units are spatially aligned one-to-one, forming 5000 independent closed-loop control loops. An industrial-grade real-time control computer (Beckford CX2040) running a real-time operating system (TwinCAT) is deployed. This computer is connected to the data acquisition card and actuator drive circuit via a PCIe bus, responsible for performing the aforementioned data comparison calculations. Data acquisition is synchronously initiated by the real-time control computer when the press starts and the upper die begins to descend. Voltage signals output from 5000 sensing units are continuously acquired at a sampling frequency of 2kHz and converted into normal contact pressure values ​​(in MPa) using a calibration formula. Every 0.5ms, a real-time pressure distribution field with 5000 data points is generated and stored in memory as a two-dimensional array. To ensure data synchronization, the acquisition trigger signal is taken from the position pulse of the punch press encoder to ensure that each frame of data corresponds strictly to the current position of the upper die (i.e., the forming stage). After each frame of real-time pressure distribution field is acquired, the real-time control computer immediately performs the following operations: (1) Based on the current position of the upper die (converted from the encoder reading to the forming progress percentage, such as 30% before the bottom dead center), retrieve the pressure range matrix (lower and upper limits of 5000 regions) from the established pressure distribution map. (2) Compare the real-time pressure distribution field with the ideal pressure range matrix pixel by pixel: For each sensing unit position, compare the real-time pressure value P. real The lower limit of pressure P at that location low and upper limit P high If Preal > P high Then mark that location as a pressure over-limit area; if P real <P low If the pressure is within the acceptable range, it is marked as an insufficient pressure area; if it is within the acceptable range, it is marked as a normal area. (3) Calculate the deviation value for each location: For the out-of-range area, the deviation ΔP = P real -P high For the insufficient region, the deviation ΔP = P low -P real The deviation in the normal region is zero. The deviation value is used as the input for the subsequent control algorithm. (4) The entire comparison process is completed within 0.3ms (accelerated by GPU parallel processing), plus the acquisition time of 0.5ms, totaling about 0.8ms, which meets the requirement of a 10ms control cycle.

[0044] Based on the millisecond-level real-time comparison results of the pressure distribution map at the corresponding molding stage, the real-time control computer immediately calculates the local compensation force that should be applied to each actuator. Specifically, this embodiment uses a proportional-integral-derivative (PID) control algorithm. For each independent control loop (each sensor unit-actuator pair), the deviation value ΔP at that position is used as input, and the output is the compensation force F that the actuator needs to apply. compThe PID parameters are pre-tuned: proportional gain Kp = 0.8, integral time Ti = 5ms, and derivative time Td = 0.5ms. The control cycle is set to 2ms (i.e., the output force command is updated every 2ms), which is slightly longer than the data acquisition cycle (0.5ms) to allow for calculation and actuator response. For areas with excessive pressure, a reverse local compensation force is applied, i.e., the actuator moves slightly upward (away from the plate material) to relieve the pressure there; for areas with insufficient pressure, a positive local compensation force is applied, i.e., the actuator moves slightly downward (towards the plate material) to increase the pressure there. The amplitude of the compensation force is calculated in real time by the PID controller based on the deviation, but the upper limit is limited to within 10% of the main forming force. In this embodiment, the main forming force is approximately 2000kN (press tonnage 2500kN), therefore the maximum compensation force of each actuator does not exceed 200N (10% × 2000kN / 5000 ≈ 40N, actually rounded to 200N as a safe upper limit). In practice, the compensation force in most areas is between 5N and 50N. The application of this compensation force is dynamic and continuous. Within each control cycle (2ms), the control computer sends a new force command to the drive circuit of each actuator. The actuator completes its action within 1ms of receiving the command. Therefore, the total delay from pressure acquisition to the compensation force taking effect is approximately 3ms (0.5ms acquisition + 0.3ms comparison + 1ms calculation + 1ms actuator response), far less than the 10ms control cycle requirement, achieving synchronization. Throughout the process, the press's main forming force (i.e., the punch's downward force) acts continuously, unaffected by local compensation. Local compensation forces are only superimposed on the main forming force, allowing for fine-tuning without altering the overall forming process. In this embodiment, the control cycle frequency is set to 500Hz (i.e., one cycle every 2ms), higher than the minimum requirement of 100Hz. The entire forming cycle (from the start of the upper die's downward movement to the end of its upward movement) lasts approximately 1.2 seconds, during which approximately 600 control cycles are executed. Each cycle collects, compares, and compensates for the pressure in 5,000 areas to ensure that the pressure distribution always closely approximates the ideal spectrum.

[0045] After a single molding cycle, the real-time control computer packages and archives all data from the entire molding cycle into a single data file. This file contains pressure-time data: pressure values ​​from 5000 sensor units collected every 0.5ms from the start to the end of molding, totaling approximately 2.4 million data points (1.2 seconds × 2000Hz × 5000 ≈ 1.2 billion data points; actual storage is compressed, storing only the average value every 2ms, approximately 3 million data points). Compensation force application log: compensation force values ​​applied by each actuator in each control cycle (every 2ms), totaling approximately 3 million data points. Molding process parameters: upper mold position-time curve, main molding force-time curve, blank holder force, etc. Part identification: part serial number, production time, material batch number. After demolding, the part enters the online inspection station for 3D scanning (detecting springback and contour), surface defect visual inspection (wrinkling, cracking), and thickness reduction rate sampling inspection. The test results (pass / fail, and specific defect type and location) are linked with the aforementioned data files to form a complete digital process twin file. This file is then uploaded to the factory's MES system or a cloud database.

[0046] After each week or after a certain number of productions (e.g., 1000 pieces), the accumulated digital process twin archives are used to train a machine learning model to optimize the stress distribution map.

[0047] This embodiment employs a random forest regression algorithm. Input features include: the pressure-time curve for each micro-region (dimensionality-reduced features, such as peak pressure, pressure rise rate, average pressure during the holding phase, etc.), and compensation force application logs (cumulative compensation energy and maximum compensation force for each region). Labels represent the final part's quality inspection results (0-1 representing pass probability, or coded by defect type). Training data volume: approximately 10,000 samples (from production data of the past week). The dataset is divided into training and test sets at 80% / 20%. The random forest contains 500 decision trees, each with a maximum depth of 20. After training, the model can predict the expected quality of parts under a given pressure distribution map. Using this model, a pressure distribution map that maximizes the expected quality is found through grid search or Bayesian optimization, serving as the updated ideal map. The new map replaces the old map and is applied to subsequent production. The old map is also retained as a backup. After several iterations, the pressure distribution pattern gradually converges to the optimal solution, which can adapt to batch differences in materials (such as yield strength fluctuation of ±20MPa and thickness tolerance of ±0.05mm), significantly reducing the dependence on operator experience.

[0048] The above embodiment uses the inner door panel as an example, but the present invention is also applicable to other complex stamped parts, such as the outer panel of the engine hood, the outer panel of the side panel, and the B-pillar reinforcement plate. For different parts, only the simulation model and pressure distribution map in step S1 need to be adjusted, and the hardware system (sensor and actuator array) can be rearranged according to the shape of the mold cavity. In addition to piezoresistive thin-film sensors, capacitive thin-film sensors or fiber Bragg grating (FBG) sensors can also be used. Capacitive sensors have higher sensitivity, but slightly worse temperature stability; FBG sensors have strong anti-electromagnetic interference capabilities and are suitable for use in large press environments. Those skilled in the art can select the appropriate sensor type according to the actual working conditions. In addition to piezoelectric ceramic actuators, miniature hydraulic cylinders or electromagnetic actuators can also be used. Miniature hydraulic cylinders can provide larger compensation forces (up to several thousand Newtons), but the response speed is slower (about 10ms); electromagnetic actuators have a fast response (<1ms) but lower force density. For applications requiring large compensation forces (such as thick plate stamping), a scheme of mixed arrangement of miniature hydraulic cylinders and piezoelectric ceramics can be adopted. Besides PID control, Model Predictive Control (MPC) can also be used. MPC can predict pressure distribution changes several control steps in advance and optimize the compensation force sequence, making it suitable for situations with strong nonlinearity in the molding process. In this embodiment, PID control can meet the requirements for medium-complexity parts such as car door inner panels. Besides random forests, deep neural networks (DNNs) or support vector regression (SVRs) can also be used. DNNs can fit more complex nonlinear relationships but require more training data; SVRs perform better with small sample sizes. Those skilled in the art can choose the appropriate algorithm based on the amount of data. To verify the effectiveness of this invention, a comparative experiment was conducted. On the same press, using the same set of molds, 500 car door inner panels were produced using both the traditional open-loop process and the closed-loop process of this invention. The test results are as follows:

[0049] Wrinkling defect rate 3.2% 0.4% 87.5% Cracking defect rate 1.8% 0.1% 94.4% standard deviation of rebound 0.42mm 0.08mm 81.0% Fluctuation in thickness reduction rate ±4.5% ±1.2% 73.3% Batch pass rate 94.7% 99.3% 4.6%

[0050] The results show that the closed-loop stamping process of this invention significantly reduces the defect rate and improves dimensional accuracy and consistency, thus verifying its technical effectiveness.

[0051] The above embodiments detail the specific implementation of the closed-loop stamping forming process based on global pressure sensing and real-time dynamic compensation of the present invention. By establishing a pressure distribution map in step S1, achieving high-speed closed-loop control within the forming cycle in step S2 (including high-density pressure acquisition, millisecond-level comparison, and adaptive local compensation), and iteratively optimizing the map using machine learning in step S3, the present invention upgrades the traditional open-loop stamping process into a closed-loop intelligent process integrating real-time sensing, decision-making, and execution. This process can proactively suppress defect generation, adapt to batch material differences, and continuously self-optimize, significantly improving the quality and production efficiency of stamped parts.

[0052] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A closed-loop stamping forming process based on global pressure sensing and real-time dynamic compensation, characterized in that, Includes the following steps: S1. For the target part, through numerical simulation process data, a dynamic pressure range map of each micro-region on the contact surface between the mold and the sheet metal is predefined to be reached during the entire molding cycle. The map is associated with the pressure target value of the molding stage and the spatial position. S2. During the entire forming cycle from the start of the punch press and the descent of the upper die to the end of the descent of the upper die, the following sub-steps are executed simultaneously: S2.

1. At a sampling frequency of not less than 1kHz, normal contact pressure data of multiple discrete points on the contact surface between the mold and the sheet metal are collected in real time to generate a real-time pressure distribution field. S2.

2. The real-time pressure distribution field generated in step S2.1 is compared with the pressure distribution map of the corresponding molding stage defined in step S1 in millisecond-level real-time, and the pressure over-limit area and pressure under-limit area on the current contact surface are identified in real time through calculation. S2.

3. Based on the comparison results of step S2.2, under the continuous action of the main forming force, a reverse local compensation force is applied to the identified pressure over-limit area to reduce the real-time normal contact pressure value collected in step S2.1 at the pressure over-limit area, so that it falls back to the pressure range defined in step S1. At the same time, a positive local compensation force is applied to the identified pressure-deficient area to increase the real-time normal contact pressure value collected in step S2.1 at the pressure-deficient area, so that it rises to within the pressure range defined in step S1. The amplitude of the local compensation force is no greater than 10% of the main forming force, and the application of the local compensation force is dynamic, continuous and synchronous with the main forming process; S3. After a single molding cycle is completed, archive the pressure-time data, compensation force application log, and final part quality inspection results for the entire molding cycle. Use the data to iteratively optimize the pressure distribution map in step S1 using a machine learning algorithm, so that the map adapts to material batch differences and achieves continuous self-optimization of the process.

2. The closed-loop stamping forming process based on global pressure sensing and real-time dynamic compensation according to claim 1, characterized in that, In step S1, the numerical simulation adopts the finite element analysis method to simulate the contact process between the sheet and the mold cavity during the entire forming cycle, extracts the normal contact pressure data of each forming stage and each micro-region, and establishes the dynamic pressure range map in the form of a two-dimensional pressure cloud map or a three-dimensional pressure surface.

3. The closed-loop stamping forming process based on global pressure sensing and real-time dynamic compensation according to claim 1, characterized in that, In step S2.1, the normal contact pressure data of the multiple discrete points are collected in real time by a thin-film pressure sensor array or a piezoresistive pressure sensing unit embedded in the surface of the mold cavity. The density of the sensors or sensing units on the surface of the mold cavity is not less than 1 unit / cm².

4. The closed-loop stamping forming process based on global pressure sensing and real-time dynamic compensation according to claim 1, characterized in that, In step S2.2, the millisecond-level real-time comparison is performed by a real-time computing unit deployed locally or at the edge of the stamping equipment. The real-time computing unit compares the real-time pressure distribution field with the pressure distribution map pixel by pixel and calculates the pressure deviation value at each corresponding position. When the pressure deviation value exceeds a preset threshold, the position is marked as a pressure over-limit area or a pressure under-limit area.

5. The closed-loop stamping forming process based on global pressure sensing and real-time dynamic compensation according to claim 1, characterized in that, In step S2.3, the reverse local compensation force and the positive local compensation force are applied independently by a miniature hydraulic cylinder, piezoelectric ceramic actuator or electromagnetic actuator embedded in the mold cavity. Each actuator corresponds one-to-one with the sensor or sensing unit that collects pressure data in step S2.1, forming a closed-loop control circuit.

6. The closed-loop stamping forming process based on global pressure sensing and real-time dynamic compensation according to claim 5, characterized in that, The output force resolution of the micro hydraulic cylinder, piezoelectric ceramic actuator or electromagnetic actuator is 0.1% to 1% of the rated pressure, the response time is no more than 5ms, and the amplitude of the local compensation force is no more than 10% of the main forming force.

7. The closed-loop stamping forming process based on global pressure sensing and real-time dynamic compensation according to claim 1, characterized in that, In step S2.3, the application of the local compensation force adopts a proportional-integral-derivative control algorithm or a model predictive control algorithm. The pressure deviation value calculated in step S2.2 is used as the input, and the output force of each actuator is used as the output. The output force command is updated once in each control cycle so that the real-time pressure distribution field approximates the pressure distribution spectrum.

8. The closed-loop stamping forming process based on global pressure sensing and real-time dynamic compensation according to claim 1, characterized in that, In step S3, the machine learning algorithm is a neural network, support vector regression, or random forest regression algorithm. The pressure-time data and compensation force application log are used as input features, and the final part quality inspection results are used as labels to train a process model for predicting the optimal pressure distribution pattern.

9. The closed-loop stamping forming process based on global pressure sensing and real-time dynamic compensation according to claim 1, characterized in that, In step S2, the closed-loop control frequency within the molding cycle is not less than 100Hz, that is, a complete control cycle of acquisition-comparison-compensation is completed every 10ms.

10. The closed-loop stamping forming process based on global pressure sensing and real-time dynamic compensation according to claim 1, characterized in that, In step S3, the pressure-time data, the compensation force application log, and the final part quality inspection results are correlated to form a digital process twin file for the part. The digital process twin file is used for quality traceability, process analysis, and continuous improvement.