A flexible stamping process method suitable for multi-specification cargo box skin

Abstract

This invention discloses a flexible stamping process applicable to multi-specification cargo box skins, belonging to the field of hydraulic press stamping forming control technology. Addressing the problem of pre-wrinkling of sheet metal before the pressure ring closes due to differences in filling time and valve response delay when the four corner pressure cylinders of existing hydraulic presses independently build up pressure, this invention obtains the estimated pressure building time for each of the four pressure cylinders. This estimated value is calculated based on the target pressure value, the opening delay time of the proportional relief valve, and the filling time of the pressure oil circuit. The starting delay time of each cylinder is determined based on the estimated pressure building time, and they are started sequentially. During the pressure building process, the actual pressure value of each cylinder is collected in real time. The difference between the actual value and the target value is used as the tracking error, and the difference between the actual values ​​of each cylinder is used as the synchronization error. Control commands are generated to drive the proportional relief valve based on this. After the pressure of each cylinder reaches the target value, the steady-state deviation is collected and stored for subsequent target pressure value correction during the stamping stroke. This method is applicable to the flexible stamping production of multi-specification workpieces such as automotive cargo box skins.

Description

Technical Field

[0001] This invention belongs to the field of hydraulic press stamping and forming control technology, specifically a flexible stamping process method applicable to multi-specification cargo box skins. Background Technology

[0002] In the field of hydraulic press stamping control, some progress has been made. For example, the 2013 paper by Wei Xiang, Lu Hong, and Li Xiuzhu, on the program control of the blanking system of a thin-plate forming hydraulic press, discloses a four-corner blanking force follow-up adjustment blanking system. This system uses independent proportional relief valves to control the output pressure of each blanking cylinder, and adjusts the output pressure of each of the four corner hydraulic cylinders in real time according to the signal from the slider displacement sensor. The follow-up hydraulic pressure control of the four corner blanking force is achieved by comparing the feedback signal from the pressure sensor with the program curve. However, the above-mentioned existing technology still has the following problems when dealing with the flexible stamping production of multi-specification cargo box skins (such as asymmetrical workpieces such as the rear lower panel and rear pillar skin): First, there is a synchronization error in the pressure-building phase of the four corner pressing cylinders. The independent PID control architecture for the four corners disclosed in the literature "Program Control of the Edge Pressing System of Thin Plate Forming Hydraulic Press" only allows each pressing cylinder to independently adjust its closed-loop control based on the deviation between its target pressure and the actual pressure; there is no information exchange or coordination mechanism between the cylinders. In actual production, when the target pressure values ​​set for the four corners are different, the filling time and proportional relief valve opening delay time of each cylinder differ, resulting in inconsistent pressure-building completion times for each cylinder. The pressing cylinder that completes pressure building first has already partially compressed the sheet metal, while the corresponding area of ​​the cylinder that builds pressure later remains in a free state. Irreversible pre-deformation or pre-wrinkling occurs in the sheet metal before the pressing ring is fully closed, and this defect cannot be eliminated in subsequent stamping strokes.

[0003] Secondly, the efficiency of quickly optimizing process parameters after mold change is low. Although some fully automated mold-changing technologies achieve rapid physical replacement of molds and automatic parameter recall, the historical parameters recalled are only "usable" parameters, not the optimal parameters adapted to the current working conditions (sheet material batch, lubricant type, ambient temperature changes). In actual production, after mold change, process personnel still need to perform multiple trial runs (usually 3 to 10 times) to adjust the parameters from "usable" to "optimal." Each trial run consumes sheet material and occupies press time, and the debugging efficiency is limited by the operator's experience level. Existing parameter optimization methods based on finite element simulation or response surface methodology (such as optimization methods based on Kriging interpolation and genetic algorithms), although proven effective in academic research, require the pre-establishment of high-precision simulation models and multiple iterative calculations. The modeling cycle is long and cannot be quickly executed on-site, making it unsuitable for production scenarios with frequent mold changes for multiple varieties and specifications.

[0004] Third, there is a lack of rapid online methods for determining the peak value ratio of blank holder force. For complex-shaped parts such as aluminum alloy cargo box skins, the position of the peak value of the blank holder force during the stamping stroke (i.e., the moment when the peak appears in the "increase-decrease" curve) is extremely sensitive to the forming quality—a peak value that is too early will inhibit the initial material flow, leading to thinning or even tearing of the sidewalls; a peak value that is too late will cause excessive material flow in the flange area at the end, resulting in wrinkling. Although existing academic research has recognized that there is a reasonable range for the time when the blank holder force reaches its maximum value, it has not provided an engineering method to quickly determine this optimal peak value moment on the production line. Process engineers usually rely on finite element simulation calculations (which take several hours to several days) or repeated trial molding to explore this parameter, which is inefficient. Although the "slider displacement-blank holder force" follow-up control system disclosed in the literature "Program Control of Blank Holder System of Thin Plate Forming Hydraulic Press" can output the blank holder force according to the preset curve, it does not involve how to determine the optimal peak value ratio online based on the flow response of the sheet metal edge.

[0005] The aforementioned problems are particularly prominent in the flexible stamping production of multi-specification cargo box skins—there are many types of workpieces, frequent switching, and aluminum alloy materials are sensitive to fluctuations in process parameters. Existing technologies cannot balance mold changing efficiency and forming quality. Summary of the Invention

[0006] One objective of this invention is to address the problem of inconsistent pressure build-up times in existing hydraulic presses where the four corner pressing cylinders are controlled independently by PID controllers. Differences in filling time and valve response delays during the pressure build-up process lead to inconsistent pressure completion times for each cylinder. The pressing cylinder that completes pressure build-up first locally presses the sheet metal, while the corresponding area of ​​the cylinders that complete pressure build-up later remains in a free state, causing irreversible pre-wrinkling defects in the sheet metal before the pressing ring closes. This defect cannot be eliminated in subsequent stamping strokes, affecting the finished product yield. Existing technology does not provide a control method for achieving synchronous pressure build-up at all four corners during the pressure build-up stage.

[0007] To achieve the above objectives, the present invention provides a flexible stamping process method suitable for multi-specification cargo box skins, comprising a hydraulic press having four edge-pressing cylinders distributed at the four corners and a controller for controlling the four edge-pressing cylinders. The method includes the following steps in the edge-pressing ring closing stage: Step 1: Obtain the estimated pressure build-up time for each of the four pressure cylinders. The estimated pressure build-up time for each pressure cylinder is calculated based on the target pressure value of the pressure cylinder, the opening delay time of the proportional relief valve to which the pressure cylinder belongs, and the filling time of the pressure oil circuit to which the pressure cylinder belongs. Step 2: Based on the estimated pressure build-up time of each of the four pressure cylinders, and with the goal of making the pressure build-up completion time of each pressure cylinder approach the same time, determine the start-up delay time of each of the four pressure cylinders, and start the four pressure cylinders sequentially according to the start-up delay time. Step 3: During the pressure building process of each pressing cylinder, the actual pressure value of each pressing cylinder is collected in real time. The difference between the actual pressure value of each pressing cylinder and the target pressure value of the pressing cylinder is used as the tracking error. The difference between the actual pressure value of each pressing cylinder and the actual pressure values ​​of the other three pressing cylinders is used as the synchronization error. Based on the tracking error and the synchronization error, the control command of the pressing cylinder is generated to drive the proportional relief valve to which the pressing cylinder belongs to adjust the valve core opening. Step 4: After the actual pressure values ​​of the four pressing cylinders have all reached their respective target pressure values, the steady-state pressure values ​​of each pressing cylinder are collected, the steady-state deviation between the steady-state pressure value and the target pressure value is calculated, and the steady-state deviation is stored in the controller for correction of the target pressure value of the corresponding pressing cylinder in the subsequent stamping stroke.

[0008] Preferably, the stamping stroke further includes the following steps: Step 5: Real-time acquisition of the displacement and velocity values ​​of the hydraulic press slider, and prediction of the slider's predicted displacement value after a preset delay time based on the displacement and velocity values; Step 6: Calculation of the corresponding feedforward blank holder force target value based on the predicted displacement value and a preset displacement-blank force mapping relationship, and issuance of a feedforward control command to the proportional relief valve of the blank holder cylinder according to the feedforward blank holder force target value; Step 7: Using the current displacement value of the slider as the feedback displacement value, calculation of the feedback blank holder force target value based on the feedback displacement value and the displacement-blank force mapping relationship. A feedback control command is generated based on the deviation between the target value of the feedback blank holder force and the actual pressure value of the blank holder cylinder; Step 8: The feedforward control command and the feedback control command are superimposed to generate a composite control command to drive the proportional relief valve to adjust the valve core opening; Step 9: After each stamping stroke, historical data of the actual pressure value of the blank holder cylinder changing with the slider displacement during that stroke are collected, the historical data are compared with the displacement-blank holder force mapping relationship, the tracking error at each displacement point is calculated, and the preset delay time is corrected based on the tracking error for the calculation of the predicted displacement value of the next stamping stroke.

[0009] Preferably, before determining the target pressure values ​​of the four blank holder cylinders, the following steps are included: Step a: Install the aluminum alloy cargo box skin sheet to be formed on the worktable of the hydraulic press, and install displacement sensors at multiple preset positions on the edge of the sheet. The displacement sensors are used to measure the displacement of the sheet edge sliding into the die during the stamping process; Step b: Perform a trial stamping stroke. During the trial stamping stroke, the controller controls the target pressure value of each blank holder cylinder according to a preset scanning curve. The scanning curve is the ratio of the peak blank holder force moment to the total stamping stroke, which continuously changes from a first ratio value to a second ratio value; Step c: During the trial stamping stroke, the displacement signals of each displacement sensor are collected in real time, and the rate of change of each displacement signal with time is calculated. When any rate of change exceeds a preset threshold range, the proportion of the peak blank holder force moment corresponding to that moment is recorded as a critical ratio value; Step d: Determine the optimal proportion of the peak blank holder force moment based on the multiple recorded critical ratio values, and store the optimal proportion in the controller for setting the blank holder force curve in subsequent batch stamping strokes.

[0010] Preferably, before determining the target pressure value of each of the four pressing cylinders, the method further includes the following steps: Step e: Install at least one non-contact thickness sensor in the loading area or workbench of the hydraulic press. The thickness sensor is used to measure the thickness value of each measuring point on the sheet metal; Step f: After placing the sheet metal to be formed at a predetermined position on the workbench, control the thickness sensor to move along a preset scanning path on the surface of the sheet metal, collect the thickness values ​​of multiple measuring points on the sheet metal, and calculate the thickness deviation between the thickness value of each measuring point and the nominal thickness value; Step g: Input the thickness deviation into a preset pressing force correction model. The correction model outputs the correction coefficient of the target pressure value of each pressing cylinder according to the correspondence between the thickness deviation and the pressing force correction coefficient; Step h: Multiply the original target pressure value of each pressing cylinder by the corresponding correction coefficient to obtain the corrected target pressure value, which is used for pressing force control in the subsequent pressing ring closing stage.

[0011] Preferably, the stamping stroke further includes the following steps: Step i: At the beginning stage of the stamping stroke, control each blank holder cylinder to apply blank holder force according to a preset initial target pressure value, and simultaneously collect pressure sensor signals of each blank holder cylinder and displacement sensor signals of the slider in real time; Step j: Extract pressure overshoot from the pressure sensor signals, wherein the pressure overshoot is the difference between the peak pressure when the actual pressure of the blank holder cylinder first reaches the initial target pressure value and the initial target pressure value; extract displacement fluctuation amplitude from the displacement sensor signals, wherein the displacement fluctuation amplitude is the root mean square value of the high-frequency fluctuation component of the displacement signal of the slider during the period when the blank holder force remains constant; Step k: Input the pressure overshoot and the displacement fluctuation amplitude into a preset friction coefficient estimation model, wherein the friction coefficient estimation model outputs the current friction coefficient estimation value between the sheet metal and the die; Step l: Calculate the dynamic compensation coefficient of the target pressure value of each blank holder cylinder according to the ratio of the friction coefficient estimation value to the nominal friction coefficient; Step m: Multiply the current target pressure value of each blank holder cylinder by the dynamic compensation coefficient to obtain the compensated target pressure value, which is used for blank holder force control in subsequent stamping stroke stages.

[0012] Preferably, after mold change and before batch stamping, the following steps are further included: Step n: Read the specification identifier of the currently introduced workpiece from the controller, and retrieve the historical blank holder force parameter record corresponding to the specification from the historical parameter database according to the workpiece specification identifier. The historical parameter record includes at least the blank holder force curve shape parameter, sheet metal batch number, lubricant type, and ambient temperature; Step o: Take the blank holder force curve shape parameter in the historical parameter record that is closest to the current sheet metal batch number, lubricant type, and ambient temperature as the initial blank holder force parameter, and execute the first trial stamping stroke; Step p: After the first trial stamping stroke, measure the quality index value of the formed sheet metal. The quality index includes at least one of the following: sheet metal edge wrinkling height, critical fracture displacement, or surface waviness. Step 1: Calculate the deviation between the quality index value and the preset target value; Step 2: Calculate the correction direction and first step length of the blank holder force curve shape parameter according to the deviation and the preset adjustment rules to obtain the blank holder force parameter after the first correction, and execute the second trial punch stroke; Step 3: After the second trial punch stroke, measure the quality index value of the formed sheet again, and combine the quality index values ​​of the first and second trial punches and the corresponding blank holder force parameters to fit a local linear relationship between the quality index and the blank holder force parameter; Step 4: Calculate the blank holder force parameter that makes the quality index value reach the preset target value according to the local linear relationship, and use it as the optimal blank holder force parameter for the current batch production, and store the optimal parameter and the current working conditions in the historical parameter database.

[0013] Preferably, the following steps are included before the closing stage of the pressure ring: Step A: For each specification of cargo box skin workpiece, the stiffness characteristic value and area distribution characteristic value of the workpiece are obtained through offline calibration, and the stiffness characteristic value and area distribution characteristic value are associated with the workpiece specification identifier and stored in the workpiece parameter library of the controller; Step B: A first mapping relationship is established between the tracking error gain coefficient Kp and the stiffness characteristic value and area distribution characteristic value, and a second mapping relationship is established between the synchronization error gain coefficient Ks and the area distribution characteristic value, and the parameters of the first mapping relationship and the second mapping relationship are stored in the controller; Step C: After the mold change is completed, the controller obtains the specification identifier of the currently replaced workpiece, and reads the corresponding stiffness characteristic value and area distribution characteristic value from the workpiece parameter library according to the specification identifier; Step D Step E: Substitute the read stiffness characteristic value and area distribution characteristic value into the first mapping relationship to calculate the Kp value, and substitute the area distribution characteristic value into the second mapping relationship to calculate the Ks value. Use the calculated Kp value and Ks value as the tracking error gain coefficient and synchronization error gain coefficient in step three. Step E: During the batch stamping process, the controller collects the pressure overshoot and synchronization error convergence time of each stamping stroke after the stroke ends. When the pressure overshoot exceeds the preset first threshold or the synchronization error convergence time exceeds the preset second threshold for N consecutive stamping strokes, the parameter fine-tuning process is triggered. The extreme value search method is used to make a trial adjustment based on the current Kp value and Ks value with a step size not exceeding 5% of the current value. In subsequent strokes, the Kp value and Ks value are updated to the workpiece parameter library according to the adjusted control effect.

[0014] Preferably, the stamping stroke further includes the following steps: offline extraction of inflection point displacement values ​​on a preset displacement-blank force mapping curve, wherein the inflection point is defined as a point where the first derivative of the mapping curve is discontinuous or a point where the sign of the second derivative changes, and storing the displacement values ​​of each inflection point in the controller; during the stamping stroke, real-time acquisition of the slider displacement value, calculation of the displacement difference between the slider displacement value and each inflection point displacement value, and determination that the slider has entered the inflection point proximity region when the displacement difference is less than a preset first threshold; within the inflection point proximity region, the calculation method for the predicted displacement value in step five is switched from a first-order prediction model based on the current displacement and current speed to a time prediction model based on the numerical integration of the current displacement and speed sequence, i.e., the predicted displacement value S_pred = S_current + v_current × τ + 0.5 × a_current × τ 2Where τ is the preset delay time and a_current is the current acceleration of the slider, which is calculated by the controller by the difference between the two most recent velocity sampling values; after the slider passes the inflection point displacement value, the calculation method of the predicted displacement value is switched back to the first-order prediction model in step five; after each stamping stroke, the tracking deviation between the actual blank holder force curve and the preset displacement-blank holder force mapping curve in the area near the inflection point is collected. When the tracking deviation exceeds the preset second threshold, the values ​​of the first threshold and τ are adjusted for use in the calculation of the predicted displacement value of subsequent stamping strokes.

[0015] Preferably, the stamping stroke further includes the following steps: Step I: Within the initial displacement range after the slider begins to descend, the process of changing the blank holder force command from the steady-state target value after the blank holder ring is closed to the initial target value of the follow-up control is used as a natural identification window; Step II: Within the natural identification window, the time t0 when the blank holder force command begins to change is recorded, the actual pressure value of the blank holder cylinder is collected in real time, and the time t1 is recorded when the actual pressure value first reaches the preset percentage of the blank holder force command target value. The identification delay time τ_est = t1- t0- Δt_model is calculated, where Δt_model is a preset correction term; Step III: The preset delay time in Step V is replaced with the identification delay time τ_est, which is used for calculating the predicted displacement value of the remaining part of this stamping stroke; Step IV: The identification delay time τ_est is stored in the historical data sequence in Step IX, which is used to update the preset delay time by weighted average.

[0016] Preferably, the following steps are included between step q and step r: Step q1: Record the parameter value of the first trial punch as x1 and the quality index value as y1, record the parameter value of the second trial punch as x2 and the quality index value as y2, and calculate the quality index change Δy = y2 - y1 and the unit step change rate δ1 = Δy / |x2 - x1|; Step q2: Read the reference change rate δ0 corresponding to the workpiece specification from the memory of the controller, calculate the comparison result of |δ1 - δ0| with the preset first threshold T1, and the comparison result of |δ1 - δ0| with the preset second threshold T2, where T2 > T1; Step q3: When |δ1 - δ0| ≤ T1, determine that the current region is linear, and calculate the optimal blank holder force parameters according to steps r and s; Step q4: When T1 < |δ1 - δ0| ≤ When T2 and y1 and y2 are both located on the same side of the target quality index value, it is determined that the current region is nonlinear, and step r is stopped. Taking the parameter corresponding to the better quality index of x1 and x2 as the benchmark, the third step size Δx3 is calculated in the direction of improving the quality index according to the preset step size reduction coefficient, and the third trial stroke is performed to obtain the third trial stroke point (x3, y3); Step q5: Fit the three trial stroke points (x1, y1), (x2, y2) and (x3, y3) into a quadratic curve y = a·x 2 + b·x + c, calculate the minimum point x_opt = -b / (2a) of the quadratic curve, and use it as the optimal blank holder force parameter; Step q6: When |δ1- δ0|>T2, determine that the current region is in a strong nonlinear region or has exceeded the optimal interval, and use the parameter corresponding to the one with better quality index between x1 and x2 as the optimal blank holder force parameter, and enable the quality monitoring and fine-tuning mechanism in steps p to s.

[0017] The present invention has at least the following beneficial effects: 1. By estimating the pressure build-up time, planning the start-up sequence, and cross-coupling control (introducing both tracking and synchronization errors), the actual pressure of the four pressing cylinders reaches their respective target pressure values ​​synchronously during the pressure build-up stage. Compared with existing technologies where each cylinder is controlled independently and the pressure build-up completion time is inconsistent, this method eliminates the pre-wrinkling defect of the sheet metal caused by localized pre-pressing, and does not require reducing the pressure build-up speed or increasing the production cycle time.

[0018] 2. By employing a feedforward-feedback composite control, the blank holder force command is issued in advance based on predicted displacement, overcoming the phase lag caused by the inherent response delay of the hydraulic system. When the slider descends at high speed, the actual output curve of the blank holder force remains synchronized with the preset displacement-blank holder force curve, avoiding wrinkling or tearing caused by pressure lag. Combined with the method of rapidly identifying τ online using an active perturbation identification window, a small step with an amplitude not exceeding 5% of the steady-state pressure is actively applied before the start of each stamping stroke (when the slider is stationary). This obtains the actual response delay of the current hydraulic system and immediately uses it for feedforward prediction in this stroke. Correction can be completed in the first stroke after a sudden disturbance (such as a sudden rise in oil temperature or valve core jamming), avoiding continuous scrap. This active perturbation has been experimentally verified to not affect the sheet metal forming quality. Combined with the inflection point region prediction strategy switching, when there is an inflection point in the displacement-blank holder force mapping curve, time integration prediction replaces first-order displacement prediction, suppressing pressure spikes or dips at the inflection point, making feedforward control applicable to complex curve conditions.

[0019] 3. By employing a scanning-type trial punching method, the rate of change in sheet metal edge displacement is used as an online criterion for wrinkling / tearing, allowing the optimal peak value of the blank holder force to be determined within a single trial punch stroke. Compared to existing technologies that rely on multiple mold trials or finite element simulations, this significantly shortens the parameter debugging time after new workpieces are put into production or after mold changes.

[0020] 4. By using a non-contact thickness measurement and blank holder force correction model, the sheet metal thickness deviation is automatically measured before stamping, and the spatially distributed deviation is mapped to the correction coefficient of the target pressure of each blank holder cylinder. Compared with existing technologies that rely solely on manual adjustment based on experience or real-time detection during stamping, this method achieves pre-compensation for thickness fluctuations, avoiding batch defects caused by batch thickness differences.

[0021] 5. A method for estimating the friction coefficient by measuring pressure overshoot and displacement fluctuation amplitude utilizes the hydraulic press's existing sensor signals to obtain lubrication status online, eliminating the need for additional hardware. The estimated friction coefficient is used to dynamically compensate for the target value of the blank holder force, enabling the blank holder force to adapt to changes in lubrication conditions and avoiding fluctuations in forming quality caused by friction coefficient fluctuations.

[0022] 6. By combining historical operating condition retrieval with linear fitting of two trial strokes, the optimal parameters can be extrapolated using the closest historical parameters as a starting point and the data from two trial strokes. Combined with nonlinear region assessment and adaptive strategy switching, when a nonlinear risk is detected in the current sampling area, a third trial stroke is automatically added and quadratic curve fitting is used; or a conservative strategy is adopted and online fine-tuning is enabled in the strongly nonlinear region. Compared with existing technologies that rely on fixed-number linear extrapolation or large amounts of experimental data to establish response surfaces, this method achieves reliable optimization of the nonlinear quality index function with very few trial strokes, reducing the trial cost after mold change.

[0023] 7. By automatically calculating the cross-coupling gain coefficients Kp and Ks based on the workpiece stiffness characteristic value and area distribution characteristic value, and by performing online fine-tuning in mass production using the extreme value search method, the cross-coupling control can adapt to the mechanical properties of workpieces of different specifications. This avoids the debugging time and experience dependence caused by manual parameter tuning, and also compensates for parameter drift in the long-term operation of the hydraulic system.

[0024] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0025] Figure 1 This is a flowchart of a flexible stamping process applicable to multi-specification cargo box skins according to the present invention. Detailed Implementation

[0026] The present invention will be further described in detail below with reference to examples, so that those skilled in the art can implement it based on the description.

[0027] It should be understood that terms such as “having,” “comprising,” and “including” as used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

[0028] In a flexible stamping process applicable to multi-specification cargo box skins according to the present invention, the hydraulic press includes an integrally cast or welded machine body, a worktable fixed on the machine body, a slider that can move up and down along guide columns above the worktable, and four pressing cylinders fixed to the machine body and distributed at the four corners around the slider. Each pressing cylinder has a pilot-operated proportional relief valve connected to its oil inlet for independently adjusting the output pressure of that pressing cylinder. A pressure sensor (such as a diaphragm pressure sensor) is also installed in the oil circuit of each pressing cylinder to collect the actual pressure value inside the cylinder in real time. The controller of the hydraulic press can be a programmable logic controller (PLC, such as Siemens S7-1500 series) or an industrial control computer. This controller receives signals from each pressure sensor through an analog input module and sends control commands to each proportional relief valve through an analog output module. The controller's memory pre-stores the opening delay time of the proportional relief valve corresponding to each pressing cylinder (e.g., 50 milliseconds as measured by factory calibration) and the filling time of the pressure oil circuit to which the pressing cylinder belongs (e.g., 80 milliseconds calculated or measured based on the pipe length and inner diameter). For cargo box skin workpieces of different specifications, process engineers can input the target pressure values ​​of each pressing cylinder through the human-machine interface. These target pressure values ​​can be unequal to accommodate the pressing requirements of asymmetrical workpieces.

[0029] During the closing phase of the pressure ring, the controller first executes step one: Based on the target pressure value of each pressure cylinder, the opening delay time of the proportional relief valve for that cylinder, and the filling time of the pressure oil circuit for that cylinder, the controller calculates the estimated pressure build-up time for each cylinder. Specifically, the controller multiplies the target pressure value by a preset pressure-time conversion coefficient to obtain the basic pressure build-up time, and then adds the opening delay time and filling time to form the estimated pressure build-up time. For example, for the left front pressure cylinder, if its target pressure is 5 MPa, the opening delay time is 50 ms, and the filling time is 80 ms, then the estimated pressure build-up time can be calculated as 150 ms; while for the right rear pressure cylinder, if its target pressure is 8 MPa, its estimated pressure build-up time may reach 200 ms. Next, step two is executed: The controller aims to make the pressure build-up completion time of each pressure cylinder approach the same time, and determines the start-up delay time of each cylinder by comparing the differences between the four estimated values. For example, the cylinder with the longest pressure build-up time (200 milliseconds) starts without delay, while the cylinder with the shortest pressure build-up time (150 milliseconds) starts with a 50-millisecond delay. The controller, according to the calculated start-up delay, sequentially sends start commands to the proportional relief valves of each pressure cylinder via the digital output module, initiating pressure build-up in all four cylinders. During pressure build-up, step three is executed: the controller, with a sampling period of 1 to 5 milliseconds, collects the actual pressure values ​​of each pressure cylinder in real time via the analog input module, and calculates the tracking error (the difference between the actual and target values) for each cylinder, as well as the synchronization error between each cylinder and the other three cylinders (the actual pressure of this cylinder minus the actual pressures of the other cylinders). The controller multiplies the tracking error by a pre-set tracking error gain coefficient Kp, multiplies the sum of the synchronization errors by a synchronization error gain coefficient Ks, and then adds the two together to generate the control command for that cylinder. This control command is sent to the corresponding proportional relief valve via the analog output module, driving the valve core to adjust the opening, thereby dynamically adjusting the actual pressure of each cylinder. Once the actual pressure values ​​of each cylinder have reached their respective target pressure values ​​(with an allowable fluctuation of ±2%), step four is executed: The controller continues to collect the pressure signals of each cylinder, and after 100 milliseconds, takes the average value as the steady-state pressure value. The steady-state deviation between the steady-state pressure value and the target pressure value is calculated and stored in the controller's non-volatile memory. These steady-state deviations will be used in subsequent stamping strokes to correct the target pressure value of the corresponding pressure cylinder—that is, in the next stamping stroke, the controller subtracts this steady-state deviation from the original target pressure value to obtain the new command target value, thereby eliminating systematic static pressure deviations.

[0030] In a complete stamping cycle, after the blank holder ring closes, the slide begins its downward movement into the stamping stroke. At this time, the controller uses previously stored steady-state deviations to pre-correct the target pressure value, making the blank holder force control more precise. Due to the use of steps one through four, this implementation method can achieve synchronous pressure build-up in the four blank holder cylinders during the blank holder ring closure phase, avoiding localized pre-compression and pre-wrinkling defects in the sheet metal caused by differences in pressure build-up time between cylinders. Compared to traditional four-corner independent PID control (where each cylinder adjusts independently only based on its own tracking error), this implementation method achieves pre-synchronization through pressure build-up time estimation and start-up sequence planning, and achieves in-process coordination through cross-coupling control (considering both tracking and synchronization errors), thus fundamentally solving the technical problem of irreversible deformation of the sheet metal before the blank holder ring is fully closed.

[0031] Furthermore, in another embodiment, the hydraulic press used to perform the method described in this invention has a slider made of integral cast steel structure, which slides up and down along the guide posts at the four corners of the machine body. The lower surface of the slider is used to mount the upper mold (punch). A displacement sensor is fixedly mounted on the side or front of the slider. This displacement sensor can be a magnetic grating ruler or an optical grating ruler. Its reading head is fixedly connected to the slider, and the ruler body is attached to the machine body along the direction of movement. The measurement accuracy can reach ±0.01 mm. The central processing unit of the controller collects the pulse signal of the displacement sensor through a high-speed counting module and calculates the absolute displacement value of the slider in real time. At the same time, the controller obtains the speed value of the slider by performing differential operation on the displacement value (sampling once every 1 millisecond). The controller internally stores a "displacement-blank force" mapping relationship. This mapping relationship can be a lookup table with 1 mm intervals or a piecewise linear function. For example, the blank force increases linearly from 5 MPa to 8 MPa in the slider displacement range of 0 to 50 mm, remains at 8 MPa in the range of 50 to 120 mm, and decreases to 6 MPa in the range of 120 to 150 mm. The controller's memory also presets a delay duration τ, the initial value of which can be set to 100 milliseconds, representing the estimated pure time lag between the controller issuing a feedforward command and the actual pressure response of the pressure cylinder. Unlike traditional purely reactive control (i.e., the controller only issues pressure commands based on the current displacement), this implementation introduces a feedforward prediction mechanism to overcome the inherent response delay of the hydraulic system.

[0032] During the stamping stroke, the controller's control logic is as follows: First, step five is executed. Within each control cycle (e.g., 5 milliseconds), the controller reads the current slider displacement value S_current and velocity value v, and calculates the predicted displacement value S_pred = S_current + v × τ, where τ is the preset delay time (e.g., 100 milliseconds). Next, step six is ​​executed. The controller looks up the "displacement-blade pressure force" mapping relationship based on S_pred to obtain the corresponding feedforward blank pressure force target value P_feedforward, and sends this target value as a feedforward control command to the proportional relief valve of each blank pressure cylinder through the analog output module. Simultaneously, step seven is executed. The controller uses the current displacement S_current as the feedback displacement value, looks up the table to obtain the feedback blank pressure force target value P_feedback_target, then reads the actual pressure value P_actual fed back by the blank pressure cylinder pressure sensor, calculates the deviation e = P_feedback_target - P_actual, and generates the feedback control command U_feedback through PID calculation. Then, in step eight, the controller superimposes the feedforward control command and the feedback control command (e.g., adding the pressure target value or the valve drive voltage) to generate a composite control command U_composite, which drives the proportional relief valve to adjust the valve core opening through the analog output module. Steps five to eight are executed cyclically within each control cycle to ensure that the blank holder force can follow the change in slide displacement in real time throughout the entire stamping stroke. After each stamping stroke, step nine is executed: the controller collects the historical data stored in that stroke (displacement value and corresponding actual pressure value for each control cycle), compares the actual pressure value with the theoretical value in the preset "displacement-blank holder force" mapping relationship point by point, calculates the tracking error at each displacement point, and forms an error curve. The controller uses this error curve to correct the preset delay time τ—for example, if the tracking error shows a systematic lead (the actual pressure peak appears before the theoretical peak), the τ value is appropriately reduced; if it shows a systematic lag, the τ value is increased. The corrected τ value is stored in the controller for predicting the displacement of the next stamping stroke.

[0033] To illustrate with a specific example: A certain cargo box skin stamping process requires a slide descent speed of 80 mm / s. The hydraulic system response delay τ is approximately 100 milliseconds, corresponding to a hysteresis displacement of about 8 mm. Without feedforward compensation, the blank holder force only begins to rise when the slide reaches the drawbead inlet (displacement 50 mm), resulting in an actual peak blank holder force lag of 8 mm, causing uncontrolled sheet material flow and wrinkling. In this embodiment, when the actual slide displacement is 42 mm, the controller predicts that the displacement will reach 50 mm in 100 milliseconds, issuing a feedforward command to increase the blank holder force in advance, ensuring the blank holder cylinder reaches the required pressure precisely when the slide reaches 50 mm. Simultaneously, the feedback loop corrects prediction errors and external disturbances in real time. After the stroke ends, the controller detects that the actual pressure in the 40-60 mm displacement range is about 5% lower than the theoretical value, determining that τ is too small, and corrects τ from 100 milliseconds to 105 milliseconds, making the prediction for the next stroke more accurate. Through the aforementioned feedforward-feedback composite control and adaptive correction of τ, this embodiment effectively eliminates the pressure phase lag caused by hydraulic response delay, ensuring that the blank holder force remains precisely synchronized with the slide displacement during high-speed stamping (slide speed above 100 mm / s). Compared to existing technologies that rely solely on increasing PID response speed or decreasing slide speed, this embodiment significantly improves the accuracy of blank holder force follow-up control without sacrificing production cycle time, making it particularly suitable for multi-specification flexible stamping production such as cargo box skins that require precise control of the blank holder force curve.

[0034] In another embodiment, a hydraulic press for performing the method of the present invention has an aluminum alloy cargo box skin sheet (e.g., an aluminum-magnesium alloy sheet of grade 5A02, 1.2 mm thick) to be formed placed on its worktable. Displacement sensors are installed at multiple preset positions along the edge of the sheet. These sensors can be laser displacement sensors or wire-type displacement sensors, with their measuring heads aligned with the edge of the sheet to measure the displacement of the sheet edge sliding into the die during the stamping process. The installation positions of the sensors are determined according to the shape of the sheet profile; for example, one sensor is installed at the center of each of the four straight edges of the lower rear panel. Each sensor has a range of 0 to 50 mm and an accuracy of ±0.05 mm. The hydraulic press controller is a programmable logic controller (PLC) that acquires signals from each displacement sensor through an analog input module, with a sampling frequency set to 1 kHz. The controller internally stores a scan curve showing the continuous change of the peak value of the blank holder force from a first proportional value to a second proportional value. For example, for a cargo box skin with a large depth, the first proportion can be set to 10% (i.e., the blank holder force reaches its peak value in the first 10% of the total stamping stroke), and the second proportion can be set to 70% (i.e., the blank holder force reaches its peak value in 70% of the stroke). The scanning curve can be linear, meaning that the proportion at the peak moment increases uniformly from 10% to 70% as the slider displacement increases. Unlike the traditional method of relying on experience and repeated trial molding to determine the peak moment, this implementation method completes parameter scanning and optimal value locking through a single trial stamping stroke, significantly reducing the number of trial moldings.

[0035] During a trial punch stroke, the controller controls the target pressure value of each blank holder cylinder according to a preset scanning curve. Specifically, after the slider begins to descend, the controller reads the slider displacement sensor signal in real time and calculates the proportion of the current displacement to the total punch stroke (for example, if the total stroke is 200 mm and the current displacement is 100 mm, the proportion is 50%). Based on the scanning curve, the proportion of the blank holder force peak moment corresponding to this proportion is set to, for example, 40%. The controller then generates a curve showing the change of blank holder force with displacement: the blank holder force gradually increases from the initial value, reaches its peak at 40% of the total stroke, and then gradually decreases. Throughout the trial punch process, the controller collects the signals from each displacement sensor in real time at a 1-millisecond cycle and calculates the rate of change of the sheet metal edge sliding displacement measured by each sensor over time (i.e., the sliding speed). When any rate of change exceeds a preset threshold range, the controller records the proportion of the blank holder force peak moment corresponding to that moment as a critical proportion value. For example, the threshold range can be set from 0.5 mm / s to 5 mm / s: when the sliding speed is below 0.5 mm / s, it indicates that the flow at the edge of the sheet metal is obstructed, posing a risk of wrinkling; when it is above 5 mm / s, it indicates that the material flows in too quickly, posing a risk of tearing. In the trial punch, assuming that when the slider displacement ratio reaches 35%, the rate of change of the left edge displacement sensor drops to 0.4 mm / s, triggering the lower threshold, the controller records the peak time ratio corresponding to this moment as 35%, which is recorded as critical ratio value 1; when the slider displacement ratio reaches 52%, the rate of change of the right edge displacement sensor rises to 5.2 mm / s, triggering the upper threshold, and the peak time ratio is recorded as 52%, which is recorded as critical ratio value 2.

[0036] After the trial stamping stroke, the controller determines the optimal peak value of the blank holder force based on the recorded critical ratio values. One method is to take the median of all critical ratio values ​​as the optimal ratio. For example, if three critical ratio values ​​are recorded as 32%, 38%, and 51%, the optimal ratio can be any value between 38% and 51%, such as the median of 38% or the average of 40%. Another method is to take the midpoint between the first trigger threshold (wrinkling risk) and the last trigger threshold (tearing risk). For example, if the wrinkling risk is triggered at 35% and the tearing risk is triggered at 52%, the optimal ratio is (35% + 52%) / 2 = 43.5%. The controller stores this optimal ratio in non-volatile memory for setting the blank holder force curve in subsequent batch stamping strokes—that is, in batch production, the peak value of the blank holder force is fixed at this optimal ratio, so that the blank holder force curve matches the optimal forming window of the workpiece. Using the above method, this embodiment utilizes real-time monitoring of the edge displacement change rate during a single trial stamping stroke to automatically lock the peak moment ratio of the blank holder force that simultaneously avoids wrinkling and tearing. Compared to the existing technology that relies solely on operators observing the appearance of parts and repeatedly adjusting the mold, this method combines the parameter optimization process with the trial stamping process, significantly reducing the debugging time after new workpieces are put into production or after mold changes. It is particularly suitable for flexible stamping production lines with multiple varieties and specifications, such as cargo box skins.

[0037] In another embodiment, the hydraulic press used to perform the method described in this invention has at least one non-contact thickness sensor installed in its loading area or on the side of the worktable. This thickness sensor can be a laser triangulation displacement sensor (e.g., Keyence LK-G5000 series) or a confocal thickness sensor, with a measurement accuracy of ±0.005 mm. The sensor's installation position should avoid interference with moving parts such as molds or robotic arms; for example, it can be fixed to a gantry frame on the outside of the worktable and moved along a preset scanning path on the sheet metal surface by a servo motor. The hydraulic press's controller is a programmable logic controller (PLC) or an industrial control computer, communicating with the thickness sensor via Ethernet or RS-485 bus to receive thickness measurement data. The controller has pre-stored a nominal thickness value (e.g., 1.2 mm) and a blank holder force correction model. The correction model can be a linear proportional model, in the form of correction coefficient = 1 + k × (measured thickness - nominal thickness) / nominal thickness, where k is a sensitivity coefficient, which can be between 1.2 and 2.0 for aluminum alloy sheets. In this embodiment, k = 1.5 is preferred. Unlike the prior art where operators manually adjust the blank holder force based on experience and changes in sheet thickness, this embodiment achieves quantitative pre-correction of the blank holder force through automated thickness measurement and model calculation.

[0038] In practice, after the aluminum alloy cargo box skin sheet to be formed is placed at a predetermined position on the workbench (e.g., by positioning pins or a vision system), the controller first controls the thickness sensor to execute step f: the sensor moves along a preset scanning path on the sheet surface, which can cover the main forming area of ​​the sheet. For example, it collects one point every 50 mm along the length of the sheet and one point every 40 mm along the width, collecting approximately 20 to 30 measurement points. The sensor transmits the thickness value of each point back to the controller in real time, and the controller calculates the deviation between the thickness value of each measurement point and the nominal thickness value. For example, the nominal thickness of a batch of sheets is 1.2 mm, the actual measured thickness of the left front area is 1.23 mm (deviation +0.03 mm), and the thickness of the right rear area is 1.18 mm (deviation -0.02 mm). Then, step g is executed: the controller inputs the thickness deviation of each measurement point into a preset blank holder force correction model. This correction model can output correction coefficients for the target pressure values ​​of each blank holder cylinder based on the deviation distribution of multiple measurement points. A simple and effective model is as follows: Divide the blank holder plane into four quadrants, with each quadrant corresponding to a blank holder cylinder. Take the average thickness deviation Δh_avg of all measurement points within that quadrant, and then the correction coefficient for that cylinder = 1 + k × Δh_avg / h_nominal. For example, if the average deviation in the left front quadrant is +0.03 mm, then the correction coefficient = 1 + 1.5 × 0.03 / 1.2 = 1.0375; if the average deviation in the right rear quadrant is -0.02 mm, then the correction coefficient = 1 + 1.5 × (-0.02) / 1.2 = 0.975. For working conditions with more complex deviation distributions, the correction model can also use a weighted average or a nonlinear mapping table based on finite element pre-analysis. Finally, step h is executed: The controller multiplies the original target pressure value of each blank holder cylinder (e.g., the nominal pressure value set according to the workpiece specifications) by the corresponding correction coefficient to obtain the corrected target pressure value, which is stored in the controller for subsequent blank holder force control during the blank holder closing stage.

[0039] Here's a complete example: A batch of rear panel sheet metal has a nominal thickness of 1.2 mm. The original target pressure values ​​are 5.0 MPa for the left front cylinder, 5.2 MPa for the right front cylinder, 4.8 MPa for the left rear cylinder, and 5.0 MPa for the right rear cylinder. Thickness scanning reveals that the left front quadrant is on average 0.03 mm thicker, the right rear quadrant is on average 0.02 mm thinner, and the deviations in the other quadrants are within ±0.005 mm. The controller calculates corrected target pressure values ​​as follows: left front cylinder 5.0 × 1.0375 ≈ 5.19 MPa, right rear cylinder 5.0 × 0.975 ≈ 4.88 MPa, while the other two cylinders remain unchanged. In the subsequent blank holder closure stage, the controller uses these corrected target pressure values ​​to execute steps one through four, enabling the blank holder force to compensate for the actual thickness distribution of this batch of sheet metal. During mass production, if another batch of sheet metal with different thickness deviations is used, the thickness sensor will automatically rescan and update the correction coefficients without manual intervention. Through the above method, this embodiment achieves automatic detection of sheet metal thickness fluctuations and quantitative pre-compensation of blank holder force, avoiding wrinkling or tearing defects caused by thickness deviations. Compared with the existing technology that only passively adjusts after scrap occurs, this embodiment eliminates the impact of thickness fluctuations before stamping begins, improving the first-pass yield of mass production. It is especially suitable for multi-specification flexible stamping production of aluminum alloy cargo box skins and other products that are sensitive to thickness fluctuations.

[0040] In another embodiment, the hydraulic press used to perform the method of the present invention has the same configuration of slider, pressure cylinder, proportional relief valve, and pressure sensor as in the aforementioned embodiment. Furthermore, the displacement sensor of the slider can be a magnetic scale or optical scale, with its reading head fixedly connected to the slider and the scale body attached to the machine body along the direction of movement, achieving a measurement accuracy of ±0.01 mm. The controller is a programmable logic controller (PLC) or an industrial control computer, acquiring displacement signals through a high-speed counting module and acquiring pressure sensor signals from each pressure cylinder through an analog input module. The controller's memory pre-stores a friction coefficient estimation model, which can be a binary linear regression model or a pre-trained neural network. For example, for a certain type of aluminum alloy sheet, the friction coefficient estimation model can be expressed as μ_est = a×P_os+b×D_flu+c, where P_os is the pressure overshoot (in megapascals), D_flu is the displacement fluctuation amplitude (in millimeters), and a, b, and c are coefficients obtained through offline experimental fitting; in this embodiment, a is 0.15, b is 0.08, and c is 0.12. Meanwhile, the controller stores the nominal friction coefficient μ_nom (e.g., 0.15) and the calculation formula for the dynamic compensation coefficient, such as compensation coefficient = μ_est / μ_nom. Unlike the existing technology that passively adjusts the blank holder force based solely on the operator's observation of scrap, this embodiment utilizes the pressure overshoot and displacement fluctuation information naturally generated at the beginning of the stamping stroke to estimate the actual friction coefficient under the current lubrication state online and actively compensates for the target value of the blank holder force dynamically.

[0041] At the beginning of the stamping stroke, the controller executes step i: First, it controls each blank holder cylinder to apply blank holder force according to the preset initial target pressure value (e.g., the nominal blank holder force set according to the workpiece specifications). Simultaneously, it collects the pressure sensor signals of each blank holder cylinder and the displacement sensor signals of the slider in real time at a sampling frequency of 1 kHz. This initial stage typically corresponds to the slider displacement range of 0 to 5 mm. Within this range, the blank holder force remains constant, and the slider begins to descend but has not yet entered the complex deformation region. Next, step j is executed: The controller analyzes the collected pressure signals and extracts the pressure overshoot P_os. Specifically, when the actual pressure of the blank holder cylinder first reaches the initial target pressure value, the peak pressure at that moment is recorded, and the difference between this peak pressure and the initial target pressure value is calculated; this is the pressure overshoot. For example, if the initial target pressure is 5.0 MPa, and a peak of 5.3 MPa appears when the actual pressure first reaches 5.0 MPa, then the overshoot is 0.3 MPa. Simultaneously, the controller extracts the displacement fluctuation amplitude D_flu from the displacement sensor signal. During periods when the blank holder force remains constant (e.g., within the slider displacement range of 1 to 4 mm), the displacement signal is bandpass filtered (cutoff frequency 10 Hz to 200 Hz) to remove low-frequency motion components and high-frequency noise. The root mean square value of the filtered signal is then calculated as the displacement fluctuation amplitude. This fluctuation amplitude reflects the minute vibrations of the slider caused by changes in friction between the sheet metal and the die; a larger amplitude indicates a higher coefficient of friction. Then, step k is executed: the controller inputs the extracted pressure overshoot P_os and displacement fluctuation amplitude D_flu into the friction coefficient estimation model. The model outputs the estimated friction coefficient μ_est between the sheet metal and the die. For example, if P_os = 0.3 MPa and D_flu = 0.02 mm, substituting these values ​​into the model, μ_est = 0.15 × 0.3 + 0.08 × 0.02 + 0.12 = 0.045 + 0.0016 + 0.12 = 0.1666. Then, step l is executed: the controller calculates the dynamic compensation coefficient λ = μ_est / μ_nom = 0.1666 / 0.15 ≈ 1.11 based on the ratio of μ_est to the nominal friction coefficient μ_nom (0.15). Finally, step m is executed: the controller multiplies the current target pressure value (e.g., the corrected target pressure value) of each blank holder cylinder by the dynamic compensation coefficient λ = 1.11 to obtain the compensated target pressure value, which is used for blank holder force control in subsequent stamping stroke stages. If the compensation coefficient is less than 1, the blank holder force is reduced accordingly.

[0042] A specific example illustrates this: The lubricant coating on a batch of cargo box skin panels had a low viscosity, resulting in an actual friction coefficient higher than the nominal value. Without compensation, the blank holder force set according to the nominal friction coefficient would lead to excessive flow resistance in the sheet metal, easily causing tearing. At the beginning of the stamping stroke, the controller detected an increase in pressure overshoot from the usual 0.2 MPa to 0.35 MPa, and a displacement fluctuation amplitude from 0.015 mm to 0.025 mm. The model estimated μ_est = 0.175, and calculated a compensation coefficient λ = 1.17. The controller multiplied the target pressure values ​​of each blank holder cylinder by 1.17, correspondingly increasing the actual blank holder force to offset the additional flow resistance caused by the high friction coefficient. After compensation, the slider continued to descend to complete the entire stamping stroke, and the formed part met quality standards. Through this method, this implementation achieves online estimation of the friction coefficient and dynamic compensation of the blank holder force based on existing sensor signals, without requiring additional friction measurement hardware. Compared with existing technologies that rely on operator experience or offline friction coefficient measuring devices, this embodiment can automatically sense changes in lubrication status and adjust the blank holder force in real time at the beginning of each stamping stroke, effectively avoiding batch defects caused by fluctuations in the friction coefficient. It is especially suitable for flexible stamping production of various specifications with varying lubrication conditions.

[0043] In another embodiment, the hydraulic press used to execute the method described in this invention employs an industrial control computer or programmable logic controller (PLC) as its controller, equipped with a large-capacity non-volatile memory for storing a historical parameter database. This database records, in tabular form, the shape parameters of the blank holder force curve (such as reference pressure value, peak time ratio, and rising slope) for each workpiece specification (e.g., rear lower panel, rear pillar skin, etc.) in each production run, along with the corresponding sheet metal batch number, lubricant type (e.g., "WD-40" or "Shell S2"), ambient temperature (in degrees Celsius), and production date. The controller reads the workpiece specification identifier currently being changed into the mold via a human-machine interface (HMI) or RFID reader. An industrial camera or laser profilometer is mounted above the hydraulic press's worktable to measure the quality indicators of the stamped sheet metal, including the edge wrinkling height (obtained through 3D point cloud analysis, with an accuracy of ±0.05 mm), the critical fracture displacement (through tensile testing or visual inspection), and surface waviness (through structured light scanning). Quality indicators can be measured at the hydraulic press unloading station, where a robot delivers the stamped parts to the measurement station, or directly by an online inspection system. Unlike traditional methods that rely on operator visual observation and repeated trial molding, this implementation achieves rapid and automatic optimization of blank holder force parameters after mold change through historical data retrieval and linear fitting of two trial stamping data.

[0044] After mold change and before batch stamping, the controller executes step n: It reads the electronic tag on the mold using RFID to obtain the specification identifier of the currently inserted workpiece (e.g., "R101-Front Door Anti-collision Beam Assembly"). Then, based on this identifier, it retrieves all historical blank holder force parameter records corresponding to this specification from the historical parameter database. These records are arranged in reverse chronological order by production date. Each record contains blank holder force curve shape parameters (e.g., base pressure value P_base=4.5 MPa, peak time ratio α=42%), sheet metal batch number (e.g., "B20231201"), lubricant type (e.g., "Shell S2"), and ambient temperature (25℃). Next, step o is executed: The controller compares the current operating conditions (e.g., current sheet metal batch number "B20240315", lubricant type "Shell S2", ambient temperature 26℃) with the operating conditions in the historical records. It calculates the similarity using weighted Euclidean distance and uses the blank holder force curve shape parameter from the historical record closest to the current operating conditions as the initial blank holder force parameter. For example, a record with sheet metal batch number "B20231201", lubricant type "Shell S2", and ambient temperature of 25°C is retrieved from the historical records. Its reference pressure value is 4.5 MPa and peak value ratio is 42%, which is closest to the current working conditions. The controller loads this parameter as the blank holder force parameter for the first trial punch. Subsequently, the first trial punch stroke is executed. After the punching is completed, the workpiece is automatically transported to the inspection station. Step p is executed: The controller uses a laser profilometer to measure the edge wrinkling height of the formed sheet metal (e.g., the measured wrinkling height is 0.8 mm, the preset target value is ≤0.3 mm), and calculates the deviation as +0.5 mm (exceeding the standard). At the same time, the critical displacement for breakage is measured (e.g., the measured value is 2.1 mm, the target value is ≥1.8 mm, which is qualified) and the surface waviness (e.g., 0.12 mm, the target value is ≤0.15 mm, which is qualified). Since the wrinkling height exceeds the standard, the blank holder force is determined to be too small. Then, step q is executed: The controller calculates the correction direction and first step length of the blank holder force curve shape parameter according to the deviation and the preset adjustment rules. The adjustment rule could be: if the wrinkling height exceeds the standard, increase the reference pressure value, taking 10% of the current value for the first step, i.e., increasing it from 4.5 MPa to 4.95 MPa by 0.45 MPa; the peak value ratio can remain unchanged or be slightly adjusted. The controller applies the first corrected blank holder force parameters (reference pressure 4.95 MPa, peak value ratio 42%) and executes the second trial stroke.

[0045] After the second trial stroke, the quality index values ​​are measured again. Assume that after the second trial stroke, the wrinkling height decreases to 0.35 mm (still slightly higher than 0.3 mm), the critical fracture displacement is 1.9 mm (qualified), and the surface waviness is 0.13 mm (qualified). Step r: The controller combines the parameters from the first trial stroke (x1 = 4.5 MPa, y1 = 0.8 mm) and the parameters from the second trial stroke (x2 = 4.95 MPa, y2 = 0.35 mm) to fit a local linear relationship between the quality index (wrinkling height) and the blank holder force parameter (reference pressure). The fitting formula is y = a·x + b. The slope is determined by two points: slope a = (0.35 - 0.8) / (4.95 - 4.5) = -0.45 / 0.45 = -1.0, and intercept b = 0.8 - (-1.0) × 4.5 = 5.3, i.e., y = -x + 5.3. Then, step s is executed: based on the linear relationship, the optimal blank holder force parameter to achieve the preset target value (y_target = 0.3 mm) is calculated: x_opt = 5.3 - 0.3 = 5.0 MPa. The controller uses 5.0 MPa as the optimal blank holder force parameter for the current batch production and stores this optimal parameter and the current working conditions (sheet batch number, lubricant type, ambient temperature) in the historical parameter database for subsequent retrieval under similar working conditions. Through the above method, this embodiment utilizes historical data retrieval to provide reasonable initial parameters, and combines linear fitting of two trial stamping data to quickly converge to the optimal value, reducing the number of parameter adjustments after mold change from the traditional 5-10 times to 2 times. Compared with the existing technology that relies on repeated trial stamping based on manual experience or relies on a large number of simulation calculations, this embodiment significantly reduces the cost and time of trial stamping while ensuring optimization accuracy, and is especially suitable for flexible stamping production lines with multiple varieties and specifications, such as cargo box skins.

[0046] In another embodiment, the hydraulic press used to perform the method described in this invention includes a slider, four corner-distributed pressure cylinders, a proportional relief valve, a pressure sensor, and a controller. The controller is an industrial control computer (or a high-performance programmable logic controller) equipped with a large-capacity non-volatile memory for storing a workpiece parameter library, mapping parameters, and historical control data. For each specification of cargo box skin workpiece (e.g., rear lower panel, rear pillar skin), the stiffness characteristic value and area distribution characteristic value of the workpiece are obtained through offline calibration before initial production. The stiffness characteristic value can be obtained through finite element simulation or uniaxial compression experiments, defined as the elastic deformation displacement of the sheet metal under a unit pressure force, in millimeters per megapascal. The area distribution characteristic value quantifies the asymmetry of the workpiece by projecting the workpiece contour onto the pressure ring plane and calculating the variance of the ratio of the projected area in the four quadrants to the total area. For example, for workpieces with good symmetry, the area distribution characteristic value is close to 0; for severely asymmetrical workpieces (such as the rear pillar skin), this characteristic value can reach 0.15. The controller pre-establishes a first mapping relationship between the tracking error gain coefficient Kp and the stiffness eigenvalue and area distribution eigenvalue, and a second mapping relationship between the synchronization error gain coefficient Ks and the area distribution eigenvalue. The mapping relationships can employ a linear regression model, for example, Kp = 2.5 × stiffness eigenvalue + 0.8 × area distribution eigenvalue + 0.2, Ks = 1.2 × area distribution eigenvalue + 0.5. The parameters of the mapping relationships are obtained through offline experimental fitting and stored in the controller. Unlike existing technologies that use fixed Kp and Ks parameters or rely on manual experience for tuning, this implementation achieves adaptive matching between the gain coefficient and workpiece characteristics.

[0047] After the mold change is completed, the controller executes step C: It obtains the specification identifier of the currently inserted workpiece (e.g., "lower rear panel") by reading the electronic tag on the mold via RFID or inputting it through the human-machine interface. Then, based on this identifier, it reads the corresponding stiffness characteristic value and area distribution characteristic value from the workpiece parameter library. For example, the stiffness characteristic value is read as 0.12 mm / MPa, and the area distribution characteristic value as 0.05. Next, it executes step D: The controller substitutes the read characteristic values ​​into the first and second mapping relationships to calculate the Kp and Ks values. According to the above model, Kp = 2.5 × 0.12 + 0.8 × 0.05 + 0.2 = 0.3 + 0.04 + 0.2 = 0.54, Ks = 1.2 × 0.05 + 0.5 = 0.06 + 0.5 = 0.56. The controller uses the calculated Kp=0.54 and Ks=0.56 as the tracking error gain coefficient and synchronization error gain coefficient in step three, respectively, for cross-coupling control during the blank holder closure stage. In subsequent batch stamping processes, the controller executes step E: after each stamping stroke, it collects the pressure overshoot (the difference between the peak value of the actual pressure exceeding the target value for the first time and the target value) and synchronization error convergence time (the time required from pressure build-up to all four cylinders reaching the target value). The controller compares the pressure overshoot of N consecutive strokes (e.g., N=3) with a preset first threshold (e.g., 0.2 MPa) and the synchronization error convergence time with a preset second threshold (e.g., 200 milliseconds). When the pressure overshoot of three consecutive stamping strokes exceeds 0.2 MPa, or the synchronization error convergence time exceeds 200 milliseconds, a parameter fine-tuning process is triggered. The fine-tuning process uses an extreme value search method: based on the current Kp and Ks values, the controller makes tentative adjustments in steps not exceeding 5% of the current values. For example, if the current Kp = 0.54, the controller first attempts to increase it by 2% (i.e., 0.0108) to 0.5508, observing the change in pressure overshoot in the next stroke. If the overshoot decreases, fine-tuning continues in the same direction; if the overshoot increases, fine-tuning occurs in the opposite direction. Each adjustment step is limited to within 5%, and the cumulative adjustment range does not exceed 20% of the initial value to ensure the stability of the fine-tuning process. After several strokes of trial and convergence, the controller updates the corrected Kp and Ks values ​​to the workpiece parameter library for subsequent production of workpieces of the same specification.

[0048] To illustrate with a specific example: During the production of a batch of rear lower panels, Kp=0.54 and Ks=0.56 were calculated using offline calibrated stiffness and area distribution characteristic values. After producing 50 pieces consecutively, the control performance deteriorated due to increased hydraulic oil temperature and wear of the proportional relief valve. The controller detected pressure overshoot of 0.25 MPa, 0.27 MPa, and 0.26 MPa for three consecutive strokes, all exceeding the 0.2 MPa threshold, triggering fine-tuning. The controller increased Kp by 2.5% from 0.54 to 0.5535, reducing the pressure overshoot to 0.22 MPa in the next stroke; it then increased it by another 2% to 0.5646, further reducing the pressure overshoot to 0.18 MPa, bringing it within the threshold. The controller stored the final optimized Kp=0.5646 back into the workpiece parameter library. When producing the same specifications of workpieces after the next mold change, the controller will directly call this optimized Kp value, without needing to re-tune from the initial value. Through the above method, this embodiment achieves offline pre-adaptation and online self-optimization of cross-coupled control parameters. Compared with the existing technology that uses fixed gain or relies on repeated manual trial and error, this embodiment significantly reduces the parameter debugging time after mold change and can adaptively compensate for parameter drift in the hydraulic system caused by temperature changes, equipment wear, etc., making it particularly suitable for frequent mold changes in flexible stamping production lines for multi-specification cargo box skins.

[0049] In another embodiment, the hydraulic press used to perform the method described in this invention includes a slider, a displacement sensor (magnetic or optical scale with an accuracy of ±0.01 mm), a blank holder cylinder, a proportional relief valve, a pressure sensor, and a controller (industrial control computer or programmable logic controller). The controller internally stores a preset "displacement-blade holder force" mapping curve, typically stored in a lookup table with 1 mm intervals. In the offline phase, the controller performs inflection point extraction: it numerically differentiates the mapping curve, calculates the slope (first derivative) at each displacement point, and marks the displacement point as an inflection point when the slope change between two adjacent points exceeds a preset threshold (e.g., the rate of change of slope is greater than 100%) or the sign of the slope changes. For piecewise linear mapping curves, the inflection point is the connection point between the linear segments. For example, if the mapping curve of a cargo box skin switches from a rapidly rising segment (slope 0.2 MPa / mm) to a gentle segment (slope 0.05 MPa / mm) at a displacement of 50 mm, then 50 mm is marked as an inflection point, with a displacement value S_k = 50 mm. The controller stores the displacement values ​​of each inflection point in non-volatile memory.

[0050] During the stamping stroke, the controller acquires the slider displacement value S_current and velocity value v_current in real time at a period of 1 to 5 milliseconds, and calculates the current acceleration a_current by dividing the difference between the two most recent velocity samples by the sampling interval. Simultaneously, it calculates the absolute difference ΔS = |S_current - S_k| between the current displacement and the displacement values ​​at each inflection point. A preset first threshold is set based on the prediction step size, for example, prediction step size L = v_current × τ, where τ is the preset delay time (e.g., 100 milliseconds). If the current velocity v_current = 80 mm / s, then L = 8 mm, and the first threshold can be 1.5 × L = 12 mm. When ΔS is less than the first threshold (e.g., 12 mm), the controller determines that the slider has entered the region near an inflection point.

[0051] Once the system enters the region near the inflection point, the controller switches the method for calculating the predicted displacement in step five from the first-order prediction model S_pred = S_current + v_current × τ to a linear prediction model based on the current velocity and acceleration: S_pred = S_current + v_current × τ + 0.5 × a_current × τ 2 This model assumes that the acceleration remains constant over the next τ time interval, allowing it to more accurately reflect the velocity change trend of the slider near the inflection point. For example, the slider may have already decelerated (a_current is negative) as it approaches the inflection point; the first-order prediction model would overestimate the displacement increment, while the model incorporating the acceleration term can compensate for the deceleration effect. The controller continues to use this linear prediction model in the region near the inflection point until the slider displacement crosses the inflection point value S_k (i.e., S_current ≥ S_k). After crossing the inflection point, the controller switches back to the first-order prediction model to reduce computational overhead.

[0052] After each stamping stroke, the controller performs a post-evaluation: it collects the tracking deviation between the actual blank holder force curve and the preset mapping curve in the area near the inflection point (e.g., a 10 mm interval before and after the inflection point), and calculates the root mean square error (RMS) within this area. When the RMS error exceeds a preset second threshold (e.g., 0.2 MPa), the current first threshold or τ value is deemed inappropriate. The controller uses a gradient descent method to correct the first threshold and τ: if the tracking deviation manifests as the actual pressure lagging behind the preset curve, τ is appropriately decreased; if a peak appears at the inflection point, the first threshold is appropriately increased (entering the inflection point region earlier). The corrected parameters are stored in the controller for use in subsequent stamping strokes.

[0053] To illustrate with a specific example: the mapping curve of a cargo box skin has an inflection point at a displacement of 50 mm, transitioning from an upward segment to a flat segment. The preset delay time τ = 100 milliseconds, the current slider speed v_current = 100 mm / s, and the acceleration a_current = -0.3 m / s². 2 (Decelerating), predicted step size L = 10 mm, first threshold set to 15 mm. When the slider displacement reaches 35 mm (ΔS = 15 mm), the controller determines that it has entered the inflection point proximity region and switches to the linear prediction model. Calculate the predicted displacement increment: v_current×τ = 100 mm / s × 0.1 s = 10 mm, 0.5 × a_current×τ 2 = 0.5×(-300 mm / s) 2 )×0.01 seconds 2 = -1.5 mm, total increment = 8.5 mm, predicted displacement S_pred = 35 + 8.5 = 43.5 mm. The first-order model would mistakenly calculate 45 mm. Using the linear prediction model, the feedforward command predicts a displacement of 43.5 mm, corresponding to a blank holder force still in the middle of the rising segment, avoiding the pressure spike caused by the first-order model skipping the inflection point and directly entering the flat segment. The slider continues to descend, and when the displacement exceeds 50 mm, the controller switches back to the first-order prediction model. After the stroke ends, the controller analyzes the deviation between the actual blank holder force in the inflection point area and the preset curve, finding the deviation to be within 0.1 MPa, meeting the requirements, and requiring no parameter adjustment. If, during a production run, the hydraulic oil temperature changes, causing τ to actually become 120 milliseconds, resulting in a 0.3 MPa spike at the inflection point, the controller corrects τ from 100 milliseconds to 105 milliseconds and adjusts the first threshold from 15 mm to 18 mm, making the prediction for the next stroke smoother. By employing the above method, this embodiment effectively suppresses predicted overshoot at the inflection point of the blank holder force curve, avoiding forming defects caused by pressure spikes or dips. Compared with the passive approach in the prior art that relies solely on increasing feedback gain or reducing slider speed, this embodiment adaptively switches between curve shape recognition and an acceleration-compensated prediction model, ensuring smooth blank holder force control while maintaining high stamping speed. It is particularly suitable for complex cargo box skin stamping processes where the mapping curve has multiple inflection points.

[0054] In another embodiment, the hydraulic press used to perform the method described in this invention includes a slider, a displacement sensor (magnetic or optical scale with an accuracy of ±0.01 mm), a blank holder cylinder, a proportional relief valve, a pressure sensor, and a controller (industrial control computer or programmable logic controller). The controller internally presets a curve showing the change in blank holder force command and an initial preset delay time τ (e.g., 100 milliseconds). Unlike the prior art method of indirectly correcting τ through tracking errors over multiple strokes, this embodiment actively applies a very small pressure step before the start of each stamping stroke (when the slider is stationary), quickly identifying the response delay of the current hydraulic system online, and immediately using the identification result within the stroke, thereby achieving a single-stroke response to sudden disturbances.

[0055] After the pressure rings have closed, the actual pressure values ​​of each pressure cylinder have reached their respective target pressure values, and before the slider begins to descend, the controller executes step I: simultaneously applying a step command with an amplitude not exceeding 5% of the current steady-state target pressure value to the proportional relief valves of each pressure cylinder. For example, if the current steady-state target pressure is 5.0 MPa, then a step of +0.25 MPa (or -0.25 MPa) is applied. This amplitude has been verified by finite element simulation and experiments to not cause detectable deformation or defects to the already compressed sheet metal. This step command lasts for approximately 100 to 200 milliseconds, forming a clear active perturbation identification window.

[0056] Within the identification window, the controller executes step II: recording the time t0 when the step command is issued (the time when the controller outputs the command), and simultaneously acquiring the pressure sensor signal of the pressure cylinder in real time at a sampling frequency of 1 kHz. When the actual pressure value first reaches a preset percentage of the target value of the step command (for example, for an approximate first-order system, 63.2% corresponds to the time constant), the time t1 is recorded. The preset percentage can be selected as 63.2%, at which point the response time constant of the first-order system is τ_est = t1 - t0 - Δt_model, where Δt_model is a preset correction term used to compensate for the pure electrical delay and valve core start-up time between the issuance of the command and the actual start of pressure rise (for example, Δt_model = 10 milliseconds measured through offline testing). Assuming t1 - t0 = 120 milliseconds and Δt_model = 10 milliseconds are measured, then τ_est = 110 milliseconds.

[0057] After obtaining τ_est, the controller immediately executes step III: restoring the pressure of each pressure cylinder to the steady-state target pressure value before the step jump (usually requiring 20 to 50 milliseconds), and then controlling the slider to begin its descent. Step IV: replacing the preset delay duration τ in step V (originally 100 milliseconds) with the identified τ_est = 110 milliseconds, used for calculating the predicted displacement value of this stamping stroke. This means that from the first control cycle after the slider begins its descent, the predicted displacement calculation formula S_pred = S_current + v × τ will use the new τ value of 110 milliseconds, enabling the feedforward command to reflect the current actual response speed of the hydraulic system in a timely manner. Simultaneously, step V is executed: the controller stores the identified τ_est = 110 milliseconds in the historical data sequence in step nine, used for weighted averaging updates of the preset delay duration under the current working conditions for this specification of workpiece. For example, if the historical data sequence already contains values ​​of 100 milliseconds, 102 milliseconds, and 101 milliseconds, and 110 milliseconds is added this time, an exponentially weighted moving average (weight α=0.3) will be used for the update: New τ = 0.3×110 + 0.7×(previous average approximately 101 milliseconds) ≈103.7 milliseconds. The updated τ will be used as the default preset delay time for the next stamping stroke.

[0058] To illustrate with a specific example: During the continuous production of an aluminum alloy cargo box skin, a sudden failure occurred in the hydraulic oil cooling system, causing the oil temperature to rise from 40°C to 60°C within 30 seconds. This resulted in the proportional relief valve response delay increasing from 100 milliseconds to 130 milliseconds. Before the start of the first stamping stroke after the failure, after the blank holder closed, the controller actively applied a step of +0.25 MPa, measuring τ_est = 128 milliseconds. Subsequently, pressure was restored, and the slider began to descend. Throughout the stamping process, the feedforward prediction used a τ value of 128 milliseconds, and the blank holder force curve showed no significant deviation. After the stroke ended, the controller stored 128 milliseconds in the historical sequence and updated the preset τ to approximately 125 milliseconds. Without this implementation method, the slow correction in step nine would require multiple strokes to gradually adjust τ to 125 milliseconds. During this period, the blank holder force control accuracy would deteriorate for several consecutive strokes, potentially resulting in defective products. Through the above method, this implementation method achieves rapid response within a single stroke to sudden disturbances in the hydraulic system, effectively avoiding feedforward command interference caused by τ mismatch. Compared with the existing technology that relies solely on the cumulative error after the stroke for slow correction, this implementation actively acquires the direct measurement value of the current system response delay before the start of each stamping stroke, which significantly improves the adaptability of feedforward control to changes in operating conditions. Moreover, the amplitude of the active micro-disturbance is extremely small (not exceeding 5% of the steady-state pressure), and experimental verification has shown that it will not affect the sheet metal forming quality. It is especially suitable for multi-specification flexible stamping production lines with large fluctuations in hydraulic oil temperature, aging equipment, or occasional faults.

[0059] In another embodiment, the hydraulic press used to execute the method described in this invention includes a controller (industrial control computer or programmable logic controller), a human-machine interface, a historical parameter database, a quality index detection device (such as a laser profilometer or industrial camera), and related actuators. The controller internally stores a reference rate of change δ0 corresponding to each workpiece specification. This δ0 can be obtained through offline analysis of typical curves showing the change of quality index with blank holder force parameters in historical production data, or through finite element simulation pre-analysis. For example, for the lower rear panel workpiece, near the optimal blank holder force parameter, the reference rate of change δ0 of the wrinkling height with the reference pressure can be -1.0 mm / MPa (i.e., for every 0.1 MPa increase in pressure, the wrinkling height decreases by approximately 0.1 mm). Simultaneously, the controller presets a first threshold T1 and a second threshold T2, for example, T1 = 0.2 × |δ0| = 0.2 mm / MPa, and T2 = 0.5 × |δ0| = 0.5 mm / MPa. Unlike existing technologies that use only two trial data for linear extrapolation without assessing nonlinear risks, this implementation calculates the rate of change per unit step size and compares it with a reference value to automatically determine whether the current sampling region is a linear region, and adaptively selects linear extrapolation, quadratic fitting, or a conservative strategy based on the determination result.

[0060] After the first and second trial punches are completed following the mold change, the controller executes step q1: The blank holder force parameter of the first trial punch is recorded as x1 (e.g., reference pressure 4.5 MPa), and the quality index value (wrinkling height) is recorded as y1 (0.8 mm). The parameter of the second trial punch is recorded as x2 (4.95 MPa), and the quality index value is recorded as y2 (0.35 mm). The change in quality index Δy = y2 - y1 = -0.45 mm is calculated, and the unit step change rate δ1 = Δy / |x2 - x1| = -0.45 / 0.45 = -1.0 mm / MPa. Next, step q2 is executed: The controller reads the reference change rate δ0 = -1.0 mm / MPa corresponding to the workpiece specification from the memory, calculates |δ1 - δ0| = 0, and compares it with T1 = 0.2 and T2 = 0.5. Since |δ1 - δ0| = 0 ≤ T1, it is determined that the current region is linear, and step q3 is executed: calculate the optimal blank holder force parameters according to steps r and s above, that is, obtain x_opt = 5.0 MPa through two-point linear fitting, without the need for additional trial punches. If, under another working condition, the first trial punch is x1 = 4.5 MPa, y1 = 0.8 mm; the second trial punch is x2 = 4.7 MPa, y2 = 0.5 mm, then Δy = -0.3 mm, δ1 = -0.3 / 0.2 = -1.5 mm / MPa, |δ1 - δ0| = 0.5, which is between T1 = 0.2 and T2 = 0.5, and y1 and y2 are both greater than the target value of 0.3 mm (located on the same side), then step q4 is executed: it is determined that the region is nonlinear, and linear extrapolation is stopped. Using the parameter x2=4.7 MPa corresponding to the better quality index between x1 and x2 (y2=0.5 mm is better than y1=0.8 mm) as the benchmark, the third step length Δx3 is calculated according to the preset step reduction coefficient (e.g., 0.4) in the direction of improving the quality index (increasing pressure). The result is 0.4 × |x2-x1| = 0.4 × 0.2 = 0.08 MPa, yielding x3=4.78 MPa. A third test stroke is performed, and y3=0.32 mm is measured. Then, step q5 is executed: the three points (4.5, 0.8), (4.7, 0.5), and (4.78, 0.32) are fitted into a quadratic curve y = a·x 2+ b·x + c, calculate the minimum point x_opt = -b / (2a). For example, if the fitting result is a=10, b=-90, c=202.5, then x_opt=4.5 MPa. In another extreme case, if the deviation between δ1 and δ0 exceeds T2 (e.g., δ1=-2.0, |δ1-δ0|=1.0>0.5), then execute step q6: determine that it is a strong nonlinear region or has exceeded the optimal interval. At this time, no third trial punch or linear extrapolation is performed. The parameter corresponding to the better quality index of x1 and x2 is directly used as the optimal blank holder force parameter (e.g., x2=4.7 MPa), and the quality monitoring and fine-tuning mechanism in steps p to s above is activated. In batch production, it is gradually fine-tuned to the optimal value.

[0061] Here's a complete example: For the first trial punch of a batch of post-column skin, the historical retrieval parameters x1 = 5.0 MPa and the wrinkling height y1 = 1.2 mm (target ≤ 0.4 mm) are used. For the second trial punch, the parameters are increased by 0.5 MPa to x2 = 5.5 MPa, and the wrinkling height y2 = 0.6 mm. The calculation is: δ1 = (0.6 - 1.2) / (5.5 - 5.0) = -0.6 / 0.5 = -1.2, the reference rate of change δ0 = -0.8, and |δ1 - δ0| = 0.4. If T1 = 0.16 and T2 = 0.4, then |δ1 - δ0| = 0.4 equals T2, indicating the product is on a nonlinear boundary. Since y1 and y2 are both greater than 0.4 and located on the same side of the target value, the controller executes step q4. Using the better value x2 = 5.5 MPa as a baseline, with a step size reduction factor of 0.5, the third step size is calculated to be 0.25 MPa, x3 = 5.75 MPa. The third trial stroke yields y3 = 0.35 mm (meeting the target). A quadratic curve is fitted to the three points (5.0, 1.2), (5.5, 0.6), and (5.75, 0.35), and the minimum point x_opt ≈ 5.7 MPa is found. The controller uses this value as the optimal parameter for mass production. Through this method, this implementation, requiring only two trial strokes, adaptively determines whether to add a third trial stroke by online assessment of nonlinear risk, avoiding parameter misleading caused by blind linear extrapolation. Compared with existing technologies that rely on linear fitting or large amounts of experimental data to establish response surface methods, this implementation method achieves accurate optimization of nonlinear quality index functions with very few trial stampings. It is especially suitable for flexible stamping production of various specifications, such as cargo box skins, where the forming quality is sensitive to process parameters and exhibits obvious nonlinear characteristics.

[0062] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. It can be applied to various fields suitable for the present invention. Further modifications can be readily implemented by those skilled in the art.

Claims

1. A flexible stamping process method suitable for multi-specification cargo box skins, comprising a hydraulic press, the hydraulic press having four edge-pressing cylinders distributed at the four corners and a controller for controlling the four edge-pressing cylinders, characterized in that, The method includes the following steps during the closing stage of the pressure ring: Step 1: Obtain the estimated pressure build-up time for each of the four pressure cylinders. The estimated pressure build-up time for each pressure cylinder is calculated based on the target pressure value of the pressure cylinder, the opening delay time of the proportional relief valve to which the pressure cylinder belongs, and the filling time of the pressure oil circuit to which the pressure cylinder belongs. Step 2: Based on the estimated pressure build-up time of each of the four pressure cylinders, and with the goal of making the pressure build-up completion time of each pressure cylinder approach the same time, determine the start-up delay time of each of the four pressure cylinders, and start the four pressure cylinders in sequence according to the start-up delay time. Step 3: During the pressure building process of each pressing cylinder, the actual pressure value of each pressing cylinder is collected in real time. The difference between the actual pressure value of each pressing cylinder and the target pressure value of the pressing cylinder is used as the tracking error, and the difference between the actual pressure value of each pressing cylinder and the actual pressure values ​​of the other three pressing cylinders is used as the synchronization error. Based on the tracking error and the synchronization error, the control command of the pressing cylinder is generated to drive the proportional relief valve to which the pressing cylinder belongs to adjust the valve core opening. Step 4: After the actual pressure values ​​of the four pressing cylinders have all reached their respective target pressure values, the steady-state pressure values ​​of each pressing cylinder are collected, the steady-state deviation between the steady-state pressure value and the target pressure value is calculated, and the steady-state deviation is stored in the controller for correction of the target pressure value of the corresponding pressing cylinder in the subsequent stamping stroke.

2. The method according to claim 1, characterized in that, The stamping stroke also includes the following steps: Step 5: Obtain the displacement and velocity values ​​of the hydraulic press slider in real time, and predict the slider's displacement value after a preset delay time based on the displacement and velocity values. Step 6: Calculate the corresponding target value of the feedforward blanking force based on the predicted displacement value and the preset displacement-blank force mapping relationship, and send a feedforward control command to the proportional overflow valve of the blanking cylinder according to the target value of the feedforward blanking force; Step 7: Using the current displacement value of the slider as the feedback displacement value, calculate the target value of the feedback blank holder force based on the feedback displacement value and the displacement-blade holder force mapping relationship, and generate a feedback control command based on the deviation between the target value of the feedback blank holder force and the actual pressure value of the blank holder cylinder. Step 8: Superimpose the feedforward control command and the feedback control command to generate a composite control command to drive the proportional relief valve to adjust the valve core opening; Step 9: After each stamping stroke, collect historical data on the change of the actual pressure value of the pressure cylinder with the displacement of the slider during that stroke. Compare the historical data with the displacement-pressure force mapping relationship, calculate the tracking error at each displacement point, and correct the preset delay time according to the tracking error for the calculation of the predicted displacement value of the next stamping stroke.

3. The method according to claim 1, characterized in that, The following steps are included before determining the target pressure values ​​for each of the four pressure cylinders: Step a: Install the aluminum alloy cargo box skin sheet to be formed on the worktable of the hydraulic press, and install displacement sensors at multiple preset positions on the edge of the sheet. The displacement sensors are used to measure the amount of displacement of the sheet edge sliding into the die during the stamping process. Step b: Perform a trial stroke. During the trial stroke, the controller controls the target pressure value of each blank holder cylinder according to the preset scanning curve. The scanning curve is the ratio of the peak blank holder force to the total stamping stroke, which changes continuously from the first ratio value to the second ratio value. Step c: During the test stroke, the displacement signals of each displacement sensor are collected in real time, and the rate of change of each displacement signal with time is calculated. When any rate of change exceeds the preset threshold range, the proportion of the peak value of the blank holder force at that moment is recorded as the critical proportion value. Step d: Determine the optimal peak time ratio of blank holder force based on the multiple critical ratio values ​​obtained from the records, and store the optimal ratio in the controller for setting the blank holder force curve in subsequent batch stamping strokes.

4. The method according to claim 1 or 3, characterized in that, The following steps are included before determining the target pressure values ​​for each of the four pressure cylinders: Step e: Install at least one non-contact thickness sensor in the loading area or on the worktable of the hydraulic press. The thickness sensor is used to measure the thickness value at each measuring point on the sheet metal. Step f: After placing the sheet material to be formed at the predetermined position on the worktable, control the thickness sensor to move along the preset scanning path on the surface of the sheet material, collect the thickness values ​​of multiple measurement points on the sheet material, and calculate the thickness deviation between the thickness value of each measurement point and the nominal thickness value. Step g: Input the thickness deviation into the preset blank holder force correction model. The correction model outputs the correction coefficient of the target pressure value of each blank holder cylinder according to the correspondence between the thickness deviation and the blank holder force correction coefficient. Step h: Multiply the original target pressure value of each pressure cylinder by the corresponding correction coefficient to obtain the corrected target pressure value, which is used for pressure force control during the subsequent pressure ring closing stage.

5. The method according to claim 1, characterized in that, The stamping stroke also includes the following steps: Step i: At the beginning of the stamping stroke, control each blank holder cylinder to apply blank holder force according to the preset initial target pressure value, and at the same time collect the pressure sensor signal of each blank holder cylinder and the displacement sensor signal of the slider in real time. Step j: Extract the pressure overshoot from the pressure sensor signal. The pressure overshoot is the difference between the peak pressure when the actual pressure of the pressure cylinder first reaches the initial target pressure value and the initial target pressure value. Extract the displacement fluctuation amplitude from the displacement sensor signal. The displacement fluctuation amplitude is the root mean square value of the high-frequency fluctuation component of the displacement signal of the slider while the pressure force remains constant. Step k: Input the pressure overshoot and displacement fluctuation amplitude into the preset friction coefficient estimation model, and the friction coefficient estimation model outputs the estimated value of the friction coefficient between the current sheet and the mold; Step 1: Calculate the dynamic compensation coefficient of the target pressure value of each pressure cylinder based on the ratio of the estimated friction coefficient to the nominal friction coefficient; Step m: Multiply the current target pressure value of each blank holder cylinder by the dynamic compensation coefficient to obtain the compensated target pressure value, which is used for blank holder force control in the subsequent stamping stroke stage.

6. The method according to claim 1, characterized in that, The following steps are included after mold changing and before batch stamping: Step n: Read the current workpiece specification identifier from the controller, and retrieve the historical blank holder force parameter record corresponding to the specification from the historical parameter database according to the workpiece specification identifier. The historical parameter record shall at least include the blank holder force curve shape parameter, sheet batch number, lubricant type and ambient temperature. Step o: Take the blank holder force curve shape parameter from the historical parameter record that is closest to the current sheet metal batch number, lubricant type, and ambient temperature as the initial blank holder force parameter, and execute the first trial stroke; Step p: After the first trial stroke, measure the quality index values ​​of the formed sheet metal. The quality index includes at least one of the following: edge wrinkling height, critical displacement for breakage, or surface waviness. Calculate the deviation between the quality index values ​​and the preset target values. Step q: Calculate the correction direction and first step length of the blank holder force curve shape parameters according to the deviation and the preset adjustment rules to obtain the blank holder force parameters after the first correction, and then execute the second trial stroke; Step r: After the second trial punch stroke is completed, measure the quality index value of the formed sheet again. Combine the quality index values ​​of the first and second trial punches and the corresponding blank holder force parameters to fit the local linear relationship between the quality index and the blank holder force parameters. Step s: Calculate the blank holder force parameter that makes the quality index value reach the preset target value based on the local linear relationship, and use it as the optimal blank holder force parameter for the current batch production. Then, store the optimal parameter and the current working conditions in the historical parameter database.

7. The method according to claim 1, characterized in that, The following steps are included before the closing stage of the blank holder: Step A: For each specification of cargo box skin workpiece, obtain the stiffness characteristic value and area distribution characteristic value of the workpiece through offline calibration, and store the stiffness characteristic value and area distribution characteristic value with the workpiece specification identifier in the workpiece parameter library of the controller. Step B: Establish a first mapping relationship between the tracking error gain coefficient Kp and the stiffness characteristic value and the area distribution characteristic value, and a second mapping relationship between the synchronization error gain coefficient Ks and the area distribution characteristic value, and store the parameters of the first mapping relationship and the second mapping relationship in the controller; Step C: After the mold change is completed, the controller obtains the specification identifier of the currently inserted workpiece and reads the corresponding stiffness characteristic value and area distribution characteristic value from the workpiece parameter library according to the specification identifier. Step D: Substitute the read stiffness characteristic value and area distribution characteristic value into the first mapping relationship to calculate the Kp value, and substitute the area distribution characteristic value into the second mapping relationship to calculate the Ks value. Use the calculated Kp value and Ks value as the tracking error gain coefficient and synchronization error gain coefficient in step three. Step E: During the batch stamping process, the controller collects the pressure overshoot and synchronization error convergence time of each stamping stroke after the stroke ends. When the pressure overshoot exceeds the preset first threshold or the synchronization error convergence time exceeds the preset second threshold for N consecutive stamping strokes, the parameter fine-tuning process is triggered. The extreme value search method is used to make tentative adjustments based on the current Kp and Ks values ​​with a step size not exceeding 5% of the current values. In subsequent strokes, the Kp and Ks values ​​are updated to the workpiece parameter library according to the adjusted control effect.

8. The method according to claim 2, characterized in that, The stamping stroke also includes the following steps: Offline extraction of inflection point displacement values ​​on the preset displacement-blade pressure force mapping curve. Inflection points are defined as points where the first derivative of the mapping curve is discontinuous or where the sign of the second derivative changes. The displacement values ​​of each inflection point are stored in the controller. During the stamping stroke, the slider displacement value S_current and the speed value v_current are acquired in real time. The displacement difference between the slider displacement value and the displacement value of each inflection point is calculated. When the displacement difference is less than a preset first threshold, it is determined that the slider has entered the inflection point proximity area. In the region near the inflection point, the method for calculating the predicted displacement value in step five is switched from a first-order prediction model based on the current displacement and current velocity to a linear prediction model based on the current velocity and current acceleration, i.e., the predicted displacement value S_pred = S_current + v_current × τ + 0.5 × a_current × τ 2 , where τ is the preset delay time, and a_current is the current acceleration of the slider, which is calculated by the controller by the difference between the two most recent velocity samples; After the slider passes the inflection point displacement value, the calculation method for the predicted displacement value is switched back to the first-order prediction model in step five. After each stamping stroke, the tracking deviation between the actual blank holder force curve and the preset displacement-blank holder force mapping curve in the area near the inflection point is collected. When the tracking deviation exceeds the preset second threshold, the values ​​of the first threshold and τ are adjusted for use in the calculation of the predicted displacement value of subsequent stamping strokes.

9. The method according to claim 2, characterized in that, The following steps are also included before the stamping stroke begins: Step 1: After the pressure ring is closed, the actual pressure value of each pressure cylinder reaches its target pressure value, and before the slider starts to descend, the controller simultaneously applies a step command with an amplitude not exceeding 5% of the current steady-state target pressure value to the proportional relief valve of each pressure cylinder, as an active perturbation identification window; Step II: Within the active perturbation identification window, record the time t0 when the step command is issued, and collect the actual pressure value of the pressure cylinder in real time. When the actual pressure value first reaches the preset percentage of the target value of the step command, record the time t1, and calculate the identification delay τ_est = t1 – t0 - Δt_model, where Δt_model is a preset correction term. Step 3: After completing Step 2, the controller restores the pressure of each pressure cylinder to the steady-state target pressure value before the step jump, and then controls the slider to start descending; Step IV: Replace the preset delay time in Step V with the τ_est identified in Step II, and use it to calculate the predicted displacement value for this stamping stroke; Step V: Store τ_est into the historical data sequence in Step 9 for weighted average updates of the preset delay duration.

10. The method according to claim 6, characterized in that, The following steps are also included between step q and step r: Step q1: Record the parameter value of the first trial run as x1 and the quality index value as y1, and the parameter value of the second trial run as x2 and the quality index value as y2. Calculate the change in quality index Δy = y2 - y1 and the rate of change per unit step δ1 = Δy / |x2 - x1|. Step q2: Read the reference change rate δ0 corresponding to the workpiece specification from the memory of the controller, calculate the comparison result of |δ1 - δ0| with the preset first threshold T1, and the comparison result of |δ1 - δ0| with the preset second threshold T2, where T2 > T1; Step q3: When |δ1 - δ0| ≤ T1, determine that the current region is linear, and calculate the optimal blank holder force parameters according to steps r and s; Step q4: When T1 < |δ1 - δ0| ≤ T2 and y1 and y2 are both located on the same side of the target quality index value, it is determined that the current region is nonlinear, and step r is stopped. Based on the parameter corresponding to the better quality index of x1 and x2, the third step length Δx3 is calculated in the direction of improving the quality index according to the preset step size reduction coefficient. The third trial is performed to obtain the third trial point (x3, y3). Step q5: Fit the three trial points (x1, y1), (x2, y2), and (x3, y3) into a quadratic curve y = a·x 2 + b·x+ c, calculate the minimum point x_opt = -b / (2a) of the quadratic curve, and use it as the optimal blank holder force parameter; Step q6: When |δ1 - δ0| > T2, it is determined that the current region is in a strong nonlinear region or has exceeded the optimal interval. The parameter corresponding to the better quality index between x1 and x2 is taken as the optimal blank holder force parameter, and the quality monitoring and fine-tuning mechanism in steps p to s is activated.

Images