Self-adjusting control method and system for bottom dead center position of servo stamping equipment

Abstract

The invention provides a bottom dead center position self-adjustment control method and system for servo stamping equipment, and the method employs a double-speed cooperation architecture of a field programmable gate array (FPGA) and a micro control unit (MCU): the FPGA carries out the high-frequency collection and preprocessing of position, direction, current, temperature, vibration and acceleration signals in a bottom dead center core region, and transmits the signals to the MCU at a low time delay; the MCU executes bottom dead center reference calibration and transmission chain back clearance recognition, constructs a comprehensive error in combination with Kalman filtering state estimation, solves the compensation amount in a rolling mode based on model predictive control (MPC), and writes the compensation amount into a servo driver fine adjustment motion curve through a bus in cooperation with dynamic dead zone control, so that the position of a bottom dead center is stably converged for a long time, and the stamping precision and consistency are improved.

Description

Technical Field

[0001] This invention relates to the field of servo stamping technology, specifically to a method and system for self-adjusting the bottom dead center position of a servo stamping equipment. Background Technology

[0002] In recent years, servo stamping equipment has been widely used in manufacturing scenarios with extremely high requirements for dimensional accuracy and consistency, such as precision electronic components, automotive safety system parts, and key aerospace structural parts. This places higher demands on the tolerance consistency, surface quality, and structural reliability of stamped parts.

[0003] The reciprocating motion of the slide in a servo stamping machine usually relies on a rigid transmission system such as a servo motor, reducer, and toggle lever. The theoretical bottom dead center is often determined by geometric dimensions and mechanism parameters during the design phase and is usually considered to be a fixed position. However, in actual mass production with high cycle time, the theoretical bottom dead center will fluctuate due to various dynamic factors, which directly restricts the stamping accuracy and stability.

[0004] Specifically: First, when the equipment is running continuously for a long time, the friction and heat generated by the servo motor and transmission system will cause thermal expansion of the machine body and key components. The accumulated thermal deformation is transmitted to the slider, causing the bottom dead center position to drift slowly but continuously relative to the cold machine reference, even reaching tens of micrometers. As a result, the mold that has been properly calibrated after startup will gradually show dimensional deviations and scrap after stable production.

[0005] Secondly, backlash in the transmission chain (such as gear meshing clearance, spherical bearing clearance, lead screw nut mating clearance, etc.) will introduce idle stroke during reverse movement, causing the slider response to lag and forming a direction-related lag deviation; moreover, the backlash will increase with wear and the direction of deviation will be random, making it difficult for traditional fixed compensation to adapt in the long term.

[0006] Third, the high-speed stamping process is accompanied by strong impact and vibration. Under high-frequency conditions, the vibration superimposed and interfered, causing the slider to be in a dynamic and unstable environment when it reaches the bottom dead center, introducing random instantaneous deviations.

[0007] Therefore, the actual bottom dead center of servo stamping equipment is closer to "dynamic parameters that change with working conditions, time and load". Its instability makes it difficult to effectively control the finished product error rate and ensure high consistency under mass production, thus creating a demand for intelligent bottom dead center control technology with dynamic real-time compensation capabilities.

[0008] The shortcomings of existing technology: 1. Treating the bottom dead center as a fixed value and relying on one-time calibration makes it difficult to cover dynamic drift: Slow drift caused by thermal deformation will accumulate during continuous production, and a single calibration / static setting cannot maintain accuracy in the long term.

[0009] 2. The directional hysteresis and random deviation caused by backlash are difficult to solve with fixed compensation: reverse motion generates idle stroke, and backlash increases with wear and is random in direction, making it difficult for fixed compensation or a single compensation coefficient to be effective continuously.

[0010] 3. The instantaneous random error introduced by high-speed impact vibration is difficult to suppress by low-frequency / single-variable control: Under high-frequency cycles, the vibration superposition interference causes instantaneous deviation at the bottom dead center. Traditional control strategies based on a single feedback quantity are difficult to balance vibration suppression and accuracy.

[0011] 4. Lack of high-resolution, high-frequency measurement and real-time processing capabilities in the "bottom dead center core region": To achieve micron or even submicron level control near the bottom dead center, higher resolution position detection and higher sampling frequency real-time acquisition and preprocessing capabilities are required; otherwise, it is difficult to capture key transient errors and vibration characteristics.

[0012] Therefore, existing technologies have shortcomings and need further improvement. Summary of the Invention

[0013] To address the problems existing in the prior art, this invention provides a method and system for self-adjusting the bottom dead center position of a servo stamping equipment.

[0014] To achieve the above objectives, the specific solution of the present invention is as follows: This invention provides a self-adjusting control method for the bottom dead center position of a servo stamping equipment, applicable to a servo stamping equipment with a slider and a servo motor drive chain. The method includes: S1, construct a dual-speed collaborative architecture, setting the FPGA unit as the high-frequency acquisition and preprocessing layer and the MCU unit as the control decision layer; S2, Bottom Dead Point Calibration: Control the slider to reach the bottom dead point, collect the position sensor output and determine the bottom dead point reference position S_0, and at the same time collect and store the initial temperature reference T_0; S3, backlash identification: The control slider performs a small reciprocating motion near the bottom dead center reference position S_0, and simultaneously collects the servo motor rotation angle signal and slider displacement signal, quantizes and stores the backlash parameter b of the servo motor transmission chain; S4, Multi-source data acquisition and preprocessing: In each stroke, the FPGA unit acquires multi-source data at high frequency, including position sensor signals, servo motor running direction signals, motor current signals, temperature signals and slider acceleration signals, and transmits the preprocessed data to the MCU unit. S5, State estimation: The MCU unit timestamps the received data and performs Kalman filtering to obtain state parameters for error calculation and prediction. S6, Comprehensive Error Calculation: The MCU unit calculates the basic position error ΔS=S_1-S_0 based on the real-time position S_1 and the bottom dead center reference position S_0, and performs directional correlation compensation on the backlash parameter b according to the running direction of the servo motor to obtain the backlash compensation amount ΔS_b; the MCU unit calculates the thermal drift compensation amount ΔS_T based on the temperature signal and the stamping frequency, and constructs the disturbance observation value d_hat based on the position measurement residual; the comprehensive error ΔS_total is obtained by superimposing the ΔS, ΔS_b, ΔS_T and d_hat. S7, MPC Multi-Step Prediction and Compensation Solution: Input ΔS_total and operating parameters (including temperature, stamping frequency, motor current and vibration amplitude) into the Model Predictive Control (MPC) module, and perform rolling optimization to obtain the optimal compensation sequence for multiple future strokes, and determine the compensation amount for the current stroke. S8, Dynamic Dead Zone Control: When |ΔS_total| is within the preset dynamic dead zone threshold, the compensation output is suppressed; when |ΔS_total| exceeds the dynamic dead zone threshold, the current stroke compensation amount is output. S9, Compensation command execution: The compensation amount is sent to the servo driver to fine-tune the motion curve of the servo motor and the slider, so that the bottom dead center position of the subsequent stroke converges to the bottom dead center reference position S_0.

[0015] Furthermore, in the bottom dead center calibration in step S2, the FPGA unit acquires the position sensor signal at a sampling frequency of 10kHz to 50kHz and transmits it to the MCU unit after multi-stage digital filtering. The MCU unit calculates the bottom dead center reference position S_0 based on the signal-displacement conversion relationship of the position sensor and locks and stores it.

[0016] Furthermore, in the backlash identification in step S3, the slider performs a micro-amplitude reciprocating trial motion within a range of ±5μm near the bottom dead center reference position S_0. The MCU unit performs differential operation on the pairwise data of "angle-displacement". When the change in angle reaches the preset threshold and the change in displacement does not exceed the preset threshold within the continuous sampling period, the corresponding displacement is determined to be the backlash parameter b of the transmission system of the servo motor drive chain. After multiple identifications and elimination of outliers, the average value is taken to form the final backlash parameter.

[0017] Furthermore, the Kalman filter is an extended Kalman filter (EKF) or an unscented Kalman filter (UKF), and the state parameters include displacement error, slider speed, slider acceleration, motor current, comprehensive temperature deviation, and backlash state variable.

[0018] Furthermore, the thermal drift compensation amount is obtained through online identification. The MCU unit updates the thermal drift model parameters based on the temperature signal and stamping frequency using recursive least squares (RLS), and incorporates the thermal drift compensation amount as a feedforward term and / or disturbance term into the calculation process of the comprehensive error ΔS_total.

[0019] Furthermore, the MPC multi-step prediction and compensation solution satisfies the constraints, which include the upper limit of servo motor current, the upper limit of single compensation amount, the upper limit of slider acceleration, and the upper limit of compensation amount change rate; and when the deviation between the actual error and the MPC prediction error meets the preset criteria, online model re-identification is triggered to update the prediction step size, weight matrix, and constraint boundaries.

[0020] Furthermore, it also includes a laser interferometric calibration step. The servo stamping equipment is equipped with a laser interferometric measurement module, which includes a laser interferometer and a reflective target set on the slider or a rigidly connected component to the slider. During bottom dead center calibration or equipment operation, when preset calibration trigger conditions are met, including mold change, temperature change relative to the initial temperature T_0 exceeding a threshold, or cumulative stroke count reaching a threshold, the slider is controlled to perform micro-displacement calibration scanning at least two points near the bottom dead center reference position S_0. At the same time, the magnetic scale displacement and laser interferometric displacement are collected and linear alignment fitting and error fitting are performed to obtain the magnetic scale scale scaling factor k, zero-point offset o, and installation pitch angle error θ. The k, o, and θ are written back to the position conversion model to correct the calculation of the real-time position S_1, and the corrected real-time position S_1 is used in the comprehensive error ΔS_total calculation and MPC multi-step prediction compensation solution to achieve online suppression of drift error and scaling error of the position measurement chain.

[0021] Furthermore, it also includes an inertial navigation fusion step, wherein the servo stamping device is equipped with an inertial measurement unit (IMU) on the slider, and the IMU includes an accelerometer and a gyroscope; The FPGA unit synchronously acquires and timestamps the IMU and magnetic scale signals before transmitting them to the MCU unit. The MCU unit executes a tightly coupled extended Kalman filter (EKF), using the IMU output as the state prediction input and the magnetic scale displacement as the measurement update input to obtain fused estimates of the slider's displacement, velocity, attitude angle, and sensor bias. Based on the attitude angle, the MCU unit calculates and compensates for the displacement projection error caused by the slider's slight tilt in real time. Simultaneously, based on the fused estimates and the position measurement residual, a disturbance observation value d_hat is constructed. The dynamic dead zone threshold is then nonlinearly and adaptively adjusted according to the vibration energy or acceleration peak measured by the IMU. This suppresses ineffective compensation output under high vibration conditions and improves compensation resolution under low vibration conditions, thereby enhancing the bottom dead center control's resistance to vibration interference and external disturbances.

[0022] This invention provides a bottom dead center position self-adjustment control system for a servo stamping equipment, used to implement the above method, including: The FPGA unit is used for high-frequency synchronous acquisition and preprocessing of multi-source sensor data, and timestamps the acquired data. The MCU unit is used to receive the multi-source data and perform bottom dead point calibration, back gap identification, state estimation, comprehensive error calculation, MPC multi-step prediction and compensation decision, dynamic dead zone control, and self-learning update of model parameters and back gap parameters. A position sensor is used to detect slider displacement and for bottom dead center calibration and real-time position measurement. Temperature sensors are used to detect the operating temperature of equipment; An inertial measurement unit (IMU) is used to acquire slider acceleration, vibration, and attitude signals. The laser interferometry module is used to perform laser interferometry calibration of the slider displacement. Servo driver, used to receive compensation instructions output by MCU unit and fine-tune the movement of servo motor and slider; The servo motor running direction detection module is used to obtain the running direction of the servo motor.

[0023] Furthermore, the position sensor is a magnetic scale, and when the slider enters the core region of the bottom dead center within ±10μm of the theoretical bottom dead center, the FPGA unit switches to a higher sampling frequency of 20kHz to 100kHz to improve the bottom dead center measurement resolution and vibration resistance. The MCU unit includes a self-learning trigger and execution module. The self-learning trigger conditions include: model parameter calibration trigger, backlash parameter update trigger, error-operating condition correlation model optimization trigger, and MPC model adaptive re-identification trigger. When any trigger condition is met, the backlash parameter, thermal drift model parameter, and MPC model parameter are automatically updated. The MCU unit writes compensation instructions to the servo driver through the fieldbus. The compensation instructions carry at least the backlash compensation amount, thermal drift compensation amount, and MPC multi-step prediction identifier (used to identify the prediction step number corresponding to the current compensation amount), so that the servo driver can precisely fine-tune the servo motor rotation angle of the next stroke according to the compensation instructions to change the slider motion curve and achieve precise stopping at the bottom dead center position.

[0024] The technical solution of this invention has the following beneficial effects: 1. Improved bottom dead center positioning accuracy and faster response: High-frequency sampling is triggered in the core area of ​​the bottom dead center (±10μm), combined with high-resolution magnetic scale and low-latency data transmission, making the measurement of the bottom dead center reference S_0 and the current position S_1 more accurate and the closed-loop control more timely, thereby achieving high-precision and stable tracking of the bottom dead center.

[0025] 2. Backlash direction-related errors are effectively eliminated: The backlash parameter b is obtained through an independent backlash identification process, and a direction-related backlash compensation correction ΔS_b=k_b·b is introduced during error calculation, where k_b is a preset backlash compensation coefficient (e.g., 1.1) to offset idle travel and reduce bottom dead center deviation and random fluctuations caused by reverse lag.

[0026] 3. Long-term accuracy drift caused by thermal drift is suppressed: A thermal drift model is constructed based on bearing temperature, die temperature and stamping frequency and ΔS_T is calculated. The model parameters are updated online using RLS, so that the system can correct drift in real time with temperature rise and changes in operating conditions, and maintain long-term stable accuracy.

[0027] 4. Enhanced resistance to vibration and external disturbances: The d_hat is constructed by a disturbance observer to characterize external disturbances such as mold impact and material hardness changes, and these disturbances are incorporated into the comprehensive error ΔS_total for unified compensation, thereby reducing the impact of instantaneous random deviations on the bottom dead center under high-speed stamping. Attached Figure Description

[0028] Figure 1 System architecture diagram; Figure 2 Control method flowchart; Figure 3 Schematic diagram of dual-speed collaborative architecture; Figure 4 Schematic diagram of the back gap identification process; Figure 5 Multi-source data acquisition and preprocessing flowchart; Figure 6 Flowchart of MPC multi-step prediction and compensation solution; Figure 7 Dynamic dead-time control logic diagram; Detailed Implementation The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, and not all of them.

[0029] Combined with appendix Figure 1-7 As shown, this invention provides a self-adjusting control method for the bottom dead center position of a servo stamping equipment, applicable to a servo stamping equipment with a slider and a servo motor drive chain. The method includes: S1, construct a dual-speed collaborative architecture, setting the FPGA unit as the high-frequency acquisition and preprocessing layer and the MCU unit as the control decision layer; S2, Bottom Dead Point Calibration: Control the slider to reach the bottom dead point, collect the output of the position sensor (magnetic scale) and determine the bottom dead point reference position S_0, and at the same time collect and store the initial temperature reference T_0 (including the initial temperature of the bearing and the initial temperature of the mold). S3, backlash identification: The control slider performs a small reciprocating motion near the bottom dead center reference position S_0, and simultaneously collects the servo motor rotation angle signal and slider displacement signal, quantizes and stores the backlash parameter b of the servo motor transmission chain; S4, Multi-source data acquisition and preprocessing: In each stroke, the FPGA unit acquires multi-source data at high frequency, including position sensor (magnetic scale) signals, servo motor running direction signals, motor current signals, temperature signals (bearing temperature and mold temperature) and slider acceleration signals, and transmits the preprocessed data to the MCU unit. S5, State estimation: The MCU unit timestamps the received data and performs Kalman filtering (EKF / UKF) to obtain state parameters for error calculation and prediction; S6, Comprehensive Error Calculation: The MCU unit calculates the basic position error ΔS=S_1-S_0 based on the real-time position S_1 and the bottom dead center reference position S_0, and performs directional correlation compensation on the backlash parameter b according to the running direction of the servo motor to obtain the backlash compensation amount ΔS_b; the MCU unit calculates the thermal drift compensation amount ΔS_T based on the temperature signal and the stamping frequency, and constructs the disturbance observation value d_hat based on the position measurement residual; the comprehensive error ΔS_total is obtained by superimposing the ΔS, ΔS_b, ΔS_T and d_hat. S7, MPC Multi-Step Prediction and Compensation Solution: Input ΔS_total and operating parameters (including temperature, stamping frequency, motor current and vibration amplitude) into the Model Predictive Control (MPC) module, and perform rolling optimization to obtain the optimal compensation sequence for multiple future strokes, and determine the compensation amount for the current stroke. S8, Dynamic Dead Zone Control: When the absolute value of |ΔS_total| is within the preset dynamic dead zone threshold, suppress compensation output; when the absolute value of |ΔS_total| exceeds the dynamic dead zone threshold, output the current stroke compensation amount. S9, Compensation command execution: The compensation amount is sent to the servo driver to fine-tune the motion curve of the servo motor and the slider, so that the bottom dead center position of the subsequent stroke converges to the bottom dead center reference position S_0.

[0030] In the bottom dead center calibration in step S2, the FPGA unit acquires the position sensor (magnetic scale) signal at a sampling frequency of 10kHz to 50kHz and transmits it to the MCU unit after multi-stage digital filtering. The MCU unit calculates the bottom dead center reference position S_0 based on the signal-displacement conversion relationship of the position sensor and locks and stores it.

[0031] In the backlash identification in step S3, the slider performs a micro-amplitude reciprocating trial motion within a range of ±5μm near the bottom dead center reference position S_0. The MCU unit performs differential operation on the pairwise data of "angle-displacement". When the change in angle reaches the preset threshold and the change in displacement does not exceed the preset threshold within the continuous sampling period, the corresponding displacement is determined to be the backlash parameter b of the transmission system of the servo motor drive chain. After multiple identifications and elimination of outliers, the average value is taken to form the final backlash parameter.

[0032] The Kalman filter is an extended Kalman filter (EKF) or an unscented Kalman filter (UKF), and the state parameters include displacement error, slider speed, slider acceleration, motor current, comprehensive temperature deviation, and backlash state variable.

[0033] The thermal drift compensation amount is obtained through online identification. The MCU unit updates the thermal drift model parameters based on the temperature signal (bearing temperature and mold temperature) and stamping frequency using recursive least squares (RLS), and incorporates the thermal drift compensation amount as a feedforward term and / or disturbance term into the calculation process of the comprehensive error ΔS_total.

[0034] The MPC multi-step prediction and compensation solution satisfies the constraints, which include the upper limit of servo motor current, the upper limit of single compensation amount, the upper limit of slider acceleration, and the upper limit of compensation amount change rate. When the absolute value of the difference between the actual error and the MPC prediction error meets the preset criteria (such as the deviation of 3 consecutive strokes exceeding 0.3μm), online model re-identification is triggered to update the prediction step size, weight matrix, and constraint boundary.

[0035] The process also includes a laser interferometric calibration step. The servo stamping equipment is equipped with a laser interferometric measurement module, which includes a laser interferometer and a reflective target mounted on the slider or a rigidly connected component to the slider. During bottom dead center calibration or equipment operation, when preset calibration trigger conditions are met, including mold change, temperature change relative to the initial temperature T_0 exceeding a threshold, or cumulative stroke count reaching a threshold, the slider is controlled to perform micro-displacement calibration scanning at least two points near the bottom dead center reference position S_0. Simultaneously, the magnetic scale displacement and laser interferometric displacement are collected and linear alignment fitting and error fitting are performed to obtain the magnetic scale scale scaling factor k, zero-point offset o, and installation pitch angle error θ. The k, o, and θ are written back to the position conversion model to correct the calculation of the real-time position S_1, and the corrected real-time position S_1 is used in the calculation of the comprehensive error ΔS_total and the MPC multi-step prediction compensation solution to achieve online suppression of drift error and scaling error of the position measurement chain.

[0036] It also includes an inertial navigation fusion step, wherein the servo stamping device is equipped with an inertial measurement unit (IMU) on the slider, and the IMU includes an accelerometer and a gyroscope; The FPGA unit synchronously acquires and timestamps the IMU and magnetic scale signals before transmitting them to the MCU unit. The MCU unit executes a tightly coupled extended Kalman filter (EKF), using the IMU output as the state prediction input and the magnetic scale displacement as the measurement update input. This yields fused estimates of the slider's displacement, velocity, attitude angles (roll angle φ, pitch angle θ), and sensor bias. Based on these attitude angles, the MCU unit calculates and compensates for displacement projection errors caused by slider micro-tilts in real time. Simultaneously, based on the fused estimates and position measurement residuals, a disturbance observation value d_hat is constructed. The dynamic dead zone threshold is then nonlinearly and adaptively adjusted according to the vibration energy or acceleration peak measured by the IMU. This suppresses ineffective compensation output under high vibration conditions and improves compensation resolution under low vibration conditions, thereby enhancing the bottom dead center control's resistance to vibration interference and external disturbances.

[0037] The present invention also provides a bottom dead center position self-adjustment control system for a servo stamping equipment, used in the above method, comprising: The FPGA unit is used for high-frequency synchronous acquisition and preprocessing of multi-source sensor data, and timestamps the acquired data. The MCU unit is used to receive the multi-source data and perform bottom dead point calibration, back gap identification, state estimation, comprehensive error calculation, MPC multi-step prediction and compensation decision, dynamic dead zone control, and self-learning update of model parameters and back gap parameters. The position sensor, specifically a magnetic scale, is used to detect slider displacement and for bottom dead center calibration and real-time position measurement. The laser interferometry module is used to perform laser interferometry calibration of slider displacement, including a laser interferometer and a reflective target (which can be a single target or a double target structure). Servo driver, used to receive compensation instructions output by MCU unit and fine-tune the movement of servo motor and slider; An inertial measurement unit (IMU) is used to acquire slider acceleration, vibration, and attitude signals; Servo motor running direction detection module, used to obtain the running direction of servo motor; Temperature sensors, including bearing temperature sensors and mold temperature sensors, are used to detect the operating temperature of equipment.

[0038] When the slider enters the core region of the bottom dead center (BDC) within ±10μm of the theoretical BDC, the FPGA unit switches to a higher sampling frequency of 20kHz to 100kHz (higher than the 10kHz to 50kHz used in the BDC calibration stage) to improve the BDC measurement resolution and vibration resistance. The MCU unit includes a self-learning trigger and execution module. The self-learning trigger conditions include: model parameter calibration trigger, backlash parameter update trigger, error-operating condition correlation model optimization trigger, and MPC model adaptive re-identification trigger. When any trigger condition is met, the backlash parameter, thermal drift model parameter, and MPC model parameter are automatically updated. The MCU unit writes compensation instructions to the servo driver via the fieldbus. The compensation instructions carry at least the backlash compensation amount, thermal drift compensation amount, and MPC multi-step prediction identifier (used to identify the prediction step number corresponding to the current compensation amount). This allows the servo driver to precisely fine-tune the servo motor angle of the next stroke according to the compensation instructions to change the slider motion curve and achieve precise stopping at the BDC position.

[0039] Working principle: The working principle of this invention (bottom dead center position closed loop "acquisition - estimation - prediction - compensation - self-learning") This invention is based on a dual-speed collaborative architecture of "FPGA high-frequency data acquisition layer + MCU control decision layer": the FPGA is responsible for high-precision sensor signals at a high frequency of 10kHz to 50kHz near the bottom dead center and preprocessing them, and then transmitting them to the MCU via SPI with a microsecond delay of ≤10μs; the MCU performs state estimation, comprehensive error calculation and MPC multi-step prediction compensation decision at 500Hz to 1000Hz, and then writes the compensation instructions to the servo driver via CANopen to realize the fine-tuning closed-loop control of the slider's bottom dead center.

[0040] 1) Initialization: Bottom dead center benchmark calibration + independent backlash identification (establishing "target" and "systematic error parameters") Bottom Dead Point Calibration: In cold operating mode, the FPGA acquires the magnetic scale reference potential value at 10kHz to 50kHz, and simultaneously acquires the initial bearing temperature T_bearing0 and the initial mold temperature T_die0, using these as the initial temperature reference T_0 = (T_bearing0, T_die0). After digital filtering, the data is transmitted to the MCU via SPI. The MCU calculates the reference bottom dead point position S_0 based on the "potential value - distance" conversion relationship and locks it as the target value.

[0041] Backlash identification: The MCU triggers the backlash identification process, controlling the slider to perform ±5μm micro-amplitude reciprocating probes near S_0; the FPGA synchronously collects the motor rotation angle and slider displacement and transmits them to the MCU. The MCU determines the comprehensive backlash b based on the section where "the rotation angle changes but the displacement does not respond", takes the average value multiple times and stores it in the parameter library as a key parameter for subsequent reverse motion compensation.

[0042] 2) Online operation: High-frequency acquisition and preprocessing of multi-source data (ensuring the "observability" of the lower dead zone); Core area encrypted sampling: The theoretical bottom dead center ±10μm is defined as the "bottom dead center core area". When the slider enters this area, the FPGA automatically switches to the highest sampling frequency (20kHz~100kHz) to perform high-frequency sampling of signals such as magnetic scale and acceleration to ensure the capture of instantaneous deviations under high-speed stamping.

[0043] Multi-source acquisition + structured transmission: The FPGA high-frequency acquisition of magnetic grating ruler potential value, running direction, motor current, bearing temperature T_bearing, mold temperature T_die, slider acceleration, etc., and transmits the data packet of "potential value + direction + current + temperature + acceleration" to the MCU with a delay of ≤10μs via SPI to ensure the real-time performance of control data.

[0044] Multi-rate Kalman filter state estimation: The MCU timestamps the high-frequency displacement and low-frequency operating condition signals, and uses EKF / UKF to estimate the state vector (error / velocity / acceleration / current / temperature deviation / backlash state, etc.), outputting noise-resistant observable state quantities, providing stable input for error calculation and prediction.

[0045] 3) Comprehensive error modeling: unify "position error + backlash + thermal drift + external disturbance" into a single error quantity; When each stroke reaches the bottom dead center, the MCU compares the current position S_1 calculated by the magnetic scale with the target S_0 to form the basic error ΔS, and further superimposes three types of key corrections to obtain the final comprehensive error ΔS_total: Backlash directional correction ΔS_b: Based on the motor's running direction, if there is a reverse segment (upward to downward), a backlash compensation correction amount ΔS_b=k_b·b (k_b is a preset backlash compensation coefficient, such as 1.1) is introduced. When moving in the same direction, ΔS_b=0, which is used to offset the lag deviation caused by "idle travel".

[0046] Thermal drift compensation ΔS_T: Based on T_bearing, T_die and stamping frequency f, the thermal drift model parameters are updated online using RLS, the displacement offset caused by thermal drift is calculated and incorporated into the control model as a feedforward / disturbance term.

[0047] Disturbance observation d_hat: A disturbance observer is constructed by the residual between the filtered displacement value and the predicted displacement value to characterize external disturbances such as mold impact and material hardness change and incorporate them into the compensation.

[0048] 4) MPC Multi-Step Prediction: Upgrades "current error" to "predictive forecast of error trend in the next 3-5 strokes"; The MCU constructs an "error-operating condition" correlation model based on historical data from nearly a hundred strokes. The prediction objective is to minimize the sum of squared comprehensive errors over the next 3 to 5 strokes. Constraints such as current, single compensation amount, and acceleration are set, and the optimal compensation amount sequence is obtained through rolling optimization (achieving lead control and avoiding the lag of traditional linear compensation).

[0049] The MCU inputs the real-time ΔS_total, operating conditions, and filter status values ​​into the MPC to calculate the compensation amount to be executed in the current stroke and the compensation trend in subsequent strokes, thus achieving "early prediction of error trends".

[0050] 5) Execution and stability: Dynamic dead zone + servo fine-tuning (both stable and accurate, avoiding jitter); Dynamic dead-zone logic: When |ΔS_total|≤0.5μm, the MCU does not write compensation instructions to the servo driver to avoid frequent oscillations caused by small errors; at the same time, it continues to run filtering and MPC prediction to monitor trends and ensure smooth switching between inside and outside the dead zone.

[0051] Dead zone out-of-zone compensation output: When |ΔS_total| exceeds the threshold, the MCU generates a control output (carrying ΔS_b, ΔS_T and multi-step prediction flags) based on the current stroke compensation amount output by the MPC, and writes it to the servo driver through CANopen; the servo driver then fine-tunes the servo motor rotation angle / motion curve accordingly, so that the slider returns to the target S_0 at the bottom dead center in the next stroke, thereby offsetting the deviation caused by backlash, thermal drift, vibration and external disturbance.

[0052] 6) Long-term accuracy maintenance: self-learning triggering and model updates (adapting to wear, mold changes and changes in operating conditions) The system continuously monitors the deviation between "actual error and prediction error" and automatically performs self-learning when the trigger conditions are met. This includes model parameter calibration, back gap re-identification, error correlation model optimization, and MPC online re-identification (which can be combined with CUSUM to detect error drift trends), ensuring that the control model is adaptively updated as equipment wears down, molds are changed, and materials change.

[0053] Meanwhile, after startup and mold change or after continuous operation for a certain period of time, the system automatically updates the back clearance b and thermal drift model parameters, and synchronously updates the MPC associated model to ensure the stability of the bottom dead center control accuracy under long-term production.

[0054] Example 1: The following embodiment describes a "self-adjusting control system / method for the bottom dead center position of a servo stamping equipment". In this embodiment, the servo stamping equipment includes a frame, a slide, a servo motor, a reducer, a transmission mechanism such as a toggle lever, and a servo driver; dynamic compensation control of the bottom dead center position is achieved through a dual-speed architecture of "FPGA high-frequency acquisition layer + MCU control decision layer".

[0055] I. Hardware Configuration and Installation Layout; Control hardware: The FPGA unit uses Artix-7 or equivalent devices for high-frequency acquisition and preprocessing; the MCU unit uses STM32H7 or equivalent devices for control calculation and decision output; the FPGA and MCU communicate bidirectionally via SPI (≥10Mbps) with a transmission delay ≤10μs.

[0056] Sensors and Interfaces: High-precision position sensors (specifically magnetic scales with a resolution ≤ 0.1μm) are installed on the side of the slider or on the mold, and their signals are connected to the FPGA high-speed sampling interface; a servo motor running direction detection module (sampling frequency ≥ 1kHz) is set at the servo motor encoder, and its signal is connected to the FPGA digital input; a bearing temperature sensor is set at the bearing, a mold temperature sensor is set on the mold surface, and a slider accelerometer is set on the slider, and its signal is connected to the FPGA analog acquisition interface; the MCU communicates with the servo driver via CANopen (500kbps~1Mbps) to write compensation instructions.

[0057] Definition of the core region of the bottom dead center: The range of ±10μm from the theoretical bottom dead center is defined as the core region of the bottom dead center; when the slider enters this region, the FPGA automatically switches to the highest sampling frequency (20kHz~100kHz) to perform high-frequency sampling of the magnetic scale and accelerometer.

[0058] II. Initialization: Bottom Dead Point Calibration and Back Gap Identification; Cold-run bottom dead center calibration (establishing reference S_0): When the equipment is in cold-run state, the control slider runs at low speed to the bottom dead center; the FPGA collects the reference potential value of the bottom dead center of the magnetic scale at 10kHz to 50kHz (preferably 20kHz to 40kHz), and at the same time collects the initial temperature of the bearing T_bearing0 and the initial temperature of the mold T_die0, and uses them as the initial temperature reference T_0=(T_bearing0, T_die0); after 16-level digital filtering, it is sent to the MCU via SPI; the MCU calculates the reference position S_0 according to the potential value-distance conversion relationship of "100mV corresponds to 1μm", stores it in non-volatile memory and displays it on the human-machine interface, and locks it after user confirmation.

[0059] Independent backlash identification (establishing backlash parameter b): After the MCU issues a backlash identification command, the MCU controls the slider to perform ±5μm micro-amplitude reciprocating probes near S_0 (probe speed 0.5° / s, dwell time 50ms in each direction), simultaneously acquiring the motor output rotation angle (≤0.0005°) and slider displacement (≤0.01μm), and uploading them to the MCU as "rotation angle-displacement" data packets; the MCU performs data differential analysis: when the rotation angle change is ≥0.001° and the displacement change is ≤0.1μm, it is determined that the segment is an empty stroke, corresponding to the comprehensive backlash b; repeating ≥3 times and taking the average value, the backlash parameter b with an identification accuracy ≤0.2μm is obtained and written into the parameter database.

[0060] III. Operational Phase: Dual-speed data acquisition, state estimation, error synthesis, and MPC prediction and compensation; Multi-source data acquisition and preprocessing (FPGA): The FPGA continuously acquires data at 10kHz to 50kHz, including magnetic scale potential value, running direction, motor current, bearing temperature T_bearing, mold temperature T_die, and slider acceleration. After preprocessing, the data is sent to the MCU with a delay of ≤10μs to ensure real-time performance.

[0061] Multi-rate Kalman filtering and state estimation (MCU): The MCU timestamps the high-frequency displacement sequence and the low-frequency operating condition signal, and uses EKF / UKF to establish state estimation. The state vector can be set as x=[e,v,a,i,T,b_state], where e is the displacement error estimate, v is the velocity, a is the acceleration, i is the current, T is the comprehensive temperature deviation, and b_state is the backlash / hysteresis state variable; the noise-resistant state of the filtered output is used as the input for subsequent MPC.

[0062] Comprehensive error calculation (including backlash, thermal drift, and disturbance): The current position S_1 of the bottom dead center is obtained by converting the potential value, and the basic error ΔS = S_1 - S_0; Backlash correction is performed according to the running direction: ΔS_b = k_b·b (k_b is the preset backlash compensation coefficient, such as 1.1) when moving in the opposite direction, and ΔS_b = 0 when moving in the same direction; The thermal drift model parameters α, β, and γ are updated online based on RLS to obtain the thermal drift compensation ΔS_T = α·(T_bearing-T_bearing0) + β·(T_die-T_die0) + γ·f (f is the stamping frequency); The disturbance observer is constructed using the displacement residual to obtain d_hat; The comprehensive error is: ΔS_total = ΔS + ΔS_b + ΔS_T + d_hat.

[0063] MPC Multi-Step Prediction and Rolling Optimization: The MCU has a built-in MPC module that establishes an "error-operating condition" correlation model based on historical data from nearly 100 strokes. The prediction target is to minimize the sum of squared comprehensive errors for the next 3 to 5 strokes. Constraints are set: motor current ≤ rated × 0.9, single compensation amount ≤ 10μm, and slider acceleration ≤ 5m / s². The rolling optimization outputs the optimal compensation amount sequence for the next 3 to 5 strokes and executes the first compensation amount in the current stroke.

[0064] IV. Dynamic dead zone, command issuance and self-learning update; Dynamic dead zone control: Set the dynamic dead zone threshold |ΔS_total|≤0.5μm (it can also be adaptively adjusted in the range of 0.3μm~1μm); when the error is within the dead zone, no compensation command is issued to avoid high-frequency oscillation; outside the dead zone, compensation is performed according to the MPC result.

[0065] Compensation command output and servo execution: When ΔS_total exceeds the dead zone, the MCU writes "compensation amount + backlash correction amount ΔS_b + thermal drift compensation ΔS_T + MPC multi-step prediction flag (used to identify the prediction step number corresponding to the current compensation amount)" to the servo driver via CANopen; the servo driver then fine-tunes the motor angle to change the slider motion curve of the next stroke, achieving precise bottom dead center stopping and offsetting the effects of thermal deformation / backlash / vibration / disturbance.

[0066] Self-learning triggering and execution (ensuring long-term accuracy): Set at least four types of triggers: When the absolute value of the difference between the actual error and the predicted error exceeds 0.3μm for 3 consecutive strokes: trigger MPC parameter calibration; When changing molds or running continuously for 100 hours: trigger backlash re-identification and update b; When the cumulative number of strokes reaches 500: trigger the iteration of the "error-operating condition" correlation model; When |actual error - predicted error| exceeds the threshold N times (N=3) or CUSUM detects an offset trend: trigger online model re-identification (preferably executed within the stop window or dead zone).

[0067] Human-computer interaction and parameter visualization: The human-computer interface displays parameters such as the bottom dead center reference position S_0, back clearance parameter b, comprehensive error ΔS_total, MPC prediction compensation amount, bearing temperature T_bearing, mold temperature T_die, thermal drift compensation amount ΔS_T, and disturbance observation value d_hat in real time. It also supports "cyclic test mode" (e.g., automatically repeating 20 to 50 cycles) for online adjustment of MPC weights, prediction step size and trigger threshold, so that the system can quickly and stably converge to the target bottom dead center position.

[0068] The above embodiments provide a complete closed-loop process from hardware installation, calibration and backlash identification, dual-speed acquisition and state estimation, comprehensive error synthesis, MPC multi-step prediction compensation, dynamic dead zone and self-learning update to servo execution. The parameters and steps are clear and can be directly implemented.

[0069] Example 2 (Detailed Implementation of Laser Interference Calibration + MPC Compensation) This embodiment, based on the "magnetic scale measurement + FPGA high-frequency acquisition preprocessing + MCU comprehensive error calculation and MPC multi-step prediction compensation" described in Embodiment 1, further adds a laser interferometry calibration function, which is used to correct the proportional error and zero drift of the magnetic scale measurement chain online / periodically (and correct the installation pitch angle error θ when necessary), making the position measurement more accurate, thereby improving the long-term stability of the comprehensive error ΔS_total calculation and MPC compensation solution.

[0070] The basic processes such as obtaining S_0, high-frequency acquisition of magnetic scale, low-latency transmission of SPI, and back gap identification are still consistent with those in Example 1 (for example: the FPGA acquires the potential value of the magnetic scale at 10kHz to 50kHz and transmits it to the MCU after filtering. The MCU calculates S_0 according to the formula of "potential value - distance" and locks it. Then, it probes and identifies the back gap b in the vicinity of S_0 within ±5μm).

[0071] I. Hardware Supplement: Installation and interface of the laser interferometry module; 1. A heterodyne laser interferometer (or a linear laser displacement interferometer with equivalent resolution) is fixedly mounted on a rigid part of the frame (away from the impact point and close to the guide rail base). Its beam axis is preferably parallel to the direction of slider movement. The interferometer output interface is a high-speed counting / serial data interface, connected to the FPGA's high-speed I / O or a dedicated counting interface.

[0072] 2. The reflective target is fixedly installed on the side of the slider or on a component rigidly connected to the slider as a reflective target (using a single target or double target structure); the mounting surface of the reflective target must be rigidly connected to the slider to avoid loosening.

[0073] A dual-target structure is preferred: a first reflective target and a second reflective target are arranged on the same plane of the slider, with the two targets spaced laterally by a baseline length L (e.g., L=80mm~150mm, preferably 100mm), which is used to estimate the installation pitch angle error θ caused by the slight tilt of the slider.

[0074] 3. Synchronous data acquisition and time alignment; During the calibration scan, the FPGA synchronously acquires and timestamps the following signals: Magnetic grating potential value (or converted magnetic grating displacement S_mag_raw); Laser interference displacement (S_laser_raw for single-channel; S_laser1_raw and S_laser2_raw for dual-channel). The acquired data packets are sent to the MCU via SPI, ensuring a high-speed link and time alignment mechanism consistent with the original system (≤8μs latency).

[0075] II. Calibration triggering conditions and calibration safety prerequisites; 1. Define the trigger condition (any one of the conditions must be met); First power-on after mold change; The change in bearing temperature or mold temperature relative to the initial temperature reference T_0, ΔT (where ΔT = max(|T_bearing-T_bearing0|, |T_die-T_die0|)) exceeds a threshold (e.g., ΔT ≥ 5℃). The cumulative number of strokes reaches the threshold (e.g., N≥200000 strokes). Alternatively, maintenance personnel can actively trigger the "calibrate" button on the human-machine interface.

[0076] Note: The acquisition and storage of T_0 follow the cold machine calibration process of Example 1 (bearing initial temperature T_bearing0 and mold initial temperature T_die0).

[0077] 2. Calibrate safety prerequisites The calibration scan is performed only if the following conditions are met: The stamping equipment is in "non-stamping / no-load calibration mode" (i.e., no material is being dropped and no stamping is being triggered). The servo drive is in position control mode with a speed limit of ≤50mm / s (consistent with the calibration phase).

[0078] If the safety conditions are not met, the MCU will refuse to enter the calibration process and display the reason on the interface.

[0079] III. Calibration Scanning Procedure (Multi-point Scanning with Micro-displacement Near S_0) To enter the calibration start point, the MCU controls the slider to move to the calibration start position: S_start = S_0 - A, where A is the half-width of the scan (e.g., A = 20 μm). After the slider reaches the position, it remains stable for 0.2s to 0.5s to ensure vibration attenuation.

[0080] Multi-point scanning trajectory: The MCU controls the slider to perform a step-like scanning with a step size d: Step size d: 2μm~5μm (preferably 2μm); Number of points M: 11 points~21 points (preferably 21 points); The scanning point sequence is: S_i=S_0-A+(i-1)·d, i=1…M, covering the S_0±A interval.

[0081] Data acquisition and denoising at each scanning point: At each scanning point S_i, the dwell time t_hold is maintained (e.g., 20ms to 50ms). Within the dwell window, the FPGA performs high-frequency sampling of the magnetic scale and laser interference output, and calculates the mean / median as the steady-state reading of the point: Magnetic scale reading: V_i (potential value) or S_mag_raw,i (converted by the formula "potential value - distance"); Laser reading: S_laser_raw,i (single channel) or S_laser1_raw,i, S_laser2_raw,i (dual channel). Bearing temperature and mold temperature are recorded simultaneously as part of the calibration record for easy subsequent traceability and model management.

[0082] IV. Solving for calibration parameters (k, o, and optional θ) and determining their validity: 4.1 Estimation of Installation Pitch Angle Error θ (Preferred Scheme for Dual Targets) When using dual targets, the MCU calculates the average displacement of the two laser channels for each scanning point: S_laser_avg,i=(S_laser1_raw,i+S_laser2_raw,i) / 2 Displacement difference: ΔS_laser,i=S_laser1_raw,i-S_laser2_raw,i The slight tilt angle of the slider near the scanning point (installation pitch angle error) is approximated by a small angle: θ_i≈arctan(ΔS_laser,i / L), where L is the distance between the two targets (measured and input into the system during installation, e.g., 100mm). A robust estimate of the full scan points is taken as θ: θ=median(θ_i) (or the mean after removing outliers). The laser displacement is then cosine corrected to obtain an approximate true displacement: S_laser_true,i=S_laser_avg,i / cos(θ); If a single target (single channel) scheme is used, θ can be incorporated into the scaling factor k for processing (without outputting θ separately), and effective correction of scaling / zero point error can still be achieved.

[0083] 4.2 Fitting the scaling factor k and the zero offset o: A linear calibration model is established between the original displacement S_mag_raw,i obtained by the magnetic scale conversion and the laser true displacement S_laser_true,i: S_laser_true,i=k·S_mag_raw,i+o. The MCU uses least squares fitting to solve for k and o to minimize the objective function: minΣ_i(k·S_mag_raw,i+o-S_laser_true,i)². After obtaining k and o, they are written into the "position conversion model parameter area" and a version number ver_n (with timestamp) is generated.

[0084] 4.3 Calibration Validity Judgment and Rollback MCU Calculation of Fitting Residual RMS: RMS=sqrt((1 / M)·Σ_i(k·S_mag_raw,i+o-S_laser_true,i)²) If RMS≤0.2μm (can be adjusted within the range of 0.1μm~0.5μm according to actual accuracy requirements), the calibration is deemed valid, and k, o (and θ) are saved; if RMS exceeds the limit or laser signal loss / saturation occurs, the calibration is deemed to have failed: do not update parameters, roll back to the previous version ver_(n-1), and display "Calibration failure reason + suggested handling" on the interface.

[0085] V. Application of calibration parameters in control closed loop (“Correct measurement → Recalculate ΔS_total → MPC compensation”) During normal operation, the FPGA still acquires the magnetic scale potential value and transmits it to the MCU via the original acquisition link. The MCU first obtains S_mag_raw using the original "potential value - distance" formula, and then corrects it according to the calibration parameters to obtain the current position: S_1 = k·S_mag_raw + o (If θ is obtained separately and further correction of installation projection error is needed, cos(θ) is introduced into S_1 or projection correction is performed according to the system definition.) The corrected S_1 is used for the basic error calculation in the calculation of the comprehensive error ΔS_total and the MPC solution: ΔS=S_1-S_0; and it is combined with the back gap compensation, thermal drift compensation and disturbance observation value to form ΔS_total. The subsequent dynamic dead zone determination and MPC multi-step prediction compensation process remains unchanged (including stopping the compensation output when the dead zone threshold is ≤0.5μm, and writing to the servo driver outside the dead zone via CANopen to perform fine-tuning, etc.).

[0086] After successful calibration data management and self-learning collaborative calibration, the MCU writes (k, o, θ), temperature (T_bearing, T_die), scan range A, step size d, RMS, etc. into the historical calibration database. When the "actual error - MPC prediction error" continuously exceeds the threshold (for example, the deviation of 3 consecutive strokes exceeds 0.3μm), in addition to triggering MPC model parameter calibration, a fast calibration scan can also be triggered first to eliminate systematic errors caused by measurement chain drift.

[0087] VI. Implementation Results (Description of Reproducible Engineering Results) By performing multi-point calibration via laser interferometry near S_0 and updating k, o (θ if necessary), the measurement errors caused by magnetic scale installation deviation, proportional error, and zero-point drift can be suppressed online, making the S_1 measurement closer to the true displacement. Therefore, the calculation of ΔS_total is more accurate, the error prediction and compensation output of MPC for the next 3 to 5 strokes is more stable, the bottom dead point position converges faster and drifts less during long-term operation, and invalid jitter compensation is avoided under the dynamic dead zone strategy.

[0088] Example 3 (IMU tightly coupled EKF fusion + perturbation observation + dynamic dead zone adaptation) This embodiment, based on the overall framework of "magnetic scale displacement measurement - FPGA high-frequency acquisition preprocessing - MCU comprehensive error ΔS_total calculation - MPC multi-step prediction compensation - dynamic dead zone - CANopen writing to servo driver" described in Embodiment 1, adds an inertial measurement unit (IMU) (accelerometer + gyroscope) on the slider, and uses a tightly coupled extended Kalman filter (EKF) on the MCU side for multi-sensor fusion; at the same time, it uses the fusion residual to construct the disturbance observation value d_hat, and adaptively adjusts the dynamic dead zone threshold according to the IMU vibration intensity to improve the vibration resistance and external disturbance resistance of the bottom dead point control under high-speed stamping.

[0089] I. Hardware Addition and Installation IMU Selection and Mounting Method A 6-axis IMU (3-axis accelerometer + 3-axis gyroscope) is fixedly mounted on the slider body or a mounting base rigidly connected to the slider. Requirements: Acceleration range ≥ ±16g, noise density ≤ 0.2mg / √Hz; Angular velocity range ≥ ±2000° / s, noise density ≤ 0.02° / s / √Hz; Output frequency ≥ 2kHz (4kHz or 8kHz recommended).

[0090] The IMU mounting surface is rigidly connected to the slider to prevent loosening; the Z-axis of the IMU should be aligned with the nominal downward direction of the slider as much as possible (small angle installation error is allowed, which will be estimated and compensated by EKF).

[0091] The original system sensors and sampling remain unchanged. The hardware acquisition links such as magnetic grating ruler (resolution ≤0.1μm), servo motor running direction detection module (sampling frequency ≥1kHz), motor current, bearing temperature T_bearing, mold temperature T_die, and slider acceleration signal are still retained. The FPGA acquires the data at a high frequency of 10kHz to 50kHz and transmits it to the MCU with a delay of ≤10μs after filtering.

[0092] II. FPGA-side synchronous acquisition and hardware timestamp (to ensure fusion usability) The unified time base FPGA internally sets a 1MHz free-running counter as the unified time base (time resolution 1μs) and performs hardware marking of all sampling points.

[0093] Synchronous acquisition strategy: Magnetic scale: sampling at 10kHz to 50kHz, switching to a higher frequency range (20kHz to 100kHz, following the "core area encrypted sampling" logic of Example 1) when entering the core region of the bottom dead center. IMU: sampling at 4kHz or 8kHz (8kHz recommended for easy capture of impact transients). Direction, current, and temperature: acquired at the original sampling frequency.

[0094] The FPGA packages each sampled data into: {timestamp, S_mag_raw or potential value V, a_IMU, ω_IMU, dir, I, T_bearing, T_die} and sends it to the MCU via SPI, enabling the MCU to perform time alignment and tight coupling fusion (the original solution already clearly has timestamp alignment and multi-rate Kalman filtering mechanisms).

[0095] III. MCU-side tightly coupled EKF fusion (core: IMU for prediction, magnetic scale for measurement update) Objective: Under strong impact and vibration conditions, relying solely on magnetic grating rulers may result in transient noise / jitter; introducing an IMU can provide high-bandwidth dynamic information, while an EKF outputs noise-resistant, continuous, and observable fused displacement / velocity / attitude and bias.

[0096] 3.1 Initialization and Static Calibration (performed once after power-on or mold change) Static bias estimation: When the device is stationary (slider stops for ≥2s), collect N=2000 sets of IMU data (approximately 0.25s for 8kHz, or approximately 0.5s for 4kHz), and calculate: Acceleration zero bias b_a0=mean(a_IMU)-g·e_z (e_z is the nominal vertical unit vector, g=9.80665m / s²); Gyroscope zero bias b_ω0=mean(ω_IMU); Write b_a0 and b_ω0 into the parameter area as initial values ​​for EKF.

[0097] Initial state settings: initial position p0 = S_mag_raw(0) (or initial displacement obtained by converting potential value to distance); initial velocity v0 = 0; initial attitude q0 is a quaternion (or a small Euler angle) that aligns static acceleration with the direction of gravity; initial bias b_a = b_a0, b_ω = b_ω0.

[0098] 3.2 EKF State Vector and Discrete Update To facilitate engineering implementation, this embodiment adopts a tightly coupled model that primarily uses one-dimensional displacement control while simultaneously estimating small-angle attitude. The state vector is defined as: x=[p,v,φ,θ,b_ax,b_ay,b_az,b_ωx,b_ωy,b_ωz]^T where: p: the fused displacement (μm or m) of the slider along the nominal downward direction; v: the fused velocity; φ, θ: the roll angle and pitch angle of the slider's slight tilt (rad, satisfying |φ|, |θ|≤5°); b_a, b_ω: the IMU acceleration and angular velocity bias.

[0099] (1) Prediction step (driven by IMU) In each IMU sampling period Δt (e.g., 1 / 8000s or 1 / 4000s): Attitude update (small angle approximation): φ_n=φ_{n-1}+(ω_x-b_ωx)·Δt; θ_n=θ_{n-1}+(ω_y-b_ωy)·Δt; The zero-biased acceleration vector (a_IMU-b_a) is transformed to the gantry base by the rotation matrix R(φ_n, θ_n). The system is defined, and the components on the vertical unit vector e_z are taken to obtain the vertical acceleration: a_z = e_z^T·R(φ_n, θ_n)·(a_IMU-b_a)-g; the velocity and position integrals are: v_n = v_{n-1} + a_z·Δt; p_n = p_{n-1} + v_{n-1}·Δt + 0.5·a_z·Δt^2; the zero bias of acceleration b_a and the zero bias of gyroscope b_ω are updated according to the random walk model.

[0100] (2) Measurement update (driven by magnetic scale): When a magnetic scale sample is received (e.g., 10kHz to 50kHz), the EKF is updated with the magnetic scale displacement measurement z=S_mag_raw. The measurement model can be expressed as z=p+n_z, where p is the fused displacement predicted / estimated by the EKF, and n_z is the measurement noise. The EKF outputs the fused displacement p, fused velocity v, attitude angle φ, θ, and bias estimate.

[0101] Explanation: The state vector is estimated using EKF / UKF and noise-resistant state parameters are output. Based on this, the IMU is used as the prediction input and tightly coupled.

[0102] IV. Compensation for Displacement Projection Errors Caused by Tilt (Converting "Attitude Angle" into "Displacement Correction") There is usually an installation offset between the reading head and the geometric center of the slider of the magnetic scale (for example, the reading head is offset by r in the horizontal direction relative to the center of mass of the slider, which is 10mm to 50mm, and is measured and recorded in the HMI after installation). When the slider has pitch θ or roll φ, the displacement at the reading head will be superimposed with the equivalent displacement error caused by the rotation.

[0103] This embodiment uses a small-angle approximation for compensation: Let the head be offset by r_y in the pitch direction and r_x in the roll direction. Then the equivalent displacement error caused by tilt is: Δp_tilt≈r_y·θ+r_x·φ. The displacement after tilt compensation is: S_1=p-Δp_tilt. Where S_1 is the current position of the bottom dead center and participates in subsequent error calculation (based on the attitude angle to compensate for the displacement projection error).

[0104] V. Construction of perturbation observations d_hat (using fused residuals to characterize external perturbations such as impact / material changes) The disturbance observer can be constructed from the "residual between the filtered displacement value and the predicted displacement value" to characterize external disturbances such as mold impact and material hardness changes, and incorporate them into the error and MPC model.

[0105] This embodiment uses an innovative amount between the EKF-predicted displacement and the displacement measured by the magnetic grating ruler as the residual: Let r_n = S_mag_raw,n - p_hat(n|n-1), where p_hat(n|n-1) is the EKF-predicted fused displacement at time n. The disturbance observation value is filtered using a first-order low-pass filter: d_hat(n) = λ·d_hat(n-1) + (1-λ)·r_n, where λ is taken as 0.9~0.99 (0.95 recommended), so that d_hat mainly reflects the slow / medium speed components caused by impact disturbances and sudden changes in operating conditions.

[0106] VI. Overall Error ΔS_total and MPC Input When the MCU reaches bottom dead center in each stroke, it calculates the overall error as follows: S_1 uses the result of "fusion + tilt compensation": Basic error: ΔS = S_1 - S_0; Backlash compensation: When moving in the opposite direction, ΔS_b = k_b·b (k_b is the preset backlash compensation coefficient, such as 1.1); when moving in the same direction, ΔS_b = 0. Thermal drift compensation: ΔS_T = α·(T_bearing-T_bearing0) + β·(T_die-T_die0) + γ·f, where α, β, and γ are updated online by RLS; Perturbation observation: Add d_hat; Overall error: ΔS_total=ΔS+ΔS_b+ΔS_T+d_hat; Then, ΔS_total, motor current, stamping frequency, T_bearing, T_die, acceleration / IMU characteristics, and filtered state parameters are input into the MPC model for future processing.

[0107] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. All equivalent structural transformations made under the inventive concept of the present invention using the contents of the present invention specification and drawings, or direct / indirect applications in other related technical fields, are included within the protection scope of the present invention.

Claims

1. A method for self-adjusting the bottom dead center position of a servo stamping device, characterized in that, The method includes the following steps: S1, construct a dual-speed collaborative architecture, setting the FPGA unit as the high-frequency acquisition and preprocessing layer and the MCU unit as the control decision layer; S2, Bottom Dead Point Calibration: Control the slider to reach the bottom dead point, collect the position sensor output and determine the bottom dead point reference position S_0, and at the same time collect and store the initial temperature reference T_0; S3, backlash identification: The control slider performs a small reciprocating motion near the bottom dead center reference position S_0, and simultaneously collects the servo motor rotation angle signal and slider displacement signal, quantizes and stores the backlash parameter b of the servo motor transmission chain; S4, Multi-source data acquisition and preprocessing: In each stroke, the FPGA unit acquires multi-source data at high frequency, including position sensor signals, servo motor running direction signals, motor current signals, temperature signals and slider acceleration signals, and transmits the preprocessed data to the MCU unit. S5, State estimation: The MCU unit timestamps the received data and performs Kalman filtering to obtain state parameters for error calculation and prediction. S6, Comprehensive Error Calculation: The MCU unit calculates the basic position error ΔS=S_1-S_0 based on the real-time position S_1 and the bottom dead center reference position S_0, and performs directional correlation compensation on the backlash parameter b according to the running direction of the servo motor to obtain the backlash compensation amount ΔS_b; the MCU unit calculates the thermal drift compensation amount ΔS_T based on the temperature signal and the stamping frequency, and constructs the disturbance observation value d_hat based on the position measurement residual; the comprehensive error ΔS_total is obtained by superimposing the ΔS, ΔS_b, ΔS_T and d_hat. S7, MPC Multi-Step Prediction and Compensation Solution: Input ΔS_total and operating parameters, including temperature, stamping frequency, motor current and vibration amplitude, into the Model Predictive Control Module (MPC), and perform rolling optimization to obtain the optimal compensation sequence for multiple future strokes, and determine the compensation amount for the current stroke. S8, Dynamic Dead Zone Control: When |ΔS_total| is within the preset dynamic dead zone threshold, the compensation output is suppressed; when |ΔS_total| exceeds the dynamic dead zone threshold, the current stroke compensation amount is output. S9, Compensation command execution: The compensation amount is sent to the servo driver to fine-tune the motion curve of the servo motor and the slider, so that the bottom dead center position of the subsequent stroke converges to the bottom dead center reference position S_0.

2. The method according to claim 1, characterized in that: In the bottom dead center calibration in step S2, the FPGA unit acquires the position sensor signal at a sampling frequency of 10kHz to 50kHz and transmits it to the MCU unit after multi-stage digital filtering. The MCU unit calculates the bottom dead center reference position S_0 based on the signal-displacement conversion relationship of the position sensor and locks and stores it.

3. The method according to claim 1, characterized in that: In the backlash identification in step S3, the slider performs a micro-amplitude reciprocating trial motion within a range of ±5μm near the bottom dead center reference position S_0. The MCU unit performs differential calculation on the "angle-displacement" pair data. When the angle change reaches a preset threshold and the displacement change does not exceed the preset threshold within the continuous sampling period, the equivalent displacement obtained by converting the angle change according to the angle-displacement conversion relationship is determined as the backlash parameter b of the servo motor drive chain. After multiple identifications and elimination of outliers, the average value is taken to form the final backlash parameter.

4. The method according to claim 1, characterized in that: The Kalman filter is an extended Kalman filter (EKF) or an unscented Kalman filter (UKF), and the state parameters include displacement error, slider speed, slider acceleration, motor current, comprehensive temperature deviation, and backlash state variable.

5. The method according to claim 1, characterized in that: The thermal drift compensation amount is obtained through online identification. The MCU unit updates the thermal drift model parameters based on the temperature signal and stamping frequency using recursive least squares (RLS), and incorporates the thermal drift compensation amount as a feedforward term and / or disturbance term into the calculation process of the comprehensive error ΔS_total.

6. The method according to claim 1, characterized in that: The MPC multi-step prediction and compensation solution satisfies the constraints, which include the upper limit of servo motor current, the upper limit of single compensation amount, the upper limit of slider acceleration, and the upper limit of compensation amount change rate. Furthermore, when the deviation between the actual error and the MPC prediction error meets the preset criteria, online model re-identification is triggered to update the prediction step size, weight matrix, and constraint boundaries.

7. The method according to claim 1, characterized in that: The process also includes a laser interferometric calibration step. The servo stamping equipment is equipped with a laser interferometric measurement module, which includes a laser interferometer and a reflective target mounted on the slider or a rigidly connected component to the slider. During bottom dead center calibration or equipment operation, when preset calibration trigger conditions are met, including mold change, temperature change relative to the initial temperature T_0 exceeding a threshold, or cumulative stroke count reaching a threshold, the slider is controlled to perform micro-displacement calibration scanning at least two points near the bottom dead center reference position S_0. Simultaneously, the magnetic scale displacement and laser interferometric displacement are collected and linear alignment fitting and error fitting are performed to obtain the magnetic scale scale scaling factor k, zero-point offset o, and installation pitch angle error θ. The k, o, and θ are written back to the position conversion model to correct the calculation of the real-time position S_1, and the corrected real-time position S_1 is used in the calculation of the comprehensive error ΔS_total and the MPC multi-step prediction compensation solution to achieve online suppression of drift error and scaling error of the position measurement chain.

8. The method according to claim 1, characterized in that: It also includes an inertial navigation fusion step, wherein the servo stamping device is equipped with an inertial measurement unit (IMU) on the slider, and the IMU includes an accelerometer and a gyroscope; The FPGA unit synchronously acquires and timestamps the IMU and magnetic scale signals before transmitting them to the MCU unit. The MCU unit executes a tightly coupled extended Kalman filter (EKF), using the IMU output as the state prediction input and the magnetic scale displacement as the measurement update input to obtain fused estimates of the slider's displacement, velocity, attitude angle, and sensor bias. Based on the attitude angle, the MCU unit calculates and compensates for the displacement projection error caused by the slider's slight tilt in real time. Simultaneously, based on the fused estimates and the position measurement residual, a disturbance observation value d_hat is constructed. The dynamic dead zone threshold is then nonlinearly and adaptively adjusted according to the vibration energy or acceleration peak measured by the IMU. This suppresses ineffective compensation output under high vibration conditions and improves compensation resolution under low vibration conditions, thereby enhancing the bottom dead center control's resistance to vibration interference and external disturbances.

9. A self-adjusting control system for the bottom dead center position of a servo stamping equipment, used to implement the method described in any one of claims 1-8, characterized in that, include: The FPGA unit is used for high-frequency synchronous acquisition and preprocessing of multi-source sensor data, and timestamps the acquired data. The MCU unit is used to receive the multi-source data and perform bottom dead point calibration, back gap identification, state estimation, comprehensive error calculation, MPC multi-step prediction and compensation decision, dynamic dead zone control, and self-learning update of model parameters and back gap parameters. A position sensor is used to detect slider displacement and for bottom dead center calibration and real-time position measurement. Temperature sensors are used to detect the operating temperature of equipment; An inertial measurement unit (IMU) is used to acquire slider acceleration, vibration, and attitude signals. Servo motor running direction detection module, used to obtain the running direction of servo motor; The laser interferometry module is used to perform laser interferometry calibration of the slider displacement. Servo driver, used to receive compensation instructions output by MCU unit and fine-tune the movement of servo motor and slider.

10. The system according to claim 9, characterized in that: The position sensor is a magnetic scale, and when the slider enters the core region of the bottom dead center (BDC) within ±10μm of the theoretical BDC, the FPGA unit switches to a higher sampling frequency of 20kHz to 100kHz to improve the BDC measurement resolution and vibration resistance. The MCU unit includes a self-learning trigger and execution module. The self-learning trigger conditions include: model parameter calibration trigger, backlash parameter update trigger, error-operating condition correlation model optimization trigger, and MPC model adaptive re-identification trigger. When any trigger condition is met, the backlash parameter, thermal drift model parameter, and MPC model parameter are automatically updated. The MCU unit writes compensation instructions to the servo driver via the fieldbus. The compensation instructions carry at least the backlash compensation amount, thermal drift compensation amount, and MPC multi-step prediction identifier to identify the prediction step number corresponding to the current compensation amount. This allows the servo driver to precisely fine-tune the servo motor angle of the next stroke according to the compensation instructions to change the slider motion curve and achieve precise stopping at the BDC position.

Images