Position and pressure closed loop control method for a servo press

Abstract

This invention discloses a closed-loop control method for position and pressure of a servo press-fitting machine, belonging to the field of mechatronics precision control technology. The single-workpiece personalized control of this invention achieves a press-fitting position accuracy of ±0.5μm and a press-fitting force accuracy of ±0.05%FS, more than 20 times higher than traditional methods. It completely eliminates the influence of material property fluctuations. Predictive control based on real-time identification can accurately predict the press-fitting process, with force overshoot less than 0.1%FS, achieving true zero overpressure and a 100% press-fitting pass rate. Early anomaly detection based on constitutive parameters can identify defects in the early stages of press-fitting, avoiding batch scrapping and significantly reducing production costs. Federated learning-driven model evolution ensures that the accuracy decreases by no more than 0.2μm after 10,000 hours of continuous operation, eliminating the need for periodic manual calibration. The constitutive parameters, press-fitting data, and quality results of each workpiece are recorded, achieving quality traceability throughout the product's entire lifecycle.

Description

Technical Field

[0001] This invention belongs to the field of mechatronics precision control technology, specifically relating to a closed-loop control method for the position and pressure of a servo press machine. Background Technology

[0002] Servo press fitting is a core technology in precision manufacturing for achieving processes such as interference fit, riveting, and stamping. Its control precision directly determines the product's service life and reliability. As the manufacturing industry develops towards high-end and intelligent manufacturing, the requirements for press fitting quality have been raised from "qualified" to "zero defects," and traditional control methods are facing insurmountable technical bottlenecks.

[0003] The fundamental flaw in existing technologies lies in the fact that all control methods are based on a "general model + feedback correction" paradigm. This involves first establishing an average model applicable to all workpieces, and then correcting errors through sensor feedback. However, the press-fitting process is a highly nonlinear and irreversible elastoplastic deformation process. Each workpiece exhibits slight but non-negligible differences in material properties and initial state. A general model can never accurately describe the true deformation characteristics of a single workpiece, resulting in a theoretical upper limit to control accuracy. Furthermore, existing anomaly detection methods rely on comparing the press-fitting curve with a standard curve, only detecting anomalies in the later stages or after press-fitting, at which point defects have already occurred.

[0004] Although technologies such as digital twins and machine learning have been applied to press-fitting control in recent years, these technologies are essentially still optimizations of general models, without changing the basic architecture of "modeling first, then controlling," and cannot fundamentally solve the problems caused by the differences in the characteristics of individual workpieces. Therefore, a completely new control paradigm is urgently needed to break through the theoretical bottlenecks of existing technologies.

[0005] Based on this, the present invention provides a closed-loop control method for position and pressure of a servo press machine. Summary of the Invention

[0006] The purpose of this invention is to completely overturn the traditional press-fitting control technology and provide a closed-loop control method for the position and pressure of a servo press-fitting machine based on online characteristic identification of a single workpiece. By sensing the material characteristics of the current workpiece in real time during the press-fitting process, a unique personalized control strategy is generated, realizing a new paradigm of "identification before control". This fundamentally solves the accuracy problems caused by material characteristic fluctuations and workpiece tolerances, while also enabling early detection of press-fitting anomalies and full lifecycle model evolution.

[0007] The technical solution adopted by this invention to solve its technical problem is: a closed-loop control method for position and pressure of a servo press machine, comprising the following steps: S1 Initial Model Pre-training and System Calibration: Construct a general initial model that includes basic mechanical dynamics and servo system characteristics, collect press-fitting data of different batches of standard workpieces, complete the static calibration and dynamic response identification of the system, and establish a material property benchmark database; S2 Online Real-Time Identification of Single Workpiece Characteristics in the Initial Stage of Press Fitting: During the elastic deformation stage of press fitting, force-displacement data for the first 50ms is collected at a frequency of 10kHz. The recursive least squares method with forgetting factor is used to fit the three constitutive parameters of the current workpiece in real time: elastic modulus, yield strength and hardening index, to generate a personalized constitutive model for the workpiece. S3 Adaptive Predictive Control Based on Personalized Constitutive Model: The constitutive parameters identified in real time are input into the model predictive controller to solve the optimal pressing trajectory and control sequence of the current workpiece online. At the same time, the pressing energy is used as the third control variable to realize the three-variable coordinated constraint control of position, pressure and energy. S4 High-Order Disturbance Observer and Hierarchical Error Compensation: A high-order disturbance observer is designed to estimate unmodeled dynamics and external disturbances in the system in real time. A three-layer compensation architecture of "identification feedforward-predictive control-microfeedback" is adopted to compensate for model error, dynamic error and residual error respectively. S5 Early Anomaly Detection and Classification Handling Based on Constitutive Parameters: The constitutive parameters identified in real time are compared with the material property benchmark database. Anomalies such as workpiece defects and material non-compliance are identified within 30% of the stroke before the pressing is completed. Based on the anomaly level, parameter self-adjustment, deceleration warning or emergency shutdown are executed. S6 Federated Learning-Driven Full Lifecycle Model Evolution: After each compression is completed, constitutive parameters, compression data, and quality results are uploaded to the cloud-based federated learning platform. Without leaking local data, the global initial model is collaboratively optimized, enabling model evolution across multiple devices and scenarios.

[0008] Preferably, the online real-time identification of single workpiece characteristics in step S2 specifically includes: After the pressure head contacts the workpiece, S21 starts the characteristic identification program when the pressure reaches the preset threshold F0 (usually 1% to 2% of the rated pressure). The S22 continuously acquires force-displacement data for 50ms at a sampling frequency of 10kHz, obtaining 500 sets of sample points. S23 establishes a linearized model of the force-displacement relationship based on the theory of elastic-plastic deformation: force equals equivalent stiffness multiplied by displacement plus initial force; S24 uses a recursive least squares method with a forgetting factor to estimate the equivalent stiffness and initial force in real time, and converts them into elastic modulus and yield strength. Based on the subsequent 10ms of plastic deformation data, S25 fitted the hardening index and completed the construction of the personalized constitutive model.

[0009] Preferably, the objective function for the position-pressure-energy three-variable coordinated control in step S3 is: Minimize the weighted sum of squared errors over the next Np control cycles. The error term includes position error, pressure error, and energy error. At the same time, a weighted penalty term for the control quantity is added. Among them, the pressure-fitting energy is defined as the integral of force with respect to displacement; The constraints include: upper and lower limits of the control quantity, upper and lower limits of the rate of change of the control quantity, maximum pressing force limit, and maximum pressing energy limit.

[0010] Preferably, the design of the higher-order perturbation observer in step S4 is as follows: An extended state equation containing system state and disturbance terms is constructed, and the system state variables and total disturbance are estimated in real time using the observer gain matrix and disturbance gain matrix. The bandwidth of the disturbance observer is set to 5 to 10 times the control bandwidth, enabling real-time estimation of all disturbances with frequencies below 500 Hz.

[0011] Preferably, the three-layer compensation architecture in step S4 is as follows: Identify feedforward compensation: Based on the constitutive model identified in real time, calculate the feedforward control quantity to compensate for more than 90% of the model error, with a compensation frequency of 1 time per press. Predictive control compensation: Based on model predictive control, it compensates for system dynamic errors and slow-varying disturbances at a frequency of 1000Hz. Micro-feedback compensation: A high-frequency PI controller is used to compensate for residual errors and rapid disturbances, with a compensation frequency of 10000Hz.

[0012] Preferably, the early anomaly detection based on constitutive parameters in step S5 specifically includes: S51 establishes a material property benchmark database, which includes the range of elastic modulus, yield strength, and hardening index of qualified workpieces; S52 compares the constitutive parameters identified in real time with the reference range; If the parameter exceeds the reference range by less than 10%, it is judged as a slight deviation and the control parameter is automatically adjusted. If the parameters of S54 exceed the reference range by 10% to 30%, it is judged as a moderate abnormality, an early warning is issued and the pressing speed is reduced. If the parameters of S55 exceed the reference range by more than 30%, it is judged as a serious abnormality and an emergency shutdown is initiated immediately.

[0013] Preferably, the federated learning-driven full lifecycle model evolution described in step S6 specifically includes: After each press-fitting process, the S61 local device generates a local model update quantity and does not transmit the original production data. The S62 cloud server aggregates local updates from multiple devices and uses a federated averaging algorithm to update the global initial model. The updated global initial model is downloaded locally on the S63 device and used for feature identification in the next press-fitting process. S64 performs a global model evaluation every 1000 press cycles. If the model accuracy improves by more than 1%, it triggers a model update for all devices.

[0014] Preferably, this method is applicable to scenarios such as press-fitting of automotive parts, press-fitting of new energy vehicle battery modules, press-fitting of precision bearings, and press-fitting of aerospace parts. The press-fitting position accuracy is ≤ ±0.5μm, the press-fitting force accuracy is ≤ ±0.05%FS, the press-fitting energy accuracy is ≤ ±0.2%, and the early anomaly detection accuracy is ≥99.9%.

[0015] The advantages of this invention are: 1. Individual workpiece personalized control enables pressing position accuracy to reach ±0.5μm and pressing force accuracy to reach ±0.05%FS, which is more than 20 times higher than traditional methods. It completely eliminates the influence of material property fluctuations. Based on real-time identification, predictive control can accurately predict the pressing process. The force overshoot is less than 0.1%FS, achieving true zero overpressure and a pressing qualification rate of 100%.

[0016] 2. Early anomaly detection based on constitutive parameters can identify defects in the early stages of pressing, avoiding batch scrapping and significantly reducing production costs. Federated learning-driven model evolution ensures that the accuracy decreases by no more than 0.2μm after the equipment has been running continuously for 10,000 hours. No periodic manual calibration is required. The constitutive parameters, pressing data and quality results of each workpiece are recorded, realizing quality traceability throughout the entire product lifecycle. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a block diagram of the overall architecture of the ultra-precision servo press fitting system of the present invention; Figure 2 This is a flowchart of the online identification process for single workpiece characteristics according to the present invention; Figure 3 This is the overall flow chart of the press-fit control of the present invention; Figure 4 This is a block diagram of the three-variable collaborative MPC control of the present invention; Figure 5 This is a block diagram of the three-layer compensation architecture of the present invention; Figure 6 This is an evolutionary block diagram of the federated learning model of this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] The core principle of this invention is that the force-displacement relationship in the initial elastic stage of press fitting contains all the constitutive property information of the workpiece material. By rapidly collecting this information in the early stage of press fitting, the elastic modulus, yield strength, and hardening index of the current workpiece can be fitted in real time, thereby accurately predicting the force-displacement relationship in the subsequent plastic deformation stage, and generating the optimal press fitting trajectory for the workpiece, achieving true personalized control.

[0021] The specific technical solution is as follows: 1. Single-workpiece online real-time characteristic identification technology This is the core innovation of this invention. Abandoning traditional offline batch identification methods, it collects force-displacement data at an ultra-high sampling frequency of 10kHz during the elastic deformation stage of the pressing process (typically the first 5% to 10% of the pressing stroke). Using a recursive least squares method with a forgetting factor, it fits the three key constitutive parameters of the current workpiece—elastic modulus, yield strength, and hardening index—in real time within 50ms, generating a unique elastoplastic constitutive model for that workpiece. This technology achieves a leap from "one mold per batch" to "one mold per piece," eliminating errors caused by fluctuations in material properties at the source.

[0022] 2. Position-Pressure-Energy Three-Variable Coordinated Control Technology For the first time, pressing energy is introduced as an independent control variable into pressing control. Pressing energy is the integral of force over displacement, directly reflecting the degree of workpiece deformation and bonding strength, and is a more fundamental quality indicator than position and pressure. A model predictive control objective function incorporating three variables—position, pressure, and energy—is constructed to achieve coordinated constraint control among the three. This technology not only ensures the geometric accuracy of pressing but also guarantees its mechanical properties and bonding strength, solving the problem of "formal but not substantive" control in traditional methods.

[0023] 3. High-order perturbation observer and three-layer compensation architecture A high-order disturbance observer is designed to estimate unmodeled dynamics, friction, vibration, and other external disturbances in the system in real time. The disturbance estimates are then fed forward to the control input, significantly improving the system's disturbance rejection capability. Simultaneously, a three-layer compensation architecture of "identification feedforward - predictive control - micro-feedback" is constructed: identification feedforward compensates for over 90% of the model error, predictive control compensates for dynamic errors and slowly varying disturbances, and micro-feedback compensates for residual errors and rapidly varying disturbances. This architecture balances control accuracy and response speed.

[0024] 4. Early anomaly detection techniques based on constitutive parameters Anomaly detection is performed using constitutive parameters obtained in real time. Anomalies such as material defects and workpiece non-compliance directly cause constitutive parameters to deviate from the normal range. Therefore, anomalies can be identified within 30% of the stroke before press-fitting is completed, which is more than 70% earlier than traditional curve-based comparison methods. A graded handling strategy is implemented according to the anomaly level to avoid further losses from continued press-fitting and to reduce unnecessary downtime.

[0025] 5. Federated Learning-Driven Full Lifecycle Model Evolution By employing federated learning technology, collaborative model evolution across multiple devices and scenarios is achieved without leaking local production data. Each local device uploads its model updates to the cloud, where updates from all devices are aggregated, the global initial model is optimized, and then distributed to all devices. This technology enables the accuracy of the initial model to continuously improve with usage time, achieving self-learning and self-evolution of the system.

[0026] The invention will now be described in detail with reference to specific embodiments and accompanying drawings of turbine blade tenon press-fitting. Turbine blade tenon press-fitting is one of the most critical processes in aero-engine manufacturing, requiring a press-fitting position accuracy of ±2μm, a press-fitting force accuracy of ±0.5%FS, a press-fitting force range of 20kN~80kN, and 100% pass rate and traceability.

[0027] 1. System Hardware Components The hardware architecture of the ultra-precision servo press-fitting machine system used in this embodiment is as follows: Figure 1 As shown, the specific components are as follows: Servo press machine main unit: adopts gantry frame, rated press force 100kN, stroke 300mm, and overall rigidity ≥100N / μm; Servo drive system: It adopts a 5kW permanent magnet synchronous servo motor, paired with a high-precision planetary reducer and ball screw, with an encoder resolution of 0.01μm and a response frequency of 2kHz; Sensor system: including a high-precision piezoelectric force sensor (range 100kN, accuracy 0.03%FS, sampling frequency 20kHz), a laser interferometer (range 300mm, accuracy 0.01μm, sampling frequency 10kHz), and 3 PT1000 temperature sensors (accuracy ±0.01℃). Motion controller: It adopts a self-developed FPGA+ARM+GPU heterogeneous architecture motion controller with a control cycle of 0.1ms and supports parallel computing and hardware acceleration; Industrial computer: equipped with an Intel Core i9-13900K processor, 64GB DDR5 memory, and an NVIDIA RTX 4070 graphics card, used for data processing and federated learning clients; Cloud-based federated learning platform: It uses Alibaba Cloud servers to deploy the federated learning framework and is responsible for the aggregation and updating of the global model.

[0028] 2. Initial Model Pre-training and System Calibration (1) System static calibration A laser interferometer was used to calibrate the positional accuracy of the press-fitting machine, and a positional error compensation table was established. The force sensor was calibrated using a standard force measuring instrument, and a force error compensation table was established. The dynamic response characteristics of the servo system are calibrated to obtain the system's transfer function.

[0029] (2) Initial model pre-training Collect press-fit data for 1,000 standard turbine blades, including force-displacement curves, temperature, vibration and other parameters; Offline fitting of constitutive parameters for each workpiece to establish a material property benchmark database; The initial general constitutive model and anomaly detection model are trained for characteristic identification in the early stages of press assembly.

[0030] 3. Complete press-fitting control process The overall press-fit control process of this invention is as follows: Figure 3 As shown, the specific steps are as follows: (1) System initialization After the system is powered on, it performs operations such as zeroing the press machine, calibrating the sensors, loading the initial model, and connecting to the federated learning client.

[0031] (2) Workpiece loading and positioning The turbine blades and disk are placed on a special tooling to complete the positioning and clamping.

[0032] (3) Online real-time identification of single workpiece characteristics in the initial stage of press fitting The driving pressure head slowly approaches the workpiece at a speed of 1 mm / s. When the pressure reaches 0.5 kN (0.5% of the rated pressure), it is determined to be in contact with the workpiece. The pressure head continues to press down at a speed of 0.5 mm / s, entering the elastic deformation stage; When the pressure reaches 2kN (2% of the rated pressure), the characteristic identification program is started to continuously collect force-displacement data for 50ms at a frequency of 10kHz, resulting in 500 sets of sample points. The equivalent stiffness and initial force are estimated in real time using the recursive least squares method with a forgetting factor set to 0.98. The elastic modulus is calculated based on the equivalent stiffness. The formula is: elastic modulus equals equivalent stiffness multiplied by workpiece length divided by cross-sectional area. Continue collecting plastic deformation data for 10ms, and fit the yield strength and hardening index to complete the construction of the personalized constitutive model.

[0033] (4) Adaptive predictive control based on personalized constitutive models The real-time identified elastic modulus, yield strength, and hardening index are input into the model prediction controller; The prediction time domain is set to 30 control cycles, the control time domain is set to 5 control cycles, the location weight coefficient is 10, the pressure weight coefficient is 5, and the energy weight coefficient is 20. The quadratic programming problem is solved online to obtain the optimal pressing trajectory and control sequence. The constraints include: control voltage range -10V~10V, control voltage change rate range -50V / s~50V / s, maximum pressing force 85kN, and maximum pressing energy 500J. The first control input is executed to drive the servo motor to move.

[0034] (5) Higher-order disturbance observers and hierarchical error compensation A high-order disturbance observer estimates system disturbances in real time. The observer gain matrix is ​​set to [100, 10000], and the disturbance gain matrix is ​​set to 100000. The three-layer compensation architecture operates in parallel: Identify feedforward compensation: Calculate the feedforward control quantity based on the personalized constitutive model to compensate for model errors; Predictive control compensation: Generates control variables based on MPC solution results to compensate for dynamic errors; Micro-feedback compensation: A high-frequency PI controller is used to generate feedback control quantities to compensate for residual errors; The total control quantity is the sum of the three compensation quantities mentioned above.

[0035] (6) Early anomaly detection and graded treatment based on constitutive parameters The constitutive parameters obtained in real time are compared with the benchmark database. The acceptable range is: elastic modulus 190GPa~210GPa, yield strength 800MPa~900MPa, and hardening index 0.1~0.2. Tiered handling: If the parameters are within the acceptable range, continue with normal pressing. If the parameters exceed the acceptable range by 5% to 10%, the MPC weighting coefficient will be automatically adjusted to increase the weighting of pressure and energy. If the parameters exceed the acceptable range by 10% to 20%, an audible and visual warning will be issued, and the pressing speed will be reduced to 0.5 mm / s; If the parameters exceed the acceptable range by more than 20%, immediately stop the machine and record the abnormal information.

[0036] (7) Pressing completed and data uploaded When the pressing position, pressure, and energy all meet the requirements, the pressing is deemed qualified. Record the constitutive parameters, force-displacement curves, energy curves, temperature, vibration, and other data for this press fitting process; Generate local model update data and upload it to the cloud-based federated learning platform.

[0037] (8) Federated learning model update The cloud server aggregates the local update data from all devices and uses a federated averaging algorithm to update the global initial model. The local device downloads and updates the global initial model for feature identification in the next pressing process; A global model evaluation is performed every 1,000 press cycles. If the model accuracy improves by more than 1%, a model update is triggered for all devices.

[0038] 4. Experimental Results and Analysis To verify the effectiveness of this invention, a comparative experiment was conducted. Experimental conditions: pressing speed 5 mm / s, target pressing force 50 kN, target pressing position 20 mm, target pressing energy 300 J. 1000 pressing experiments were performed using traditional PID control, digital twin MPC control, and the method of this invention, respectively. The results are as follows:

[0039] Experimental results show that the method of this invention significantly outperforms traditional methods and existing digital twin methods in all performance indicators, fully meeting the ultra-precision requirements of turbine blade tenon press-fitting for aero-engines. In particular, the early anomaly detection accuracy reaches 99.9%, avoiding batch scrapping due to workpiece defects and saving enterprises a significant amount of costs.

[0040] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A closed-loop control method for position and pressure of a servo press machine, characterized in that, Includes the following steps: S1 Initial Model Pre-training and System Calibration: Construct a general initial model that includes basic mechanical dynamics and servo system characteristics, collect press-fitting data of different batches of standard workpieces, complete the static calibration and dynamic response identification of the system, and establish a material property benchmark database; S2 Online Real-Time Identification of Single Workpiece Characteristics in the Initial Stage of Press Fitting: During the elastic deformation stage of press fitting, force-displacement data for the first 50ms is collected at a frequency of 10kHz. The recursive least squares method with forgetting factor is used to fit the three constitutive parameters of the current workpiece in real time: elastic modulus, yield strength and hardening index, to generate a personalized constitutive model for the workpiece. S3 Adaptive Predictive Control Based on Personalized Constitutive Model: The constitutive parameters identified in real time are input into the model predictive controller to solve the optimal pressing trajectory and control sequence of the current workpiece online. At the same time, the pressing energy is used as the third control variable to realize the three-variable coordinated constraint control of position, pressure and energy. S4 High-Order Disturbance Observer and Hierarchical Error Compensation: A high-order disturbance observer is designed to estimate unmodeled dynamics and external disturbances in the system in real time. A three-layer compensation architecture of "identification feedforward-predictive control-microfeedback" is adopted to compensate for model error, dynamic error and residual error respectively. S5 Early Anomaly Detection and Classification Handling Based on Constitutive Parameters: The constitutive parameters identified in real time are compared with the material property benchmark database. Anomalies such as workpiece defects and material non-compliance are identified within 30% of the stroke before the pressing is completed. Based on the anomaly level, parameter self-adjustment, deceleration warning or emergency shutdown are executed. S6 Federated Learning-Driven Full Lifecycle Model Evolution: After each compression is completed, constitutive parameters, compression data, and quality results are uploaded to the cloud-based federated learning platform. Without leaking local data, the global initial model is collaboratively optimized, enabling model evolution across multiple devices and scenarios.

2. The position and pressure closed-loop control method for a servo press machine according to claim 1, characterized in that, The online real-time identification of single workpiece characteristics in step S2 specifically refers to: After the pressure head contacts the workpiece, S21 starts the characteristic identification program when the pressure reaches the preset threshold F0 (usually 1% to 2% of the rated pressure). The S22 continuously acquires force-displacement data for 50ms at a sampling frequency of 10kHz, obtaining 500 sets of sample points. S23 establishes a linearized model of the force-displacement relationship based on the theory of elastic-plastic deformation: force equals equivalent stiffness multiplied by displacement plus initial force; S24 uses a recursive least squares method with a forgetting factor to estimate the equivalent stiffness and initial force in real time, and converts them into elastic modulus and yield strength. Based on the subsequent 10ms of plastic deformation data, S25 fitted the hardening index and completed the construction of the personalized constitutive model.

3. The position and pressure closed-loop control method for a servo press machine according to claim 1, characterized in that, The objective function for the position-pressure-energy three-variable coordinated control described in step S3 is: Minimize the weighted sum of squared errors over the next Np control cycles. The error term includes position error, pressure error, and energy error. At the same time, a weighted penalty term for the control quantity is added. Among them, the pressure-fitting energy is defined as the integral of force with respect to displacement; The constraints include: upper and lower limits of the control quantity, upper and lower limits of the rate of change of the control quantity, maximum pressing force limit, and maximum pressing energy limit.

4. The position and pressure closed-loop control method for a servo press machine according to claim 1, characterized in that, The design of the higher-order perturbation observer in step S4 is as follows: An extended state equation containing system state and disturbance terms is constructed, and the system state variables and total disturbance are estimated in real time using the observer gain matrix and disturbance gain matrix. The bandwidth of the disturbance observer is set to 5 to 10 times the control bandwidth, enabling real-time estimation of all disturbances with frequencies below 500 Hz.

5. The position and pressure closed-loop control method for a servo press machine according to claim 1, characterized in that, The three-layer compensation architecture described in step S4 is as follows: Identify feedforward compensation: Based on the constitutive model identified in real time, calculate the feedforward control quantity to compensate for more than 90% of the model error, with a compensation frequency of 1 time per press. Predictive control compensation: Based on model predictive control, it compensates for system dynamic errors and slow-varying disturbances at a frequency of 1000Hz. Micro-feedback compensation: A high-frequency PI controller is used to compensate for residual errors and rapid disturbances, with a compensation frequency of 10000Hz.

6. The position and pressure closed-loop control method for a servo press machine according to claim 1, characterized in that, The early anomaly detection based on constitutive parameters mentioned in step S5 specifically refers to: S51 establishes a material property benchmark database, which includes the range of elastic modulus, yield strength, and hardening index of qualified workpieces; S52 compares the constitutive parameters identified in real time with the reference range; If the parameter exceeds the reference range by less than 10%, it is judged as a slight deviation and the control parameter is automatically adjusted. If the parameters of S54 exceed the reference range by 10% to 30%, it is judged as a moderate abnormality, an early warning is issued and the pressing speed is reduced. If the parameters of S55 exceed the reference range by more than 30%, it is judged as a serious abnormality and an emergency shutdown is initiated immediately.

7. The position and pressure closed-loop control method for a servo press machine according to claim 1, characterized in that, The federated learning-driven full lifecycle model evolution described in step S6 specifically refers to: After each press-fitting process, the S61 local device generates a local model update quantity and does not transmit the original production data. The S62 cloud server aggregates local updates from multiple devices and uses a federated averaging algorithm to update the global initial model. The updated global initial model is downloaded locally on the S63 device and used for feature identification in the next press-fitting process. S64 performs a global model evaluation every 1000 press cycles. If the model accuracy improves by more than 1%, it triggers a model update for all devices.

8. The position and pressure closed-loop control method for a servo press machine according to claim 1, characterized in that, This method is applicable to scenarios such as press-fitting of automotive parts, press-fitting of new energy vehicle battery modules, press-fitting of precision bearings, and press-fitting of aerospace parts. The press-fitting position accuracy is ≤ ±0.5μm, the press-fitting force accuracy is ≤ ±0.05%FS, the press-fitting energy accuracy is ≤ ±0.2%, and the early anomaly detection accuracy is ≥99.9%.

Images