A load force control method of servo torque feedforward and linear active disturbance rejection control

Abstract

The application provides a loading force control method of servo torque feedforward and linear active disturbance rejection control, and belongs to the technical field of servo hydraulic equipment control. The method comprises the following steps: constructing a servo hydraulic tensile pad dynamics model, building a simulation model of the servo tensile pad based on the servo tensile pad dynamics model and structural design parameters of a servo tensile pad test bench; collecting simulation data sets and experimental data sets, merging, weighting and preprocessing the simulation data sets and the experimental data sets to obtain a standardized feature matrix; constructing an XGBoost-based servo feedforward torque prediction model, taking the standardized feature matrix as the model input and outputting the feedforward torque; constructing a servo tensile pad second-order system model; combining the feedforward torque output by the servo feedforward torque prediction model and the feedback torque calculated by the linear active disturbance rejection control to form the final torque instruction of the servo motor. The application improves the control precision and robustness of the servo hydraulic tensile pad in the production process.

Description

Technical Field

[0001] This application belongs to the field of servo hydraulic equipment control technology, specifically relating to a loading force control method with servo torque feedforward and linear active disturbance rejection control. Background Technology

[0002] As the manufacturing industry moves towards intelligence and high precision, stamping technology, as one of the core processes in automobile manufacturing, aerospace, and other fields, plays a vital role in equipment manufacturing. As the core actuator of stamping equipment, its essence is to achieve a highly dynamic force control system. The core objective of this system is traditionally defined as the precise control of the applied force to suppress wrinkling of the sheet metal, which is crucial for ensuring the quality of stamped parts.

[0003] For the operating conditions of servo stretching pad equipment, it is necessary to consider the nonlinear factors of the hydraulic and mechanical systems, including the dead zone and hysteresis of proportional valves, the compressibility of hydraulic fluid, internal leakage of hydraulic cylinders, and nonlinear friction of piston-gear-rack cylinders, as well as factors such as oil temperature changes during actual operation. For servo stretching pad stamping systems, it is difficult to establish accurate models. Traditional control algorithms, such as PID control, cannot effectively solve the complex hydraulic system control, and the slow response speed of hydraulic systems makes it difficult to achieve precise dynamic tracking and control of the applied force for high-speed stamping actions. Furthermore, model predictive control relies on accurate physical models; for the current equipment and operating conditions, it is impossible to accurately model the equipment, and the uncertainty of the model directly leads to a sharp decline in the performance of these algorithms, resulting in a lack of robustness when applied to industrial equipment.

[0004] Most existing control schemes are based on traditional feedback control, which uses real-time monitored loading force to drive the lower die movement through feedback adjustment during real-time control. The shortcomings of current control schemes are: firstly, they ignore the fact that the stamping process is a coupled process of upper and lower die forces and positions; secondly, they simply perform feedback control on the system, ignoring the slow response speed of hydraulic systems; and finally, the real-time control process of hydraulic systems is subject to various internal and external disturbances, and the real-time control system needs to remain robust against these disturbances. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of traditional control schemes, such as difficulty in system modeling, slow response, and weak anti-interference ability, which result in insufficient loading force control accuracy. This invention aims to improve the control accuracy and robustness of loading force in the production process of servo hydraulic tension pads by proposing a loading force control method based on servo torque feedforward and linear active disturbance rejection control.

[0006] The method includes the following steps:

[0007] S1. Construct a dynamic model of the servo hydraulic tension pad. Based on the dynamic model of the servo tension pad and the structural design parameters of the servo tension pad test bench, build a simulation model of the servo tension pad using simulation tools.

[0008] S2. Collect simulation model data of servo tensile pad as simulation dataset, collect test bench data of servo tensile pad as experimental dataset, merge, weight and preprocess the simulation dataset and experimental dataset to obtain the standardized feature matrix;

[0009] S3. Construct a servo feedforward torque prediction model based on XGBoost, using the standardized feature matrix as the model input and outputting the feedforward torque;

[0010] S4. Construct a second-order system model of the servo tensioning pad, design a discrete linear tracking differentiator (LTD) through linear active disturbance rejection control (LADRC) to process the target loading force input signal, design a linear extended observer (LESO) to observe the total disturbance as a new state variable, and design a linear state error feedback control law (LSEF) to adjust the feedback torque output of the servo motor.

[0011] S5. The feedforward torque output by the servo feedforward torque prediction model is combined with the feedback torque calculated by the linear active disturbance rejection control (LADRC) to form the final torque command of the servo motor, which controls the loading force during the stamping process.

[0012] Furthermore, in step S1, constructing the dynamic model of the servo hydraulic tension pad includes constructing the dynamic equations, constructing the friction equations, and identifying the unknown parameters of the servo tension pad. The specific steps are as follows:

[0013] S11. Construct the dynamic equations:

[0014]

[0015]

[0016]

[0017]

[0018]

[0019]

[0020] In the formula, The main oil circuit oil pressure level, For the equivalent mass of the lower mold, For the displacement of the lower mold, For the lower die piston rod chamber hydraulic pressure, The effective working area of ​​the upper mold piston. For load capacity, For friction, The gear ratio of the reducer. For the output torque, The equivalent stiffness of the shaft is . This refers to the rotation angle of the servo motor. The rotation angle of the gear. n The transmission ratio is... The frictional torque of the gear and rack is... For the rotational inertia of the gear, For gear angular acceleration, For gear friction, For the gear radius, This refers to the oil pressure on the side of the gear and rack cylinder that is not connected to the main oil circuit. The other side of the gear and rack cylinder is the main oil circuit pressure. The effective area of ​​the gear and rack cylinder. The engagement angle, The internal friction of the gear and rack cylinder. The piston area of ​​the gear and rack cylinder. For gear rack displacement, To control the volume of the lower cavity of the lower mold piston. The initial volume of the lower cavity. This represents the internal leakage coefficient. For bulk modulus, The applied force is the control objective;

[0021] S12. Construct the friction force equation:

[0022]

[0023]

[0024] In the formula, Coulomb friction, The coefficient of viscosity is... The coulomb friction torque of the gear and rack is... The coefficient of viscous friction of the gear and rack;

[0025] S13. Based on the structural design parameters of the servo tensile testing bench, the unknown parameters of the servo tensile pad are identified.

[0026] Furthermore, in step S1, the simulation model of the servo stretching pad includes: the sliding module (1), the servo motor (2), the reducer (3), the main oil circuit (4), and the gear rack (5).

[0027] Furthermore, the specific steps of step S2 are as follows:

[0028] S21. Construct simulation datasets and experimental datasets through simulation and experimentation;

[0029] S22. Merge the simulation dataset and the experimental dataset to obtain the merged feature matrix and the merged label vector;

[0030] S23. Establish the sample weight vector;

[0031] S24. Standardize the merged data to obtain the standardized feature matrix:

[0032]

[0033] In the formula, The merged feature matrix, The feature mean vector, is the characteristic standard deviation vector.

[0034] Furthermore, the specific steps of step S3 are as follows:

[0035] S31. Construct the XGBoost model and weighted objective function;

[0036] S32. Approximate and optimize the weighted objective function to obtain the optimized objective function;

[0037] S33. Redefine the objective function according to the leaf node aggregation to obtain the final objective function;

[0038] S34. Solve for the optimal weight of each leaf node, and bring the optimal weight back into the final objective function to obtain the optimal score.

[0039] Furthermore, in step S3, the final objective function of the XGBoost model is:

[0040]

[0041] In the formula, G j For the gradient term associated with the j-th element, H j Let the coefficient of the quadratic term associated with the j-th element be . To control the score of the leaf nodes, The number of nodes in the tree. for Regularization coefficient, w j For the first One decision variable.

[0042] Furthermore, in step S3, the XGBoost model uses a greedy algorithm to construct the tree, and the splitting reward formula is:

[0043]

[0044] when If the reduction in loss is greater than the increase in complexity, then splitting occurs; when If so, it will not split.

[0045] Furthermore, step S3 also includes model validity verification:

[0046] The root mean square error (RMSE) of two modes—a baseline mode employing a feedback control strategy and a composite mode combining a servo torque feedforward prediction model with feedback control—is derived:

[0047]

[0048]

[0049] in, The root mean square error is the reference value. To control the root mean square error in a composite manner, The total number of test samples, Apply force to the target. The measured loading force for feedback control, The measured loading force under the feedforward-feedback composite control mode;

[0050] The model acceptance criterion inequality is:

[0051]

[0052] in, The coefficient is the determination coefficient; it satisfies the following conditions. This completes the verification of the model's effectiveness.

[0053] Furthermore, the specific steps of step S4 are as follows:

[0054] S41. Establish a second-order system model for the servo-driven tension pad:

[0055]

[0056] In the formula, The controlled variable is the loading force of the servo tension pad. This is a control quantity, specifically the output torque of the servo motor; External disturbance; To control the gain, specifically, how much pressure change rate can be generated by a unit torque command; The total disturbance includes both internal and external interference;

[0057] S42. Design a Discrete Linear Tracking Differentiator (LTD):

[0058]

[0059]

[0060] In the formula, Indicates the first The target applied force input signal with a sampling period of time express The applied force tracks the input signal over a sampling period of time. express The differential tracking signal of the applied force over a sampling period time To quickly track factors;

[0061] For the target loading force input signal First, the input signal is output as two discrete output signals through a discrete linear tracking differentiator. To track The output signal, for The differential tracking signal;

[0062] S43. Design a linearly extended observer (LESO). LESO will measure the total perturbation. To observe as a new state variable, define the state variable as follows: This represents the applied force value. For the rate of change of applied force, For total disturbance;

[0063] Constructing the LESO state-space equations:

[0064]

[0065]

[0066]

[0067]

[0068] In the formula, System status The estimated quantity, , For observer gain;

[0069] S44. Design a linear state error feedback control law LSEF:

[0070]

[0071]

[0072] in, For proportional gain, For the differential gain, for a second-order servo stretching pad system, and With controller bandwidth The relationship between them is derived using the pole placement method:

[0073]

[0074]

[0075] In the formula, For controller bandwidth;

[0076] S45. Adjust the feedback torque output of the servo motor according to the linear state error feedback control law LSEF.

[0077] Furthermore, in step S5, the final torque command of the servo motor includes the feedforward torque output by the servo stretching pad feedforward torque prediction model. Feedback torque calculated with LADRC The expression is:

[0078]

[0079] in, This represents the gain value of the feedforward torque.

[0080] As can be seen from the above technical solutions, the present invention has the following advantages:

[0081] In the servo torque feedforward and linear active disturbance rejection control loading force control method of this application, a servo torque feedforward prediction model is constructed, and simulation and experimental data are integrated to predict the required ideal torque of the servo motor as the feedforward torque output, thus avoiding the influence of hydraulic system nonlinearity on control. At the same time, combined with linear active disturbance rejection control (LADRC), internal and external disturbances are uniformly regarded as total disturbances and estimated and compensated in real time, avoiding the shortcomings of PID control in dealing with dynamic tracking of high-speed stamping. The feedforward control compensates for disturbances such as upper die motion coupling and slow hydraulic response in advance, while LADRC corrects model errors in real time. The composite control architecture of the two reduces loading force tracking error and has higher accuracy than single feedback control, meeting the requirements of high-precision stamping for loading force stability. Through the construction of dynamic and simulation models, data acquisition covering multiple working conditions is carried out, so that the feedforward model can be adapted to complex scenarios such as oil temperature changes and load fluctuations, ensuring that the servo hydraulic stretching pad maintains stable control performance under different production processes and expanding the application range of the equipment. This application overcomes the shortcomings of traditional control schemes, such as difficulty in system modeling, slow response, and weak anti-interference ability, which result in insufficient control accuracy of loading force, thereby improving the control accuracy and robustness of loading force of servo hydraulic tension pads during the production process. Attached Figure Description

[0082] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying 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.

[0083] Figure 1 This is a flowchart illustrating the loading force control method of servo torque feedforward and linear active disturbance rejection control in this application.

[0084] Figure 2 This is a schematic diagram of the servo hydraulic tension pad of this application.

[0085] Figure 3 This is a schematic diagram of the simulation model of the servo stretching pad in this application.

[0086] In the diagram, 1 is the sliding module, 2 is the servo motor, 3 is the reducer, 4 is the main oil circuit, and 5 is the gear rack. Detailed Implementation

[0087] The various embodiments of the invention will be described more fully in the detailed steps of the servo torque feedforward and linear active disturbance rejection control loading force control method described below. The invention may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of the invention to the specific embodiments disclosed herein, but rather the invention should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of the invention.

[0088] It should be understood that, when used in this specification, the term "comprising" indicates the presence of the described feature, integral, step, operation, element, and / or component, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0089] The terms "one embodiment" or "some embodiments" used in this application mean that one or more embodiments of this application include the specific features, structures, or characteristics described in that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this application do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized.

[0090] To make the objectives, features, and advantages of this invention more apparent and understandable, specific embodiments and accompanying drawings will be used to clearly and completely describe the technical solutions protected by this invention. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0091] like Figure 2As shown, for a servo-hydraulic stretching pad, the main body of the servo-hydraulic stretching pad includes an upper die slider and a lower die. The upper die slider is controlled by an oil pump connected to a servo proportional valve. The lower die consists of four piston cylinders, each connected to a rack and pinion cylinder via pipelines. The rack and pinion cylinder is controlled by a servo motor. During the stamping process, the system adopts a strategy of coordinated control of the upper and lower dies. The upper die slider acts as the active displacement unit, strictly following the motion path based on dynamics and process planning through closed-loop position control. The lower die is the force control actuator of the servo-hydraulic stretching pad. Its core function is to control the loading force by controlling the servo motor. The target force value in this process is the loading force, which is determined according to the production process and serves to suppress wrinkling of the sheet metal and ensure forming quality. The force control system of the servo-hydraulic stretching pad essentially converts the process target (loading force) into a loading force output that the system can accurately track and dynamically adjust.

[0092] Please see Figure 1 The diagram shows a flow chart of a loading force control method based on servo torque feedforward and linear active disturbance rejection control. The method includes:

[0093] S1. Construct a dynamic model of the servo hydraulic tension pad. Based on the dynamic model of the servo tension pad and the structural design parameters of the servo tension pad test bench, build a simulation model of the servo tension pad using simulation tools.

[0094] It should be noted that, firstly, the system composition of the servo hydraulic tension pad needs to be clearly defined, and then mathematical equations should be established based on physical laws to describe the dynamic characteristics and coupling relationships of each part. Based on the above dynamic model and the structural design parameters of the servo tension pad test bench, a suitable simulation tool (such as AMESim, MATLAB / Simulink, ADAMS, etc.) is selected to transform the mathematical equations into a visual modular connection model.

[0095] S2. Collect simulation model data of servo tensile pad as simulation dataset, collect test bench data of servo tensile pad as experimental dataset, merge, weight and preprocess the simulation dataset and experimental dataset to obtain the standardized feature matrix;

[0096] It should be noted that by merging simulation and experimental datasets, the data coverage is expanded, making up for the limitations of a single data source; weighted processing balances the importance of simulation and measured data, and standardized preprocessing eliminates differences in units and noise interference, thereby improving the model's generalization ability and prediction accuracy.

[0097] S3. Construct a servo feedforward torque prediction model based on XGBoost, using the standardized feature matrix as the model input and outputting the feedforward torque;

[0098] It should be noted that by leveraging the advantages of the XGBoost model in processing high-dimensional nonlinear data, an accurate feedforward torque prediction model can be trained using a standardized feature matrix. This model can output the compensation torque that is adapted to the operating conditions in advance, avoiding the limitations of traditional feedforward models that rely on accurate models, and providing a fast-response feedforward quantity for composite control.

[0099] S4. Construct a second-order system model of the servo tensioning pad, design a discrete linear tracking differentiator (LTD) through linear active disturbance rejection control (LADRC) to process the target loading force input signal, design a linear extended observer (LESO) to observe the total disturbance as a new state variable, and design a linear state error feedback control law (LSEF) to adjust the feedback torque output of the servo motor.

[0100] It should be noted that, through the LADRC architecture, based on a second-order model, LTD achieves smooth tracking of the target force, LESO observes and compensates for the total disturbance in real time, and LSEF adjusts the feedback torque. This approach can handle system nonlinearity and internal / external disturbances without relying on an accurate model, thus improving the accuracy of feedback control.

[0101] S5. The feedforward torque output by the servo feedforward torque prediction model is combined with the feedback torque calculated by the linear active disturbance rejection control (LADRC) to form the final torque command of the servo motor, which controls the loading force during the stamping process.

[0102] It should be noted that by using a combination of feedforward and feedback torque control, the feedforward compensates for known disturbances in advance to speed up the response, while the feedback corrects model errors and sudden disturbances in real time. The two work together to reduce the tracking error of the loading force and ensure the stability of the loading force during the stamping process.

[0103] Furthermore, as a refinement and extension of the specific implementation of the above embodiments, and to fully illustrate the specific implementation process in this embodiment, another loading force control method based on servo torque feedforward and linear active disturbance rejection control is provided, the method comprising:

[0104] S1. Construct a dynamic model of the servo hydraulic tension pad. Based on the dynamic model of the servo tension pad and the structural design parameters of the servo tension pad test bench, build a simulation model of the servo tension pad using simulation tools.

[0105] In step S1, constructing the dynamic model of the servo hydraulic tension pad includes constructing the dynamic equations, constructing the friction equations, and identifying the unknown parameters of the servo tension pad. The specific steps are as follows:

[0106] S11. Construct the dynamic equations:

[0107]

[0108]

[0109]

[0110]

[0111]

[0112]

[0113] In the formula, The main oil circuit oil pressure level, For the equivalent mass of the lower mold, For the displacement of the lower mold, For the lower die piston rod chamber hydraulic pressure, The effective working area of ​​the upper mold piston. For load capacity, For friction, The gear ratio of the reducer. For the output torque, The equivalent stiffness of the shaft is . This refers to the rotation angle of the servo motor. The rotation angle of the gear. n The transmission ratio is... The frictional torque of the gear and rack is... For the rotational inertia of the gear, For gear angular acceleration, For gear friction, For the gear radius, This refers to the oil pressure on the side of the gear and rack cylinder that is not connected to the main oil circuit. The other side of the gear and rack cylinder is the main oil circuit pressure. The effective area of ​​the gear and rack cylinder. The engagement angle, The internal friction of the gear and rack cylinder. The piston area of ​​the gear and rack cylinder. For gear rack displacement, To control the volume of the lower cavity of the lower mold piston. The initial volume of the lower cavity. This represents the internal leakage coefficient. For bulk modulus, The applied force is the control objective;

[0114] S12. Construct the friction force equation:

[0115]

[0116]

[0117] In the formula, Coulomb friction, The coefficient of viscosity is... The coulomb friction torque of the gear and rack is... The coefficient of viscous friction of the gear and rack;

[0118] S13. Based on the structural design parameters of the servo tensile testing bench, the unknown parameters of the servo tensile pad are identified.

[0119] like Figure 3 As shown, in step S1, the simulation model of the servo stretching pad includes: sliding module 1, servo motor 2, reducer 3, main oil circuit 4, and gear rack 5.

[0120] For example, based on the servo tensile pad dynamics model and the structural design parameters of the servo tensile testing bench, a simulation model is constructed using the AMESim and Simulink co-simulation platform. The specific steps are as follows:

[0121] In AMESim, a hydraulic-mechanical module for the main hydraulic circuit 4 and the gear and rack 5 is built. The main hydraulic circuit module configures the bulk modulus of hydraulic oil and the flow characteristics of the servo valve, and constructs a hydraulic cylinder force output model based on the effective working area of ​​the lower die piston. The gear and rack module integrates the gear rotational inertia, meshing angle, and cylinder working area to establish a force-displacement-torque coupling model. In Simulink, a control-transmission module for the servo motor 2 and the reducer 3 is built. The servo motor module embeds the rated torque and rotational inertia parameters, and the reducer module configures the transmission ratio. It also integrates the dynamic equation from step S11 and the friction equation from step S12 to realize the transmission of torque and speed. A co-simulation interface is set up to realize real-time interaction between AMESim hydraulic pressure, mechanical displacement, and Simulink motor control signals. The unknown parameters identified in step S13 are imported, and working conditions such as loading force and control current are set. After running the simulation, the displacement response and main hydraulic circuit pressure fluctuation characteristics of the upper sliding module 1 are analyzed.

[0122] It should be noted that the upper die slider, as an independent displacement control unit, will compress the sheet metal and cause the oil pressure in the lower die hydraulic cylinder to passively increase. This external action is both the cause of the force, but for the force control system of the lower die, this passively generated force is a major external disturbance that will disrupt the stability of the target loading force.

[0123] S2. Collect simulation model data of the servo tensile pad as the simulation dataset, and collect test bench data of the servo tensile pad as the experimental dataset. Merge, weight, and preprocess the simulation dataset and the experimental dataset to obtain the standardized feature matrix; the specific steps of step S2 are as follows:

[0124] S21. Construct simulation datasets and experimental datasets through simulation and experimentation. In the simulation dataset, the feature vector is... The label vector is In the experimental dataset, the feature vector is The label vector is ;

[0125]

[0126] in, The number of simulation samples, For the number of features, For the experimental sample size, and These are the label vectors for simulation and experiment, respectively, representing the feedforward compensation torque that the servo motor needs to output.

[0127] S22. Merge the simulation dataset and the experimental dataset to obtain the merged feature matrix. The label vector is ;

[0128] S23. Establish the sample weight vector ,in, These are the simulation weighting coefficients. These are the experimental weighting coefficients;

[0129] S24. Standardize the merged data:

[0130]

[0131] in, The standardized feature matrix, The feature mean vector, , The feature standard deviation vector, .

[0132] S3. Construct a servo feedforward torque prediction model based on XGBoost, using the standardized feature matrix as the model input and outputting the feedforward torque;

[0133] The XGBoost model is represented as a summation model of K regression trees CART:

[0134]

[0135] in, The number of regression trees, For the first A tree, The weighted objective function is the function space containing all regression trees:

[0136]

[0137] in, For loss function, For sample weights, For the first The regularization term for a tree is defined as:

[0138]

[0139] in, To control the score of the leaf nodes, The number of nodes in the tree. for Regularization coefficient, For the first The score of each node;

[0140] For the During step training, the objective function becomes:

[0141]

[0142] in, For constant terms;

[0143] make ,make Then the objective function is approximately:

[0144]

[0145] After removing the constant term, the objective function is simplified to obtain the optimized objective function:

[0146] ;

[0147] For a decision tree, consider it as a function First, the sample Mapped to a leaf node Then output the score of that leaf node. ;

[0148] Reorganize the summation in the above formula according to the leaf nodes, and assume... A leaf node, assigned to a leaf node The sample set is ;

[0149] Will and Substituting into the objective function, we get:

[0150]

[0151] make ,make ;

[0152] The final objective function is:

[0153] ;

[0154] For a tree with a fixed structure, find the optimal weight for each node. for:

[0155]

[0156] Substituting this back into the objective function, the optimal score is:

[0157] ;

[0158] It should be noted that XGBoost uses a greedy algorithm to construct the tree, and the splitting reward formula is as follows:

[0159]

[0160] In the formula, G L G is the left-hand gradient term. R H is the right-hand gradient term. L For the left-hand seasel term, H R For the right-hand sea-seal term, when If the reduction in loss is greater than the increase in complexity, then splitting occurs; when If so, it will not split.

[0161] Step S3 also includes model validity verification:

[0162] The root mean square error (RMSE) of two modes—a baseline mode employing a feedback control strategy and a composite mode combining a servo torque feedforward prediction model with feedback control—is derived:

[0163]

[0164]

[0165] in, The root mean square error is the reference value. To control the root mean square error in a composite manner, The total number of test samples, Apply force to the target. The measured loading force for feedback control, The measured loading force under the feedforward-feedback composite control mode;

[0166] The model acceptance criterion inequality is:

[0167]

[0168] in, The determination coefficient, If the above conditions are met, the validity of the model is verified.

[0169] The standardized feature matrix of step S3 includes the following eigenvalues: upper mold slider motion state, real-time system operating parameters, and target loading force.

[0170] It should be noted that an XGBoost-based machine learning model is trained as a disturbance observer to construct a servo torque feedforward prediction model. The inputs to this model are the motion state of the upper mold slider, the real-time operating parameters of the system, and the target loading force. The output is the feedforward compensation torque required to achieve the target loading force control. This torque, combined with the feedback torque calculated by the linear active disturbance rejection control, constitutes the final torque command of the servo motor, thereby achieving high-precision loading force control.

[0171] S4. Construct a second-order system model of the servo tensioning pad. Design a discrete linear tracking differentiator (LTD) using linear active disturbance rejection control (LADRC) to process the target loading force input signal. Design a linear extended observer (LESO) to observe the total disturbance as a new state variable. Design a linear state error feedback control law (LSEF) to adjust the feedback torque output of the servo motor. The specific steps of step S4 are as follows:

[0172] S41. Establish a second-order system model for the servo-driven tension pad:

[0173]

[0174] In the formula, The controlled variable is the loading force of the servo tension pad. This is a control quantity, specifically the output torque of the servo motor; External disturbance; To control the gain, specifically, how much pressure change rate can be generated by a unit torque command; The total disturbance includes both internal and external interference;

[0175] S42. Design a Discrete Linear Tracking Differentiator (LTD):

[0176]

[0177]

[0178] In the formula, Indicates the first The target applied force input signal with a sampling period of time express The applied force tracks the input signal over a sampling period of time. express The differential tracking signal of the applied force over a sampling period time To quickly track factors;

[0179] For the target loading force input signal First, the input signal is output as two discrete output signals through a discrete linear tracking differentiator. To track The output signal, for The differential tracking signal;

[0180] It should be noted that the input signal is first processed based on LTD, and the fast tracking factor is adjusted. Adjust the tracking speed. The larger the value, the faster the tracking speed; however, excessively fast tracking speeds can lead to drastic system response. The smaller the value, the slower the tracking speed, but the smoother the system response. Too small a value may result in a slow system response. Adjustments should be made during the test based on changes in the input signal and the system's response capability.

[0181] S43. Design a linearly extended observer (LESO). LESO will measure the total perturbation. To observe as a new state variable, define the state variable as follows: This represents the applied force value. For the rate of change of applied force, For total disturbance;

[0182] Constructing the LESO state-space equations:

[0183]

[0184]

[0185]

[0186]

[0187] In the formula, System status The estimated quantity, , For observer gain;

[0188] It should be noted that, using the pole placement method, all poles are placed at... ,in Given the observer bandwidth, the observation gain matrix can be obtained. . The larger the value, the faster the observation error converges, the stronger the ability to estimate disturbances, and the more rapid the system response. However, excessively large values... This will make the observer more sensitive to measurement noise, requiring... A balance is struck between dynamic response, speed, and noise immunity.

[0189] S44. Design a linear state error feedback control law LSEF:

[0190]

[0191]

[0192] in, For proportional gain, For the differential gain, for a second-order servo stretching pad system, and With controller bandwidth The relationship between them is derived using the pole placement method:

[0193]

[0194]

[0195] In the formula, For controller bandwidth;

[0196] It should be noted that, The larger the value, the faster the system response, but the more sensitive it is to noise. The smaller the value, the smoother the system will be, but the system's response speed will decrease. Therefore, a smaller value should be chosen. Start testing, gradually improve, and strike a balance between robustness and responsiveness.

[0197] S45. Adjust the feedback torque output of the servo motor according to the linear state error feedback control law LSEF.

[0198] It should be noted that Linear Active Disturbance Rejection Control (LADRC) does not rely on a precise mathematical model of the object. Instead, it uses the inner loop of a linearly extended state observer to treat all internal and external factors, such as system model uncertainty, slow hydraulic system response, and upper die motion coupling, as a total disturbance and performs real-time estimation and dynamic compensation. Meanwhile, the control law of the outer loop is responsible for quickly and accurately adjusting the torque output of the servo motor, so that the loading force applied to the sheet metal can be accurately stabilized at the target loading force set by the process.

[0199] S5. The feedforward torque output from the servo feedforward torque prediction model is combined with the feedback torque calculated by the linear active disturbance rejection control (LADRC) to form the final torque command of the servo motor, which controls the loading force during the stamping process; in step S5, the final torque command of the servo motor includes the feedforward torque output from the servo stretching pad feedforward torque prediction model. Feedback torque calculated with LADRC The expression is:

[0200]

[0201] in, This represents the gain value of the feedforward torque.

[0202] For example, the target loading force is set to 80kN, requiring precise adjustment of the servo hydraulic tension pad to avoid sheet metal forming defects. First, the XGBoost servo feedforward torque prediction model in step S3, based on current stamping parameters (upper die displacement speed 100mm / s, sheet metal thickness 0.8mm, oil temperature 45℃, etc.), outputs a feedforward torque of 120N·m to compensate for known disturbances such as upper die motion coupling and hydraulic response delay. Simultaneously, the LADRC controller in step S4 monitors the system in real time, using the linear expansion observer LESO to estimate the total disturbance (such as oil pressure fluctuations and sudden load changes). The feedback torque of 30N·m is calculated using the linear state error feedback control law LSEF to correct model errors and sudden disturbances. Combining process experience and system characteristics, the feedforward torque gain value k is adjusted to 0.95. According to the formula... The calculated final torque command is 144 N·m. This command drives the servo motor, and through the coordination of mechanical transmission and hydraulic system, the loading force stably tracks the target value of 80 kN, ensuring the forming quality of the deep drawing of the inner door panel.

[0203] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0204] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A loading force control method using servo torque feedforward and linear active disturbance rejection control, characterized in that, The method includes the following steps: S1. Construct a dynamic model of the servo hydraulic tension pad. Based on the dynamic model of the servo tension pad and the structural design parameters of the servo tension pad test bench, build a simulation model of the servo tension pad using simulation tools. S2. Collect simulation model data of servo tensile pad as simulation dataset, collect test bench data of servo tensile pad as experimental dataset, merge, weight and preprocess the simulation dataset and experimental dataset to obtain the standardized feature matrix; S3. Construct a servo feedforward torque prediction model based on XGBoost, using the standardized feature matrix as the model input and outputting the feedforward torque; S4. Construct a second-order system model of the servo tensioning pad, design a discrete linear tracking differentiator (LTD) through linear active disturbance rejection control (LADRC) to process the target loading force input signal, design a linear extended observer (LESO) to observe the total disturbance as a new state variable, and design a linear state error feedback control law (LSEF) to adjust the feedback torque output of the servo motor. S5. The feedforward torque output by the servo feedforward torque prediction model is combined with the feedback torque calculated by the linear active disturbance rejection control (LADRC) to form the final torque command of the servo motor, which controls the loading force during the stamping process.

2. The loading force control method of servo torque feedforward and linear active disturbance rejection control according to claim 1, characterized in that, In step S1, constructing the dynamic model of the servo hydraulic tension pad includes constructing the dynamic equations, constructing the friction equations, and identifying the unknown parameters of the servo tension pad. The specific steps are as follows: S11. Construct the dynamic equations: In the formula, The main oil circuit oil pressure level, For the equivalent mass of the lower mold, For the displacement of the lower mold, For the lower die piston rod chamber hydraulic pressure, The effective working area of ​​the upper mold piston. For load capacity, For friction, The gear ratio of the reducer. For the output torque, The equivalent stiffness of the shaft is . This refers to the rotation angle of the servo motor. The rotation angle of the gear. n The transmission ratio is... The frictional torque of the gear and rack is... For the rotational inertia of the gear, For gear angular acceleration, For gear friction, For the gear radius, This refers to the oil pressure on the side of the gear and rack cylinder that is not connected to the main oil circuit. The other side of the gear and rack cylinder is the main oil circuit pressure. The effective area of ​​the gear and rack cylinder. The engagement angle, The internal friction of the gear and rack cylinder. The piston area of ​​the gear and rack cylinder. For gear rack displacement, To control the volume of the lower cavity of the lower mold piston. The initial volume of the lower cavity. This represents the internal leakage coefficient. For bulk modulus, The applied force is the control objective; S12. Construct the friction force equation: In the formula, Coulomb friction, The coefficient of viscosity is... The coulomb friction torque of the gear and rack is... The coefficient of viscous friction of the gear and rack; S13. Based on the structural design parameters of the servo tensile testing bench, the unknown parameters of the servo tensile pad are identified.

3. The loading force control method with servo torque feedforward and linear active disturbance rejection control according to claim 1, characterized in that, In step S1, the simulation model of the servo stretching pad includes: the sliding module (1), the servo motor (2), the reducer (3), the main oil circuit (4), and the gear rack (5).

4. The loading force control method with servo torque feedforward and linear active disturbance rejection control according to claim 1, characterized in that, The specific steps of step S2 are as follows: S21. Construct simulation datasets and experimental datasets through simulation and experimentation; S22. Merge the simulation dataset and the experimental dataset to obtain the merged feature matrix and the merged label vector; S23. Establish the sample weight vector; S24. Standardize the merged data to obtain the standardized feature matrix: In the formula, The merged feature matrix, The feature mean vector, is the characteristic standard deviation vector.

5. The loading force control method with servo torque feedforward and linear active disturbance rejection control according to claim 1, characterized in that, The specific steps of step S3 are as follows: S31. Construct the XGBoost model and weighted objective function; S32. Approximate and optimize the weighted objective function to obtain the optimized objective function; S33. Redefine the objective function according to the leaf node aggregation to obtain the final objective function; S34. Solve for the optimal weight of each leaf node, and bring the optimal weight back into the final objective function to obtain the optimal score.

6. The loading force control method with servo torque feedforward and linear active disturbance rejection control according to claim 5, characterized in that, In step S3, the final objective function of the XGBoost model is: In the formula, G j For the gradient term associated with the j-th element, H j For the Hessian term associated with the j-th element, To control the score of the leaf nodes, The number of nodes in the tree. for Regularization coefficient, w j For the first One decision variable.

7. The loading force control method with servo torque feedforward and linear active disturbance rejection control according to claim 5, characterized in that, In step S3, the XGBoost model uses a greedy algorithm to construct the tree, and the splitting reward formula is: when If the reduction in loss is greater than the increase in complexity, then splitting occurs; when If so, it will not split.

8. The loading force control method of servo torque feedforward and linear active disturbance rejection control according to claim 5, characterized in that, Step S3 also includes model validity verification: The root mean square error (RMSE) of two modes—a baseline mode employing a feedback control strategy and a composite mode combining a servo torque feedforward prediction model with feedback control—is derived: in, The root mean square error is the reference value. To control the root mean square error in a composite manner, The total number of test samples, Apply force to the target. The measured loading force for feedback control, The measured loading force under the feedforward-feedback composite control mode; The model acceptance criterion inequality is: in, The coefficient is the determination coefficient; it satisfies the following conditions. This completes the verification of the model's effectiveness.

9. The loading force control method of servo torque feedforward and linear active disturbance rejection control according to claim 1, characterized in that, The specific steps of step S4 are as follows: S41. Establish a second-order system model for the servo-driven tension pad: In the formula, The controlled variable is the loading force of the servo tension pad. This is a control quantity, specifically the output torque of the servo motor; External disturbance; To control the gain, specifically, how much pressure change rate can be generated by a unit torque command; The total disturbance includes both internal and external interference; S42. Design a Discrete Linear Tracking Differentiator (LTD): In the formula, Indicates the first The target applied force input signal with a sampling period of time express The applied force tracks the input signal over a sampling period of time. express The differential tracking signal of the applied force over a sampling period time To quickly track factors; For the target loading force input signal First, the input signal is output as two discrete output signals through a discrete linear tracking differentiator. To track The output signal, for The differential tracking signal; S43. Design a linearly extended observer (LESO). LESO will measure the total perturbation. To observe as a new state variable, define the state variable as follows: This represents the applied force value. For the rate of change of applied force, For total disturbance; Constructing the LESO state-space equations: In the formula, System status The estimated quantity, , For observer gain; S44. Design a linear state error feedback control law LSEF: in, For proportional gain, For the differential gain, for a second-order servo stretching pad system, and With controller bandwidth The relationship between them is derived using the pole placement method: In the formula, For controller bandwidth; S45. Adjust the feedback torque output of the servo motor according to the linear state error feedback control law LSEF.

10. The loading force control method of servo torque feedforward and linear active disturbance rejection control according to claim 1, characterized in that, In step S5, the final torque command of the servo motor includes the feedforward torque output by the servo stretching pad feedforward torque prediction model. Feedback torque calculated with LADRC The expression is: in, This represents the gain value of the feedforward torque.

Images