A dynamic load cooperative control method for a robot carbon fiber joint

Abstract

The application relates to the technical field of robot joint control, and specifically discloses a dynamic load cooperative control method for a robot carbon fiber joint, constructs a flexible dynamics model of the carbon fiber joint containing online parameter updating, and corrects model mismatch caused by material property change; a generalized momentum external load dynamic observer independent of acceleration signals is designed to robustly estimate external dynamic load torque; a load multi-joint cooperative sharing function is realized by combining a Jacobian matrix and a load balancing factor in an upper cooperative planning layer; and load torque feedforward compensation is introduced in a lower model predictive control layer to realize accurate tracking of an expected instruction. The application improves model adaptability, load observation accuracy and control stability, can adapt to the flexibility of the carbon fiber joint and time-varying characteristics of the load, enhances the operation stability and load adaptability of the robot, and is suitable for various multi-joint robots with carbon fiber joints.

Description

Technical Field

[0001] This invention relates to the field of robot joint control technology, specifically to a method for dynamic load coordination control of a robot carbon fiber joint. Background Technology

[0002] As robotics technology advances towards higher precision, higher load capacity, and greater flexibility, the conflict between lightweight joint structures and rigidity requirements is becoming increasingly prominent. Carbon fiber, with its advantages of low density, high specific strength, and high specific stiffness, is widely used in the manufacture of robot joints, effectively reducing joint inertia and improving the robot's motion response speed and load capacity. However, carbon fiber itself possesses certain flexibility, and its mechanical properties are easily affected by factors such as temperature and load changes, leading to significant flexible deformation in carbon fiber joints, which in turn affects the joint's control precision.

[0003] Meanwhile, when robots perform complex tasks, each joint is subjected to time-varying external dynamic loads, and the distribution of these loads among multiple joints is uncertain. Traditional robot joint control methods are mostly based on the assumption of rigid joints, failing to fully consider the flexible characteristics of carbon fiber joints, and neglecting multi-joint collaborative load distribution for dynamic loads. This leads to problems such as large trajectory tracking errors, joint vibrations, and uneven load distribution during the control process, severely affecting the robot's operational stability and control accuracy.

[0004] In existing technologies, control methods for flexible joints mostly employ independent control strategies for individual joints, lacking load coordination mechanisms among multiple joints and failing to achieve reasonable distribution of dynamic loads across multiple joints. Furthermore, external load observations largely rely on acceleration signals, making observation accuracy susceptible to noise interference and exhibiting poor robustness. Simultaneously, they do not consider model mismatch issues caused by variations in carbon fiber material properties, resulting in insufficient control adaptability. Therefore, there is an urgent need for a method that can balance the flexibility of carbon fiber joints, robustness of dynamic load observation, and multi-joint collaborative control to address the aforementioned technical challenges. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method for dynamic load collaborative control of carbon fiber joints of robots. By constructing an accurate flexible dynamic model, a robust dynamic load observer, and an upper and lower layer collaborative control architecture, the method realizes the collaborative sharing of dynamic load among multiple joints and the accurate tracking of joint motion, thereby improving the stability and control accuracy of robot operation.

[0006] To solve the above-mentioned technical problems, the technical solution provided by the present invention is: a method for dynamic load cooperative control of a robot carbon fiber joint, comprising the following steps:

[0007] Step S1: Construct a flexible dynamic model of the carbon fiber joint, equating a single joint to a two-mass system containing nonlinear springs and damping, and establishing a flexible joint dynamic equation that includes joint position-related stiffness coefficients and joint angular velocity-related damping coefficients.

[0008] Step S2: Design an external load dynamic observer based on generalized momentum, collect joint motor current and position information, and estimate the external dynamic load torque acting on each joint in real time by combining the dynamic model of step S1.

[0009] Step S3: The upper-level collaborative planning layer performs load adaptive task allocation, taking the dynamic load torque observed in step S2 as input, and dynamically adjusting the desired motion trajectory and desired impedance parameters of each joint according to the overall kinematics and dynamics of the robot, so as to realize the collaborative sharing of dynamic load among multiple joints.

[0010] Step S4: The lower-level model predictive control layer performs precise tracking control, receives the expected command issued in step S3, constructs a model predictive controller based on the flexible dynamic model in step S1, and substitutes the load torque observed in step S2 as a feedforward compensation quantity into the predictive model to solve for the optimal control quantity and apply it to the joint motor.

[0011] Furthermore, in step S1, constructing the flexible dynamic model of the carbon fiber joint further includes: obtaining the initial values ​​of the joint stiffness coefficient and damping coefficient through offline identification; and introducing an online parameter update mechanism to use real-time feedback data to perform online correction of the stiffness coefficient using the recursive least squares method to compensate for the model mismatch caused by changes in the properties of the carbon fiber material.

[0012] Furthermore, the generalized momentum observer in step S2 is constructed without relying on the acceleration signal. It achieves robust estimation of the external dynamic load torque acting on the joint by integrating the difference between the motor torque and the internal torque calculated by the dynamic model.

[0013] Furthermore, the upper-layer collaborative planning layer in step S3 performs load adaptive task allocation, specifically including:

[0014] Based on the overall Jacobian matrix of the robot, the generalized force required by the end effector is dynamically redistributed to each joint;

[0015] A load balancing factor is introduced and calculated. The load balancing factor is adjusted in real time according to the ratio of the current load torque of each joint to the rated load capacity. It is used to dynamically adjust the output weight of each joint in the cooperative motion and the expected trajectory compensation amount.

[0016] Based on load changes and the load balancing factor, the desired stiffness and desired damping of each joint sent to the lower-level controller are dynamically adjusted.

[0017] Furthermore, the model predictive controller in step S4 has an optimization objective function that includes at least: a tracking error term between the actual and desired joint positions, a penalty term for the rate of change of the control variable, and a system state deviation term in the prediction time domain.

[0018] Furthermore, in step S4, the observed load torque is used as the feedforward compensation amount of the MPC controller. Specifically, the torque is added to the state equation as a known external disturbance in the prediction model of the MPC, so that the controller can generate a control output that resists the disturbance in advance when solving the optimal control sequence.

[0019] The advantages of this invention compared to the prior art are:

[0020] This invention constructs an accurate dynamic model of carbon fiber joint flexibility, introduces an online parameter update mechanism, and uses the recursive least squares method to correct the stiffness coefficient in real time, effectively compensating for model mismatch caused by changes in carbon fiber material properties, and improving the accuracy and adaptability of the model.

[0021] This invention designs an external load dynamic observer based on generalized momentum, which does not rely on acceleration signals and avoids noise interference caused by acceleration measurement. By integrating the difference between motor torque and internal torque and combining it with error feedback correction, it achieves robust estimation of external dynamic load torque, with high observation accuracy and strong anti-interference ability.

[0022] This invention adopts a multi-layer collaborative control architecture. The upper layer achieves dynamic load sharing among multiple joints through load adaptive task allocation, combined with Jacobian matrix and load balancing factor, thus avoiding overload of a single joint. The lower layer achieves precise tracking of joint motion through model predictive control combined with feedforward compensation, thereby improving control accuracy and system stability.

[0023] The control method of this invention can fully adapt to the flexibility of carbon fiber joints and the time-varying characteristics of external dynamic loads, significantly improving the robot's operational stability, load adaptability and control accuracy. It is applicable to a variety of multi-joint robots using carbon fiber joints and has broad application prospects. Attached Figure Description

[0024] Figure 1 This is a flowchart of a dynamic load collaborative control method for a robot carbon fiber joint according to the present invention. Detailed Implementation

[0025] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present invention.

[0026] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0027] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0028] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0029] The following detailed description of the dynamic load collaborative control method for a robot carbon fiber joint according to the present invention, with reference to the accompanying drawings, provides further details.

[0030] Combined with appendix Figure 1 The specific implementation process of the dynamic load collaborative control method for a robot carbon fiber joint of the present invention is as follows:

[0031] Step S1: Construct a flexible dynamic model of the carbon fiber joint

[0032] The flexibility of carbon fiber joints primarily stems from the internal carbon fiber transmission structure and connecting components. Its flexibility can be equivalently represented as a two-mass system incorporating nonlinear springs and damping; that is, the rotor on the motor side of the joint is one mass block, and the load side is another mass block, connected by a nonlinear flexible element (spring + damping). Based on this equivalent model, the dynamic equations of the flexible joint, including joint position-related stiffness coefficients and joint angular velocity-related damping coefficients, are established as follows:

[0033] ;

[0034] ;

[0035] in: Moment of inertia on the motor side (unit: ); Load-side moment of inertia (unit: ); These are the angular position, angular velocity, and angular acceleration of the motor-side joint (units: rad, rad / s, rad / s). 2 ); These are the angular position, angular velocity, and angular acceleration of the load-side joint (units: rad, rad / s, rad / s). 2 ); Motor-side damping coefficient (unit: ); Load-side damping coefficient (unit: ); This is the nonlinear spring stiffness term. Stiffness coefficients related to joint position (unit: Its value varies with joint angular displacement. change; For nonlinear damping terms, Damping coefficient related to joint angular velocity (unit: Its value varies with the difference in joint angular velocity. change; The output torque of the motor (unit: N·m); The external dynamic load torque acting on the joint (unit: N·m).

[0036] To improve the accuracy of the model, joint stiffness coefficients need to be obtained through offline identification. and damping coefficient The initial value is determined. Offline identification employs the step response method, applying a step torque signal to the joint motor and acquiring angular position and angular velocity data from both the motor and load sides. The data is then fitted using the least squares method to obtain the initial value. and initial value and .

[0037] Because the mechanical properties of carbon fiber materials are easily affected by factors such as temperature and load changes, the stiffness coefficient can be affected. Changes can lead to model mismatch. Therefore, an online parameter update mechanism is introduced, utilizing real-time feedback data on motor current, angular position, and angular velocity, and employing the recursive least squares (RLS) method to update the stiffness coefficients. Perform online correction, damping coefficient Because the change is small, the initial values ​​of the offline identification can be kept unchanged. The update formula for the recursive least squares method is as follows:

[0038] ;

[0039] ;

[0040] in: For the first Estimates of the stiffness coefficient at any given time; For the first Estimates of the stiffness coefficient at any given time; For the first The covariance matrix at time t; For regression vectors; This is the observation error; Forgetting factor ( ), used to balance the weights of historical and current data, typically taking .

[0041] Step S2: Design an external load dynamic observer based on generalized momentum

[0042] External dynamic load torque The load torque is a key factor affecting the accuracy of joint control, but it is difficult to measure directly. Therefore, a robust load observer needs to be designed for real-time estimation. The generalized momentum observer designed in this invention does not rely on acceleration signals, thus avoiding noise interference from acceleration measurements. It achieves robust estimation of the external dynamic load torque by integrating the difference between the motor torque and the internal torque calculated from the dynamic model.

[0043] First, we define the generalized momentum of a robot joint. for: ;

[0044] For generalized momentum Differentiating and combining this with the dynamic equations of the flexible joint in step S1, we can obtain:

[0045] ;

[0046] The external load torque can be obtained by sorting. The expression: ;

[0047] Due to generalized momentum derivative Since direct measurement is difficult, an observer is constructed using an integral approximation method. The specific observation formula is as follows:

[0048] ;

[0049] After simplification, we get:

[0050] ;

[0051] in: For the first The estimated value of the external dynamic load torque at any given time; For the first The estimated value of the external dynamic load torque at any given time; Control period (unit: seconds); For the first The observed value of generalized momentum at time t.

[0052] To improve the robustness of the observer, an observation error feedback correction term is introduced. The corrected observation formula is as follows:

[0053] ;

[0054] in: This is the observer gain (a positive number), used to adjust the convergence speed of the observation error; This is the observation error of the load side angle position. This is the estimated value of the load side angle position calculated from the dynamic model.

[0055] In practical applications, sensors are used to collect the joint motor current (used to calculate the motor's output torque). ,in The torque constant of the motor. (motor current) and joint angle position information (acquired via encoder) and By combining the modified flexible dynamic model from step S1 with the above observation formula, the external dynamic load torque acting on each joint can be estimated in real time. .

[0056] Step S3: The upper-layer collaborative planning layer performs load adaptive task allocation.

[0057] The core function of the upper-level collaborative planning layer is to take the dynamic load torque observed in step S2 as input, and dynamically adjust the desired motion trajectory and desired impedance parameters of each joint according to the overall kinematics and dynamics of the robot. This achieves collaborative load sharing among multiple joints, avoids overload on individual joints, and improves the overall load adaptability of the robot. The specific implementation process is as follows:

[0058] S3.1 Generalized Force Dynamic Redistribution

[0059] Based on the overall Jacobian matrix of the robot, the generalized forces (including force and torque) required by the end effector are dynamically redistributed to each joint. Let the generalized force vector of the robot's end effector be... (3 translational force components and 3 torque components), the Jacobian matrix of the robot is: ( Given the number of robot joints, the expected load torque vector for each joint is... It can be calculated using the following formula:

[0060] ;

[0061] in, Jacobian matrix transpose, Let be the angular position vectors of each joint of the robot.

[0062] S3.2 Load Balancing Factor Calculation

[0063] Introducing a load balancing factor This is used to dynamically adjust the force output weight and desired trajectory compensation of each joint during coordinated motion. The load balancing factor is adjusted in real time based on the ratio of the current load torque to the rated load capacity of each joint. The specific calculation formula is as follows:

[0064] ;

[0065] in: For the first The load balancing factor of each joint satisfies ; For the first Rated load torque of each joint (unit: N·m); For the first Estimates of the external dynamic load torque of each joint (obtained from step S2).

[0066] When the current load torque of a joint approaches its rated load capacity, the load balancing factor of that joint... The load will decrease, and its output weight will decrease; conversely, the load balancing factor will increase for joints with lighter loads. It will increase, its output weight will increase, thereby achieving a balanced distribution of load among multiple joints.

[0067] S3.3 Expected Parameter Dynamic Adjustment

[0068] Based on load changes and load balancing factors, dynamically adjust the desired stiffness of each joint sent to the lower-level controller. and expected damping This allows the joint's impedance characteristics to adapt to changes in dynamic load, improving control stability. The adjustment formulas for desired stiffness and desired damping are as follows:

[0069] ;

[0070] ;

[0071] in: The first Initial desired stiffness and initial desired damping of each joint; , These are the stiffness adjustment coefficient and the damping adjustment coefficient (positive numbers), used to adjust the degree of influence of load changes on the desired impedance parameters. They are typically taken as... ; For the first Load rate of each joint.

[0072] Simultaneously, the desired motion trajectory of each joint is adjusted according to the load balancing factor. For joints with heavy loads, trajectory compensation is performed to prevent increased trajectory tracking errors due to excessive load. The compensation formula is as follows:

[0073] ;

[0074] ;

[0075] in: For the first The initial expected trajectory of each joint; For the first The trajectory compensation amount of each joint; Trajectory compensation coefficient (unit: rad); This is a sign function used to determine the direction of trajectory compensation.

[0076] Step S4: The lower-level model predictive control layer performs precise tracking control.

[0077] The lower-level Model Predictive Control (MPC) layer receives the desired instructions from step S3, including the desired motion trajectory. Desired stiffness and expected damping Based on the flexible dynamic model in step S1, a model predictive controller is constructed, and the load torque observed in step S2 is used as the input. The feedforward compensation is substituted into the prediction model to solve for the optimal control quantity, which is then applied to the joint motor to achieve precise tracking of joint motion.

[0078] S4.1 Model Predictive Controller Construction

[0079] First, the dynamic equations of the flexible joint in step S1 are transformed into a state-space model. The system state vector is defined. , control input vector External disturbance vector Output vector Then the state-space model is:

[0080] ;

[0081] ;

[0082] Where matrices A, B, C, and D are respectively:

[0083] ;

[0084] ;

[0085] ;

[0086] ;

[0087] Since robot joint control uses discretized control, the above continuous state-space model is discretized using the zero-order hold method to obtain a discrete state-space model:

[0088] ;

[0089] ;

[0090] in, , , , , To control the cycle.

[0091] S4.2 Optimization Objective Function Design

[0092] The core of model predictive controllers is to predict the system state and the predictive model in the time domain within each control cycle. and control time domain ( Solve for the optimal control sequence within the time domain to make the system output track the desired trajectory. The optimization objective function should include at least the tracking error term between the actual and desired joint positions, the penalty term for the rate of change of the control variable, and the system state deviation term in the prediction time domain. The specific expression is as follows:

[0093] ;

[0094] in: This is the current control moment; To predict the first in the time domain The system output predicted value for the step; To predict the first in the time domain The expected output value of the step (determined by the expected trajectory issued in step S3); To control the rate of change of the quantity; To predict the first in the time domain The system state prediction value of the step; To predict the first in the time domain The expected state value of the step; These are the output tracking error weight matrix, the control variable change rate penalty weight matrix, and the state deviation weight matrix, respectively. They are all positive definite symmetric matrices used to adjust the weights of each objective item.

[0095] S4.3 Feedforward Compensation and Solution of Optimal Control Quantity

[0096] The load torque observed in step S2 As a feedforward compensation for the MPC controller, specifically, this torque is treated as a known external disturbance term in the MPC prediction model. By incorporating state equations, the controller can generate control outputs that resist the disturbance in advance when solving for the optimal control sequence, thereby reducing the impact of load disturbances on trajectory tracking.

[0097] Within each control cycle, the controller adjusts the current system state. Predictive models, expected commands, and external load disturbances Under the condition of satisfying the control quantity constraints ( ,in , These are the minimum and maximum values ​​of the motor output torque, respectively, and the state constraints. Under the premise of ), we solve for the minimum value of the above optimization objective function to obtain the optimal control sequence:

[0098] .

[0099] The rolling optimization strategy using model predictive control only optimizes the first control input of the optimal control sequence. The system is applied to the joint motor, and in the next control cycle, the system state and external load observations are updated. The above optimization process is repeated to achieve dynamic tracking control.

[0100] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for dynamic load coordinated control of a robot carbon fiber joint, characterized in that, Includes the following steps: Step S1: Construct a flexible dynamic model of the carbon fiber joint, equating a single joint to a two-mass system containing nonlinear springs and damping, and establishing a flexible joint dynamic equation that includes joint position-related stiffness coefficients and joint angular velocity-related damping coefficients. Step S2: Design an external load dynamic observer based on generalized momentum, collect joint motor current and position information, and estimate the external dynamic load torque acting on each joint in real time by combining the dynamic model of step S1. Step S3: The upper-level collaborative planning layer performs load adaptive task allocation, taking the dynamic load torque observed in step S2 as input, and dynamically adjusting the desired motion trajectory and desired impedance parameters of each joint according to the overall kinematics and dynamics of the robot, so as to realize the collaborative sharing of dynamic load among multiple joints. Step S4: The lower-level model predictive control layer performs precise tracking control, receives the expected command issued in step S3, constructs a model predictive controller based on the flexible dynamic model in step S1, and substitutes the load torque observed in step S2 as a feedforward compensation quantity into the predictive model to solve for the optimal control quantity and apply it to the joint motor.

2. The method for dynamic load coordinated control of a robot carbon fiber joint according to claim 1, characterized in that: In step S1, constructing the flexible dynamic model of the carbon fiber joint further includes: obtaining the initial values ​​of the joint stiffness coefficient and damping coefficient through offline identification; and introducing an online parameter update mechanism to use real-time feedback data to perform online correction of the stiffness coefficient using the recursive least squares method to compensate for the model mismatch caused by changes in the properties of carbon fiber materials.

3. The method for dynamic load coordinated control of a robot carbon fiber joint according to claim 2, characterized in that: The generalized momentum observer in step S2 is constructed without relying on acceleration signals. It achieves robust estimation of the external dynamic load torque acting on the joint by integrating the difference between the motor torque and the internal torque calculated by the dynamic model.

4. The method for dynamic load coordinated control of a robot carbon fiber joint according to claim 3, characterized in that: The upper-layer collaborative planning layer in step S3 performs load adaptive task allocation, specifically including: Based on the overall Jacobian matrix of the robot, the generalized force required by the end effector is dynamically redistributed to each joint; A load balancing factor is introduced and calculated. The load balancing factor is adjusted in real time according to the ratio of the current load torque of each joint to the rated load capacity. It is used to dynamically adjust the output weight of each joint in the cooperative motion and the expected trajectory compensation amount. Based on load changes and the load balancing factor, the desired stiffness and desired damping of each joint sent to the lower-level controller are dynamically adjusted.

5. The method for dynamic load coordinated control of a robot carbon fiber joint according to claim 4, characterized in that: The model predictive controller in step S4 has an optimization objective function that includes at least: a tracking error term between the actual and desired joint positions, a penalty term for the rate of change of the control variable, and a system state deviation term in the prediction time domain.

6. The method for dynamic load coordinated control of a robot carbon fiber joint according to claim 5, characterized in that: In step S4, the observed load torque is used as the feedforward compensation amount of the MPC controller. Specifically, the torque is added to the state equation as a known external disturbance in the prediction model of the MPC, so that the controller can generate a control output that resists the disturbance in advance when solving the optimal control sequence.

Images