Intelligent Control Method and System for Ball Screw Steering Based on Humanoid Robot

Abstract

This invention relates to an intelligent control method and system for ball screw steering based on a humanoid robot, belonging to the technical field of artificial intelligence systems in the production field. The method includes the following steps: pre-simulating the steering trajectory within a future time window in a digital twin; decomposing the steering trajectory into feedforward control quantities and, combined with real-time collected steering backlash error, calculating robust feedback compensation quantities using an adaptive sliding mode control method; at the critical moment when the ball screw is about to enter the steering dead zone, driving a dual-nut differential mechanism to unload and reconstruct the micron-level preload based on dynamic adjustment of the preload, to eliminate the hysteresis of the ball screw's elastic deformation; superimposing the feedforward control quantities and robust feedback compensation quantities and applying them to the servo drive unit to smoothly reverse the ball screw at the zero dead zone under load impact on the humanoid robot. The beneficial effects of this invention are: even with feedforward control combined with backlash error, high-precision trajectory tracking can still be achieved, improving the steering accuracy of the ball screw.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence system technology in the production field, and specifically relates to a ball screw steering intelligent control method and system based on humanoid robots. Background Technology

[0002] As a core carrier in the field of intelligent equipment, humanoid robots need to possess flexible steering capabilities to adapt to the operational needs of complex and unstructured scenarios. Ball screws, due to their advantages of high transmission efficiency, excellent positioning accuracy, high rigidity, and small backlash, have become the core transmission component of humanoid robot steering actuators, and are widely used in lower limb hip and ankle steering mechanisms, as well as torso deflection mechanisms.

[0003] Humanoid robots possess both human-like movement patterns and the ability to adapt to complex environments. Steering control is a key technology for achieving flexible movement, obstacle avoidance, and precise operation. Ball screw drives, with their advantages of high transmission efficiency, excellent positioning accuracy, high rigidity, and low backlash, are gradually replacing traditional gear and linkage transmission structures, becoming the mainstream transmission method for joint drive and steering execution in humanoid robots. They can effectively achieve efficient conversion between rotational and linear motion, meeting the high-precision steering transmission requirements of humanoid robots.

[0004] At present, there are still technical problems in the steering control technology of humanoid robots based on ball screws. Traditional steering control mostly adopts open-loop control or simple PID closed-loop control, which does not fully take into account the complex working conditions of humanoid robot's own posture, load changes and ground friction disturbances. The backlash error and flexible deformation in the ball screw transmission process cannot be compensated in real time, resulting in low steering accuracy, large overshoot, easy jitter at low speed steering and easy response lag at high speed steering. Summary of the Invention

[0005] This invention provides a ball screw steering intelligent control method and system based on a humanoid robot, which solves the technical problem of low ball screw steering accuracy in the prior art. By using a feedforward engagement clearance error, high-precision trajectory tracking can still be achieved even in the presence of mechanical dead zones, thereby improving the steering accuracy of the ball screw.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0007] The intelligent control method for ball screw steering based on humanoid robots includes the following steps:

[0008] Step S1: Construct a virtual-to-real digital twin of the humanoid robot's joint dynamics and ball screw transmission chain, and rehearse the turning trajectory in the future time window within the digital twin;

[0009] Step S2: Decompose the steering trajectory into feedforward control quantities, and combine them with the real-time collected steering backlash error to calculate the robust feedback compensation quantity using adaptive sliding mode control.

[0010] Step S3: When the ball screw is about to enter the steering dead zone, the double nut differential mechanism is driven to unload and reconstruct the micron-level preload according to the dynamic adjustment of the preload, so as to eliminate the hysteresis of the ball screw elastic deformation.

[0011] Step S4: The feedforward control quantity and the robust feedback compensation quantity are superimposed and applied to the servo drive unit to smoothly reverse the ball screw at the zero dead zone under load impact on the humanoid robot.

[0012] Optionally, in step S1, the dynamic equations are established using the Newton-Euler method, which combines inverse dynamics and forward dynamics, to calculate the joint dynamics of the humanoid robot; wherein, the dynamic equations of the Newton-Euler method are, the system inertia and internal forces = all external and internal driving forces are applied to the joint dynamics for calculation, so as to simulate the trajectory planning.

[0013] By using a rotary motor in conjunction with a ball screw, rotational motion is converted into linear thrust, thereby establishing a linear displacement mapping and a linear velocity mapping for the ball screw drive output from the joint.

[0014] Optionally, in step S1, when the steering trajectory is simulated for the future time window, the objective function in model predictive control is used to optimize the steering trajectory within the future time window, quantifying the quality of the simulated trajectory, and solving the control sequence through the optimization function.

[0015] Optionally, in step S2, the feedforward control calculates the control torque based on the desired trajectory to eliminate dynamic lag.

[0016] Since steering backlash causes a dead zone between input and output, the steering backlash error is regarded as a correction to the desired trajectory during the control process. It is a way of expressing the comprehensive tracking error and is used for high-precision motion control.

[0017] Optionally, in step S2, for the adaptive sliding mode control method, the following steps are adopted:

[0018] Step a: Approach phase; rapidly pull the system state from an arbitrary initial position toward the sliding surface;

[0019] Step b: Sliding phase; allow the system state to move stably along the sliding surface to achieve the desired dynamic performance.

[0020] Optionally, in step S3, the method for determining the critical moment of the steering dead zone is to trigger the preload adjustment by simultaneously satisfying two conditions. The speed condition is adopted, which means that when the difference between the theoretical speed and the actual load speed is less than the set threshold, it indicates that the lead screw is about to switch from forward or reverse rotation to reverse motion, entering the precursor stage of the steering dead zone.

[0021] The hysteresis condition for linear displacement is applied when the absolute value of the linear displacement hysteresis is greater than the critical deformation amount minus the advance margin, indicating that the elastic deformation is approaching its limit and is about to enter the turning dead zone.

[0022] Only when both of the above conditions are met simultaneously is it determined that the ball screw is about to enter the steering dead zone.

[0023] Optionally, in step S3, the double-nut differential mechanism is driven to perform the following steps:

[0024] Step 1: Micrometer-level preload unloading stage; immediately send a drive command to the double-nut differential mechanism to control the generation of a small relative differential displacement between the two nuts. Through differential action, the axial preload in the ball screw pair is released instantaneously at the micrometer scale, so that the elastic contact deformation between the ball and the screw raceway rebounds rapidly, thereby eliminating the elastic hysteresis error caused by the maintenance of preload, ensuring that the screw is in a free state with zero or very low preload at the moment of reversal, breaking the physical constraint of the steering dead zone;

[0025] Step II: Preload Reconstruction Stage; When the ball screw completes the reversal transition and the motion state tends to stabilize, the differential mechanism performs a reverse micro-displacement action to restore the relative position between the two nuts. The reconstruction process quickly rebuilds the rated axial preload, so that the balls and the screw raceway return to a close contact working state, thereby immediately restoring the rigid positioning accuracy of the system and seamlessly connecting to subsequent linear motion control.

[0026] Optionally, in step S3, the elastic deformation hysteresis elimination method is to use the quantitative calculation and elimination of the elastic deformation hysteresis error of the ball screw, and to establish a mapping relationship between the position deviation and the change in preload to adjust the preload for precise positioning.

[0027] Optionally, in step S3, the servo drive unit regulates the speed, torque and direction of the motor, thereby driving the ball screw transmission chain to complete the steering action. The servo drive unit smoothly outputs the commutation torque. At this time, in conjunction with the flexible adjustment of the dynamic preload, the feedforward control quantity monitors the tracking speed of the commutation trajectory, and the robust feedback compensation offsets the load impact and various error disturbances, eliminating motion sluggishness, impact noise and positioning deviation at the zero dead zone.

[0028] The intelligent control system for ball screw steering based on humanoid robots is used to execute the intelligent control method for ball screw steering based on humanoid robots described above.

[0029] The beneficial effects of this invention are:

[0030] 1. The future time window prediction of this invention predicts the robot's turning trajectory, such as turning angle, acceleration and target position, within a short period of time in the future. It also anticipates the load impacts that will be encountered in advance and uses a typical objective function from model predictive control to optimize the turning trajectory within the future time window, thus quantifying the quality of the predicted trajectory.

[0031] 2. This invention solves inversely for the acceleration, velocity, and friction characteristics in the desired trajectory, thereby providing the control torque in advance and achieving feedforward compensation for high-performance trajectory tracking. By calculating the required control quantity in advance, the inherent inertia, damping, and friction characteristics of the system are offset, enabling the output to quickly follow the command. Feedforward is mainly predictive compensation. Real-time measurement of steering backlash error is introduced. By embedding the backlash error in the feedforward + feedback architecture, high-precision trajectory tracking can still be achieved even in the presence of mechanical dead zones.

[0032] 3. This invention uses a speed condition to indicate that the ball screw is about to switch from forward or reverse rotation to reverse motion, which is the precursor stage of entering the steering dead zone. It uses a linear displacement lag condition to indicate that the elastic deformation is close to the limit and is about to enter the steering dead zone. Only when both the speed condition and the linear displacement lag condition are met simultaneously can it be determined that the ball screw is about to enter the steering dead zone. Attached Figure Description

[0033] 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.

[0034] Figure 1 This is a schematic diagram of the system structure of the present invention;

[0035] Figure 2 This is a schematic diagram of the workflow of the present invention;

[0036] Figure 3 This is a schematic diagram of the intelligent control of ball screw steering according to the present invention. Detailed Implementation

[0037] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0038] Example 1

[0039] like Figure 1 As shown, this embodiment provides an intelligent control system for ball screw steering based on a humanoid robot, including: a digital twin modeling module, an adaptive sliding mode control module, a preload dynamic adjustment module, and a servo drive execution module;

[0040] The digital twin modeling module is bidirectionally connected to the humanoid robot joint body and ball screw transmission chain. It is used to construct a virtual-real mapping digital twin of the humanoid robot joint dynamics and ball screw transmission chain. Based on the robot's steering requirements and operating conditions, the steering trajectory of the future time window is pre-simulated in the digital twin, and the generated steering trajectory is transmitted to the adaptive sliding mode control module.

[0041] The adaptive sliding mode control module is electrically connected to the digital twin modeling module and the data acquisition unit. It is used to receive the steering trajectory transmitted by the digital twin modeling module and decompose it into feedforward control quantity. At the same time, it receives the ball screw steering backlash error data collected in real time by the data acquisition unit. Based on the adaptive sliding mode control algorithm, it calculates and generates a robust feedback compensation quantity by combining the steering trajectory parameters and real-time error data, and transmits the feedforward control quantity and the robust feedback compensation quantity synchronously to the servo drive execution module.

[0042] The preload dynamic control module is electrically connected to the ball screw's double-nut differential mechanism and status detection unit. Through the real-time monitoring of the ball screw's operating status by the status detection unit, it accurately identifies the critical moment when the ball screw is about to enter the steering dead zone. Based on the preset control logic and critical moment conditions, it drives the double-nut differential mechanism to perform micron-level preload unloading and reconfiguration operations, dynamically adjusting the ball screw preload and eliminating the problem of elastic deformation hysteresis in the ball screw.

[0043] The servo drive execution module is electrically connected to the preload dynamic control module, the adaptive sliding mode control module, and the ball screw drive mechanism. It is used to receive and superimpose the feedforward control quantity and the robust feedback compensation quantity to generate the final drive command and act on the servo drive mechanism corresponding to the ball screw. In conjunction with the preload adjustment action of the preload dynamic control module, it realizes the smooth reversing control of the ball screw at the zero dead zone under load impact on the humanoid robot.

[0044] Example 2

[0045] Based on Example 1, such as Figure 2 As shown, this embodiment provides an intelligent control method for ball screw steering based on a humanoid robot, including the following steps:

[0046] Step S1: Construct a virtual-to-real digital twin of the humanoid robot's joint dynamics and ball screw transmission chain, and rehearse the turning trajectory in the future time window within the digital twin;

[0047] Among them, the virtual-real mapping digital twin is a digital model that not only includes the dynamic characteristics of the humanoid robot joints (such as inertia, friction and stiffness), but also accurately replicates all parameters of the ball screw drive chain (such as screw lead, nut clearance, material elastic modulus and transmission efficiency), ensuring the motion and force characteristics of the digital model.

[0048] The preview of the future time window is to predict the robot's turning trajectory (such as turning angle, acceleration and target position) within a short period of time (e.g. 50ms-200ms) in the future, and to anticipate the load impacts that will be encountered (such as joint load impacts and external collision impacts).

[0049] Step S2: Decompose the steering trajectory into feedforward control quantities, and combine them with the real-time collected steering backlash error to calculate the robust feedback compensation quantity using adaptive sliding mode control.

[0050] Among them, the feedforward control quantity decomposition is to break down the generated steering speed curve into feedforward control quantities of the target speed and torque of the servo motor according to time or angle. It is a command given in advance, rather than waiting for the error to occur before correction, which can reduce response lag.

[0051] Steering backlash error is collected in real time by physical sensors (such as encoders and force sensors) when the ball screw is turned. For example, the angle or displacement of the nut when the screw reverses is the core source of dead zone.

[0052] The adaptive sliding mode control calculates the feedback compensation as follows:

[0053] Sliding mode control is a highly robust control method that can ignore changes in system parameters (such as load fluctuations) and external disturbances, forcing the system to track the target trajectory. It will automatically adjust control parameters (such as sliding mode gain) based on the real-time collected steering backlash error, thus avoiding chattering problems in sliding mode control.

[0054] The robust feedback compensation output refers to the additional compensation signal or gain adjustment applied by the controller in the feedback loop to maintain system stability and achieve expected performance in the presence of model uncertainty, external interference, or sensor noise. It is a correction amount used to offset steering backlash error (such as the additional angle of motor rotation and supplementary torque).

[0055] Step S3: When the ball screw is about to enter the steering dead zone, the double nut differential mechanism is driven to unload and reconstruct the micron-level preload according to the dynamic adjustment of the preload, so as to eliminate the hysteresis of the ball screw elastic deformation.

[0056] Among them, the critical moment of the steering dead zone is the instant when the screw speed approaches zero and the torque direction is about to reverse;

[0057] The double-nut differential mechanism of the ball screw can be used to finely adjust the relative position of the two nuts through a motor or a micro-drive device of piezoelectric ceramic, thereby changing the magnitude of the preload.

[0058] Preload unloading involves slightly reducing the preload at the critical moment of steering to reduce the compressive stress between the nut and the lead screw and avoid the accumulation of elastic deformation.

[0059] Preload reconfiguration is the process of quickly restoring or adjusting the preload after unloading to eliminate the gap between the nut and the lead screw and prevent free rotation.

[0060] The micron-level adjustment achieves an adjustment precision at the micron level (μm), matching the gap of the ball screw (typically tens of microns), ensuring precise adjustment without introducing new mechanical errors.

[0061] Even with the most precise control algorithm, dead zones will still exist if the elastic deformation lag of the mechanical structure is not eliminated. This step, which actively eliminates the root cause of dead zones at the structural level, is a supplement to algorithmic control.

[0062] Step S4: The feedforward control quantity and the robust feedback compensation quantity are superimposed and applied to the servo drive unit to smoothly reverse the ball screw at the zero dead zone under load impact on the humanoid robot.

[0063] The feedforward control quantity and the robust feedback compensation quantity are weighted and superimposed to obtain the final total control quantity, such as: the target speed of the motor = feedforward speed + feedback compensation speed.

[0064] When applied to a servo driver, it converts the total control quantity into instructions that the servo driver can recognize, such as pulse signals or analog signals, to drive the motor and move the ball screw.

[0065] Due to load changes, the highly robust control and mechanical adjustment that eliminates structural hysteresis, the robot can smoothly reverse direction without dead zones or jamming when lifting heavy objects or encountering external collisions, even under sudden load changes.

[0066] Example 3

[0067] Based on Example 2, in step S1, for the humanoid robot, the joint dynamics of the digital twin are established using the Newton-Euler method, which combines inverse dynamics (i.e., finding torque from known motion) and forward dynamics (i.e., finding motion from known torque), to calculate the joint dynamics of the humanoid robot. Specifically, the Newton-Euler method's dynamic equations are based on the generalization of Newton's second law in generalized coordinates, applying system inertia and internal forces = all external and internal driving forces to the joint dynamics calculations for trajectory planning simulation.

[0068] ;

[0069] in, The term represents the system's inertia and internal forces, which is the sum of inertial force + Coriolis force or centrifugal force + gravity, and is the total internal force generated by the system's motion. This is the inertial force term; For generalized coordinates The relevant inertia matrix or mass matrix reflects the relationship between the system's mass distribution and configuration; The second derivative of the generalized coordinates is the angular acceleration or linear acceleration;

[0070] The terms are Coriolis force and centrifugal force. The Coriolis-centrifugal force matrix, with the current configuration and speed Related; The first derivative of the generalized coordinate system is the angular velocity or linear velocity.

[0071] For the gravitational term, by generalized coordinates The determination describes the torque or force of gravity about each generalized coordinate.

[0072] The term represents the external and joint driving forces, which is the sum of external forces that drive the system's motion.

[0073] This is an equivalent term for external environmental forces. The Jacobian matrix maps the Cartesian forces of the end effector to the joint space; The transpose of the Jacobian matrix completes the spatial mapping of forces; External force or torque (e.g., contact force, collision force) acting at the end (or a specified point);

[0074] This refers to the joint driving force or torque term; for The transpose of is used to map the driving torque vector in joint space to the generalized coordinate space; The selection matrix is ​​used to select either the active or driven joint; This provides the driving torque or force for each joint.

[0075] The resistance generated by the system's motion is all the internal forces or inertial forces that the system needs to overcome in order to maintain its current state of motion. It is the driving force that propels the system's movement; it is all the driving forces or external forces that the system receives from the outside.

[0076] The essence of the entire formula is the balance of forces. The sum of the inertial force, Coriolis or centrifugal force, and gravity generated by the system's motion equals the sum of the joint driving force and the external environmental force. All forces that prevent the system from moving equal all forces that make the system want to move.

[0077] Its engineering application is inverse dynamics: given a known trajectory. The joint torque needs to be calculated. Used to design robot control; positive dynamics: known joint torques and external forces The system's trajectory needs to be calculated. It is used for simulation.

[0078] For humanoid robots, the virtual-to-real mapping of the ball screw drive chain in a digital twin is achieved by using a rotary motor in conjunction with a ball screw to convert rotational motion into linear thrust, thereby establishing a complete mapping of joint outputs. This complete output mapping includes both the linear displacement mapping and the linear velocity mapping of the ball screw drive.

[0079] The linear displacement mapping of a ball screw drive is given by, assuming the motor rotation angle is θ. (Unit: rad), lead of the lead screw is (Unit: m / rev), transmission ratio is Then the linear displacement of the lead screw nut for:

[0080] ;

[0081] Among them, the linear displacement of the lead screw nut This indicates how many degrees the motor rotates. The lead screw and nut will then move a certain distance in a straight line. The transmission ratio is The lead screw rotates one revolution, and the rotation angle is... At this point, the distance the nut moves equals the lead of the lead screw. The displacement per unit radian is (Unit: m / rad), therefore, any angle The corresponding displacement is: .

[0082] The linear velocity mapping of a ball screw drive is: linear velocity With motor angular velocity Relationship:

[0083] ;

[0084] Among them, linear velocity This indicates the angular velocity of the motor per second. The lead screw and nut move at a speed Performing linear motion, the transmission ratio is .

[0085] The two formulas describe the geometric mapping relationship between rotary motion and linear motion in a ball screw drive mechanism. They are the basic kinematic models for constructing a digital twin of the motor-screw-joint virtual-real mapping.

[0086] For the future time window steering trajectory prediction, a typical objective function (cost function) from Model Predictive Control (MPC) is used to optimize the steering trajectory within the future time window. This quantifies the quality of the predicted trajectory, and the control sequence is solved through the optimization function.

[0087] ;

[0088] in, The total cost function is the target value to be optimized. The larger the value, the worse the simulated trajectory is, and the smaller the value, the better the simulated trajectory is, and the closer it is to the reference path.

[0089] The term to be summed is from the current moment. The step before predicting Horizon Accumulate at each time step;

[0090] To track error penalties, the predicted trajectory is made as close as possible to the reference trajectory; In the first At step, the system predicts the state vector (e.g., position, velocity, and heading angle); This refers to the desired reference state, such as the planned ideal trajectory point; The weight matrix (positive definite or positive semi-definite) determines the importance of different state variables; It is the square of the weighted Euclidean norm.

[0091] To control input penalties, prevent excessive control values, and avoid aggressive operations (such as: emergency rotation, rapid acceleration, and sudden stop). To control input vectors, such as steering angle and acceleration; This is the control weight matrix.

[0092] To control the rate of change penalty, it is used to suppress drastic fluctuations in the control signal, making the steering action smooth and continuous, and reducing mechanical wear; The change (increment) of the control quantity; This is the rate-of-change weight matrix.

[0093] As the terminal cost term, it ensures that the state at the end of the prediction window is close to the target, avoiding focusing only on the present and ignoring the future. This is important in stability analysis and helps to guarantee the convergence of the closed-loop system. To predict the horizon endpoint (the first) (Step) state; The desired state at the endpoint; The terminal weight matrix is ​​compared to the weight matrix. big.

[0094] In the future Within the time window of the step, the total cost function This indicates that the lead screw follows the reference path, avoiding aggressive operations such as sudden rotation, rapid acceleration, and sudden stop. The steering trajectory of the steering angle, acceleration, and target position does not deviate, and it stops steadily at the expected position.

[0095] Example 4

[0096] Based on Example 2, in step S2, the purpose of feedforward control is based on the desired trajectory. The required control torque under ideal conditions is calculated to eliminate dynamic lag. Specifically:

[0097] ;

[0098] in, This is the torque (unit: N·m) that the motor should output, used to drive the motor; This is for inertial term compensation, used to overcome the torque required to overcome the system's inertia; This is an estimate of the equivalent moment of inertia (unit: kg·m²). Acceleration at the desired steering angle (unit: rad / s) 2 ).

[0099] For damping compensation, the torque required to overcome viscous friction in the system (such as bearing lubrication resistance and hydraulic damping) is proportional to the speed. The damping is linear viscous damping. The actual system contains nonlinear components, but the feedforward term is linear. This is an estimate of the system's equivalent viscous damping coefficient (unit: N·m·s / rad). The velocity at the desired steering angle (unit: rad / s);

[0100] For friction torque compensation, it is an estimation function of the friction torque, which is related to the angle. nonlinear functions, The desired steering angle (unit: rad).

[0101] This formula utilizes the desired trajectory to inversely solve for the acceleration, velocity, and friction characteristics within that trajectory, thereby providing the ideal control torque in advance. This enables feedforward compensation for high-performance trajectory tracking. By calculating the required control quantity in advance, it counteracts the inherent inertia, damping, and friction characteristics of the system, allowing the output to quickly follow the command. Feedforward primarily provides predictive compensation.

[0102] Steering backlash causes a dead zone between the input and output. The acquired steering backlash error is set as... (Usually obtained through sensor measurements or observer estimation).

[0103] In the control process, backlash error is considered as a tracking error component, or used to correct the desired trajectory. The expression for composite tracking error is used in high-precision motion control systems, such as steer-by-wire and robot joint control. It integrates position error, velocity error, and backlash error into a unified error signal, which is the enhanced tracking error variable. As the input for subsequent sliding mode control or adaptive control, the corrected tracking error component is:

[0104] ;

[0105] in, The position tracking error is the angle deviation at the current moment, representing the basic error term, which drives the system to converge toward the target angle. The desired steering angle (command value); This is the actual steering angle (sensor measurement).

[0106] The speed tracking error weighting term is used to weight the speed deviation, suppress overshoot and oscillation in advance, improve the system damping characteristics, and speed up the dynamic response (similar to the derivative term in PD control). In sliding mode control, it helps to construct a more stable sliding surface. The velocity at the desired steering angle; This is the actual angular velocity. This is the speed error weighting coefficient (usually set by the designer based on the trade-off between response speed and stability). The value is greater than 0.

[0107] This is a backlash error compensation term for steering, which is applied in high-performance motion control systems, such as steer-by-wire, robot joints, and precision servo systems, and manifests as an active suppressor of nonlinear disturbances.

[0108] It is a saturation function; used to limit the amplitude of the gap compensation, avoid overcompensation causing oscillations, smooth switching behavior, and reduce chattering. It is equivalent to a soft limiter and only works when the gap is significant.

[0109] The saturation threshold or boundary layer thickness determines when gap compensation is initiated. When, the compensation amount increases linearly; when At that time, the compensation amount was clamped to ;

[0110] This is the gap compensation gain, used to adjust the effect of gap error on the tracking error variable. The intensity of the impact;

[0111] To monitor the real-time steering backlash error, due to gear meshing clearance and coupling looseness, the output shaft will not immediately follow when the motor reverses, resulting in idle travel. This represents the actual output lag caused by the gap at the current moment (unit: rad).

[0112] This formula defines an enhanced tracking error variable. This not only includes the traditional deviation between the expected position and the actual position, but also explicitly introduces the real-time measured steering backlash error. And it is smoothed using a saturation function. Error variable This will serve as the basis for the sliding surface design, thereby directly compensating for the nonlinear effects of gear backlash in the controller. By embedding a backlash error term in the feedforward + feedback architecture, the system can still achieve high-precision trajectory tracking even in the presence of mechanical dead zones.

[0113] For adaptive sliding mode control, the following approach is adopted:

[0114] Sliding mode control is a robust nonlinear control method that uses a sliding surface (also called a sliding mode surface) to allow the system state to rapidly approach and remain on this sliding surface under control. Once the system state is pulled onto the sliding surface and stabilizes there, the system will move according to the predetermined dynamic characteristics and will no longer be affected by external disturbances and changes in model error parameters.

[0115] The entire process consists of two steps:

[0116] Step a: Approach phase; rapidly pull the system state from any initial position toward the sliding surface.

[0117] Step b: Sliding phase; allow the system state to move stably along the sliding surface to achieve the desired dynamic performance.

[0118] Because the sliding mode control system is forced to adhere to the sliding surface, it is naturally insensitive to uncertainties and disturbances, and has strong robustness.

[0119] Among them, system state is a set of physical quantities that can completely describe the current motion of the system, such as: the position, speed and acceleration of motor control; or the coordinate position, velocity, attitude angle and attitude angular velocity of drones and robots.

[0120] In adaptive sliding mode control, to further mitigate the effects of unknown disturbances, model inaccuracies, and slow time-varying parameters, a robust feedback compensation is calculated. This robust feedback compensation is feedback-based, calculated in real-time based on the deviation of the system's current state from the sliding surface. It can compensate for all uncertainties not fully estimated or compensated by the adaptive law, including external disturbances, unmodeled dynamics, and parameter errors. During calculation, based on the magnitude of the system's deviation from the sliding surface and the upper bound of the disturbance, a sufficiently large but not overly conservative control variable is provided to pull the system back to the sliding surface and ensure it is not pushed out by disturbances again.

[0121] The robust feedback compensation amount is calculated and obtained in the following way:

[0122] Step 1: Real-time acquisition of system state and sliding surface deviation: Collect the actual state quantities of the current system, and calculate the real-time deviation value between the system state and the sliding surface based on the sliding surface.

[0123] Step 2: Determine the boundary of system disturbance and uncertainty: Based on system model error, external disturbance, and unmodeled dynamic uncertainties, determine the upper limit of the amplitude, or estimate the boundary of the uncertainty in real time through an adaptive mechanism.

[0124] Step 3: Determine the direction of compensation based on the direction of deviation: Based on the direction of the state deviating from the sliding surface, determine the direction of the robust feedback compensation, so that it always points to the sliding surface, ensuring that the system state is pulled back to the sliding surface.

[0125] Step 4: Calculate the compensation amplitude by combining boundary information: Based on the upper bound of the uncertainty term and the current deviation, calculate the compensation amplitude that can completely offset the remaining uncertainty to ensure that the compensation is strong enough to suppress the disturbance.

[0126] Step 5: Synthesize the final robust feedback compensation amount: Combine the compensation direction and compensation amplitude to form a real-time robust feedback compensation amount, and then superimpose them to obtain the total control amount.

[0127] Total control quantity = basic control quantity + robust feedback compensation quantity; where, the basic control quantity is a quantity without disturbance or error, which is the control quantity that allows the system to reach the target state and is the normal trajectory; the robust feedback compensation quantity is used to offset the quantity caused by model inaccuracies, external disturbances, unmodeled dynamics, and parameter changes.

[0128] Example 5

[0129] Based on Example 2, in step S3, the critical moment determination method for the steering dead zone is triggered by the simultaneous fulfillment of two conditions to adjust the preload, specifically:

[0130] and ;

[0131] in, As a speed condition, when the difference between the theoretical speed and the actual load speed is less than the set threshold, it indicates that the lead screw is about to switch from forward or reverse rotation to reverse motion, which is the precursor stage of entering the steering dead zone. This is the actual angular velocity of the motor; It is the reduction ratio, which is the transmission ratio from the motor to the lead screw; This represents the theoretical linear velocity of the lead screw; The actual linear velocity of the load (m / s) is measured by a position sensor or encoder. The speed threshold (unit: μm / s) is used to determine whether a turn is imminent;

[0132] The hysteresis condition for linear displacement is that when the absolute value of the linear displacement hysteresis is greater than "critical deformation amount - advance margin", it indicates that the elastic deformation is close to the limit and is about to enter the turning dead zone. It is the linear position hysteresis (command position - actual position), which reflects the elastic deformation and backlash accumulation of the ball screw; The critical elastic deformation is the maximum allowable elastic deformation of the lead screw under the rated preload. This allows for an early unloading margin (typically 2~5μm), which is used to trigger preload adjustment before actually entering the dead zone.

[0133] The ball screw is considered to be about to enter the steering dead zone only when both of the following conditions are met simultaneously:

[0134] The difference between the theoretical speed and the actual load speed is small enough (the commutation is imminent).

[0135] The positional hysteresis is large enough (elastic deformation is close to the critical value).

[0136] At this point, the micron-level preload of the dual-nut differential mechanism is unloaded and reconstructed, laying the foundation for eliminating elastic deformation hysteresis and avoiding positioning errors caused by steering dead zones.

[0137] In the process of unloading and reconfiguring the micron-level preload in the driven double-nut differential mechanism, the driven double-nut differential mechanism performs the following steps:

[0138] Step I: Micrometer-level preload unloading stage; immediately send a drive command to the double-nut differential mechanism to control a tiny relative differential displacement between the two nuts. Through differential action, the axial preload in the ball screw pair is released instantaneously at the micrometer scale, causing the elastic contact deformation between the balls and the screw raceway to rebound rapidly. This eliminates the elastic hysteresis error caused by maintaining preload, ensuring that the screw is in a free state with zero or extremely low preload at the moment of reversal, breaking the physical constraints of the steering dead zone.

[0139] Step II: Preload Reconstruction Stage; When the ball screw completes the reversal transition and the motion state tends to stabilize, the differential mechanism performs a reverse micro-displacement action to restore the relative position between the two nuts. The reconstruction process quickly rebuilds the rated axial preload, restoring the balls and screw raceways to a close contact working state, thereby immediately restoring the rigid positioning accuracy of the system and seamlessly connecting to subsequent linear motion control.

[0140] The method for eliminating elastic deformation hysteresis involves quantifying and eliminating the elastic deformation hysteresis error of the ball screw. By establishing a mapping relationship between position deviation and preload change, precise positioning can be achieved by adjusting the preload. Specifically:

[0141] ;

[0142] in, Elastic deformation hysteresis (or positioning hysteresis error) refers to the difference between the commanded position and the actual physical position at the instant of ball screw reversal (steering). It is caused by the elastic deformation of the screw, nut and balls under the rated preload, as well as the hysteresis characteristics of the contact angle, and is the root cause of inaccurate steering dead zone positioning.

[0143] It is the theoretical calculation position after eliminating hysteresis, consisting of two parts:

[0144] The initial hysteresis is the original position error before adjustment. It refers to the position deviation (i.e., the amount of deformation of the lead screw under tension or compression) that already exists under the current load before any preload adjustment is performed.

[0145] Preload compensation refers to the displacement change caused by the change in preload. The dynamic preload change refers to the difference between the actual preload of the system and the initial rated preload after the double-nut differential mechanism has performed its action. Represents the overall stiffness, which is the combined axial stiffness of the ball screw-nut-ball contact pair.

[0146] In this formula, the initial state is due to the presence of preload, causing the lead screw to be stretched or compressed. The preload was adjusted via a differential mechanism. For example: reduce the preload, i.e. .

[0147] By constructing an elastic deformation hysteresis model This model quantifies the linear relationship between positioning error and preload. Based on this model, when the critical moment of steering dead zone is detected, the required change in preload is calculated. The dual-nut differential mechanism is driven to generate a corresponding differential displacement, providing real-time compensation. To offset the initial hysteresis This allows for the dynamic elimination of positioning lag during steering.

[0148] Example 6

[0149] Based on Example 2, in step S4, the feedforward control quantity and the robust feedback compensation quantity are precisely superimposed and fused to form a comprehensive control command.

[0150] The feedforward control quantity is derived from the steering trajectory pre-simulated by the digital twin. It has foresight and predictive ability, and can match the expected steering movements of the humanoid robot joints in advance, providing basic drive commands for the ball screw transmission and realizing precise planning of steering movements.

[0151] The robust feedback compensation is calculated by combining the real-time collected steering backlash error with the adaptive sliding mode control algorithm. It has strong anti-interference and adaptive adjustment capabilities, and can specifically offset the control deviation caused by backlash, mechanical wear and external disturbances in actual operation, and make up for the lag and limitations of feedforward control.

[0152] The superposition of the two is not a simple numerical addition, but an adaptive fusion based on the characteristics of the transmission chain operation and the response characteristics of the servo drive. It retains the ability of feedforward control to accurately plan the steering trajectory, and relies on robust feedback compensation to achieve dynamic error correction. This allows the integrated control commands to not only fit the ideal running trajectory, but also adapt to the dynamic changes of actual working conditions, providing accurate and reliable command basis for subsequent servo drive execution and avoiding the control defects of a single control mode.

[0153] The integrated control commands, once superimposed, are directly applied to the servo drive unit that is paired with the ball screw, with the servo drive unit serving as the execution terminal.

[0154] like Figure 3As shown, the humanoid robot employs a smooth reversing mechanism at the zero dead zone of the ball screw under load impact. The servo drive unit precisely adjusts the motor's speed, torque, and direction according to comprehensive control commands, thereby driving the ball screw transmission chain to complete the reversing action. The servo drive unit smoothly outputs the reversing torque. At this time, combined with the flexible adjustment of dynamic preload, the feedforward control monitors the tracking speed of the reversing trajectory, and robust feedback compensation offsets the load impact and various error disturbances, eliminating motion sluggishness, impact noise, and positioning deviation at the zero dead zone. Even under continuous external load impact, the ball screw can achieve continuous, smooth, and jitter-free smooth reversing at the zero dead zone position, ensuring smooth joint reversing movements of the humanoid robot, maintaining the accuracy and stability of the overall operation, and adapting to the high-efficiency operation requirements in complex scenarios.

[0155] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope described in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for intelligent control of ball screw steering based on a humanoid robot, characterized in that, Includes the following steps: Step S1: Construct a virtual-to-real digital twin of the humanoid robot's joint dynamics and ball screw transmission chain, and rehearse the turning trajectory in the future time window within the digital twin; Step S2: Decompose the steering trajectory into feedforward control quantities, and combine them with the real-time collected steering backlash error to calculate the robust feedback compensation quantity using adaptive sliding mode control. Step S3: When the ball screw is about to enter the steering dead zone, the double nut differential mechanism is driven to unload and reconstruct the micron-level preload according to the dynamic adjustment of the preload, so as to eliminate the hysteresis of the ball screw elastic deformation. Step S4: The feedforward control quantity and the robust feedback compensation quantity are superimposed and applied to the servo drive unit to smoothly reverse the ball screw at the zero dead zone under load impact on the humanoid robot.

2. The intelligent control method for ball screw steering based on a humanoid robot according to claim 1, characterized in that, In step S1, the dynamic equations are established using the Newton-Euler method, which combines inverse dynamics and forward dynamics, to calculate the joint dynamics of the humanoid robot. The dynamic equations of the Newton-Euler method are: the system inertia and internal forces = all external and internal driving forces are applied to the joint dynamics for calculation, so as to simulate trajectory planning. By using a rotary motor in conjunction with a ball screw, rotational motion is converted into linear thrust, thereby establishing a linear displacement mapping and a linear velocity mapping for the ball screw drive output from the joint.

3. The intelligent control method for ball screw steering based on a humanoid robot according to claim 1, characterized in that, In step S1, for the future time window steering trajectory prediction, the objective function in model predictive control is used to optimize the steering trajectory within the future time window, quantifying the quality of the predicted trajectory, and solving the control sequence through the optimization function: ； in, Let be the total cost function, which is the objective value for optimization and represents the trajectory prediction result. The term to be summed is from the current moment. The step before predicting Horizon Accumulate at each time step; To track error penalties, the predicted trajectory is made as close as possible to the reference trajectory; To control input penalties, prevent excessive control values, and avoid aggressive operations; To control the rate of change penalty, it is used to suppress drastic fluctuations in the control signal, making the steering action smooth and continuous, and reducing mechanical wear; As the terminal cost term, it ensures that the state at the end of the prediction window is close to the target, avoiding focusing only on the present and ignoring the future. It is very important in stability analysis and guarantees the convergence of the closed-loop system.

4. The intelligent control method for ball screw steering based on a humanoid robot according to claim 1, characterized in that, In step S2, the feedforward control is based on the desired trajectory to calculate the control torque in order to eliminate dynamic lag. Specifically: ； in, This is the torque that the motor should output to drive it; This is for inertial term compensation, used to overcome the torque required to overcome the system's inertia; This is an estimate of the equivalent moment of inertia; The acceleration is the desired steering angle. For damping compensation, the torque required to overcome viscous friction in the system is proportional to the velocity. The damping is linear viscous damping. The actual system contains nonlinear components, but the feedforward term is linear. This is an estimate of the system's equivalent viscous damping coefficient; The velocity at the desired steering angle; For friction torque compensation, it is an estimation function of the friction torque, which is related to the angle. nonlinear functions, The desired steering angle; Because steering backlash causes a dead zone between the input and output, the steering backlash error is considered as a correction to the desired trajectory during control. It is a way of expressing the comprehensive tracking error and is used for high-precision motion control. The corrected tracking error components are: ； in, This serves as an enhanced tracking error variable, acting as an input for subsequent sliding mode control or adaptive control. The position tracking error is the angle deviation at the current moment, representing the basic error term, which drives the system to converge toward the target angle. The speed tracking error weighting term is used to weight the speed deviation, suppress overshoot and oscillation in advance, improve the system damping characteristics, speed up dynamic response, and in sliding mode control, it can construct a more stable sliding surface; This is a steering backlash error compensation term used in high-performance motion control systems.

5. The intelligent control method for ball screw steering based on a humanoid robot according to claim 1, characterized in that, In step S2, the adaptive sliding mode control method employs the following steps: Step a: Approach phase; rapidly pull the system state from an arbitrary initial position toward the sliding surface; Step b: Sliding phase; allow the system state to move stably along the sliding surface to achieve the desired dynamic performance.

6. The intelligent control method for ball screw steering based on a humanoid robot according to claim 1, characterized in that, In step S3, the determination of the critical moment of the steering dead zone is triggered by the simultaneous fulfillment of two conditions to adjust the preload: and ; in, As a speed condition, when the difference between the theoretical speed and the actual load speed is less than the set threshold, it indicates that the lead screw is about to switch from forward or reverse rotation to reverse motion, which is the precursor stage of entering the steering dead zone. For linear displacement hysteresis, when the absolute value of the linear displacement hysteresis is greater than the critical deformation amount minus the advance margin, the elastic deformation is close to its limit and is about to enter the turning dead zone; Only when both of the above conditions are met simultaneously is it determined that the ball screw is about to enter the steering dead zone.

7. The intelligent control method for ball screw steering based on a humanoid robot according to claim 1, characterized in that, In step S3, the double-nut differential mechanism is driven to perform the following steps: Step 1: Micrometer-level preload unloading stage; immediately send a drive command to the double-nut differential mechanism to control the generation of a small relative differential displacement between the two nuts. Through differential action, the axial preload in the ball screw pair is released instantaneously at the micrometer scale, so that the elastic contact deformation between the ball and the screw raceway rebounds rapidly, thereby eliminating the elastic hysteresis error caused by the maintenance of preload, ensuring that the screw is in a free state with zero or very low preload at the moment of reversal, breaking the physical constraint of the steering dead zone; Step II: Preload Reconstruction Stage; When the ball screw completes the reversal transition and the motion state tends to stabilize, the differential mechanism performs a reverse micro-displacement action to restore the relative position between the two nuts. The reconstruction process quickly rebuilds the rated axial preload, so that the balls and the screw raceway return to a close contact working state, thereby immediately restoring the rigid positioning accuracy of the system and seamlessly connecting to subsequent linear motion control.

8. The intelligent control method for ball screw steering based on a humanoid robot according to claim 1, characterized in that, In step S3, the method for eliminating elastic deformation hysteresis is to utilize the quantitative calculation and elimination of the elastic deformation hysteresis error of the ball screw, and to establish a mapping relationship between the position deviation and the change in preload to adjust the preload for precise positioning. Specifically: ； in, Elastic deformation hysteresis refers to the difference between the commanded position and the actual physical position at the instant of ball screw reversal or rotation. It is the theoretical calculation position after eliminating hysteresis, consisting of two parts: The initial hysteresis is used for the original position error before adjustment. It refers to the position deviation that already exists under the current load before any preload adjustment is performed. Preload compensation refers to the displacement change caused by the change in preload. The dynamic preload change refers to the difference between the actual preload of the system and the initial rated preload after the double-nut differential mechanism has performed its action. Represents the overall stiffness, which is the combined axial stiffness of the ball screw-nut-ball contact pair.

9. The intelligent control method for ball screw steering based on a humanoid robot according to claim 1, characterized in that, In step S3, the servo drive unit regulates the speed, torque and direction of the motor, thereby driving the ball screw transmission chain to complete the steering action. The servo drive unit smoothly outputs the commutation torque. At this time, with the flexible adjustment of the dynamic preload, the feedforward control quantity monitors the tracking speed of the commutation trajectory, and the robust feedback compensation offsets the load impact and various error disturbances, eliminating motion sluggishness, impact noise and positioning deviation at the zero dead zone.

10. A ball screw steering intelligent control system based on a humanoid robot, used to execute the ball screw steering intelligent control method based on a humanoid robot according to any one of claims 1-9, characterized in that, include: The digital twin modeling module is bidirectionally connected to the humanoid robot joint body and ball screw transmission chain. It is used to construct a virtual-real mapping digital twin of the humanoid robot joint dynamics and ball screw transmission chain. Based on the robot's steering requirements and operating conditions, the steering trajectory of the future time window is pre-simulated in the digital twin, and the generated steering trajectory is transmitted to the adaptive sliding mode control module. The adaptive sliding mode control module is electrically connected to the digital twin modeling module and the data acquisition unit. It is used to receive the steering trajectory transmitted by the digital twin modeling module and decompose it into feedforward control quantity. At the same time, it receives the ball screw steering backlash error data collected in real time by the data acquisition unit. Based on the adaptive sliding mode control algorithm, it calculates and generates a robust feedback compensation quantity by combining the steering trajectory parameters and real-time error data, and transmits the feedforward control quantity and the robust feedback compensation quantity synchronously to the servo drive execution module. The preload dynamic control module is electrically connected to the ball screw's double-nut differential mechanism and status detection unit. Through the real-time monitoring of the ball screw's operating status by the status detection unit, it accurately identifies the critical moment when the ball screw is about to enter the steering dead zone. Based on the preset control logic and critical moment conditions, it drives the double-nut differential mechanism to perform micron-level preload unloading and reconfiguration operations, dynamically adjusting the ball screw preload and eliminating the problem of elastic deformation hysteresis in the ball screw. The servo drive execution module is electrically connected to the preload dynamic control module, the adaptive sliding mode control module, and the ball screw drive mechanism. It is used to receive and superimpose the feedforward control quantity and the robust feedback compensation quantity to generate the final drive command and act on the servo drive mechanism corresponding to the ball screw. In conjunction with the preload adjustment action of the preload dynamic control module, it realizes the smooth reversing control of the ball screw at the zero dead zone under load impact on the humanoid robot.

Images