Robot screwing control method based on end excitation and feature fusion

Abstract

The application discloses a robot screwing control method based on end excitation and feature fusion. The method first applies high-frequency excitation through an end tool and performs circumferential scanning, combines contact force frequency response analysis and weighted least square fitting, and accurately perceives a screw hole normal vector and local stiffness. Secondly, in the process of guiding along the normal, based on real-time excitation-response phase difference identification, the damping and stiffness parameters of the impedance controller are adaptively adjusted to realize compliant guiding. Subsequently, in the screwing-in stage, an extended state observer is used to estimate disturbances caused by attitude deviation and the like, and a compensation motion is generated through an adaptive impedance controller. Finally, the stiffness, phase margin and friction coefficient features obtained in the previous steps are fused, the control parameters of the torque-angle double closed loop are initialized and dynamically adjusted through feedforward, and precise tightening is realized. The application realizes the integration of perception and control, and significantly improves the precision, adaptability and reliability of the robot tightening operation.

Description

Technical Field

[0001] This invention belongs to the field of robotics, specifically relating to a robot screw-tightening control method based on end-effector excitation and feature fusion. Background Technology

[0002] Current automated tightening operations using industrial robots generally rely on high-precision visual positioning or contact probes for initial hole alignment. These systems are complex, costly, and sensitive to ambient light and workpiece surface conditions. During the screwing process, traditional impedance control or torque-angle control methods use fixed parameters, making it difficult to adapt to varying workpiece stiffness, assembly clearances, and dynamic disturbances during thread engagement. This can easily lead to problems such as jamming, thread stripping, or inconsistent preload, severely impacting assembly quality and reliability. While existing literature or patents may mention some concepts from this invention, adjustments to key tightening parameters (such as impedance parameters, torque thresholds, and control gain) often rely on preset fixed parameters or simple feedback adjustments based on thresholds. These methods are essentially passive and reactive strategies. Summary of the Invention

[0003] This invention discloses a robot screw-tightening control method based on end-effector excitation and feature fusion, characterized by the following steps:

[0004] S1. Contact-based active scanning and stiffness-weighted normal vector perception: The robot's end-effector performs a circular scanning motion above the estimated screw hole, while simultaneously superimposing a high-frequency displacement excitation along the tool axis; contact force data at the scanning points are collected by force sensors, and the dynamic contact stiffness and damping of each point are calculated based on frequency domain analysis of the excitation and response; a weighted least squares method is used to fit the scanning point plane, where the weights are calculated based on the dynamic stiffness and damping of each point, resulting in a high-precision screw hole axis normal vector;

[0005] S2. Adaptive variable impedance guidance based on real-time phase identification: guide the tool along the normal vector direction obtained in step S1, and continuously superimpose multi-frequency excitation on the end during the guidance process; calculate the phase difference between the excitation displacement and the contact force response at the selected frequency in real time, and dynamically adjust the damping parameters of the impedance controller based on this phase difference, while adjusting the stiffness parameters according to the steady-state error of the contact force.

[0006] S3. Model Reference Helical Motion and Torque-Attitude Cooperative Disturbance Control: Establish a parameterized helical motion dynamic model to predict the torque and axial force curves under ideal alignment conditions; design an extended state observer to treat attitude deviations and thread anomalies during the helical process as disturbances and estimate them in real time; when the estimated disturbance exceeds the adaptive threshold, trigger and generate compensation force or torque commands, and execute compliance compensation through an adaptive impedance controller.

[0007] S4. Multi-feature feedforward fusion of torque-angle dual closed-loop precision tightening: The average dynamic stiffness and uniformity obtained in step S1, the system phase margin obtained in step S2, and the equivalent friction coefficient obtained in step S3 are fused to correct the monitoring threshold, target torque, and controller gain of torque-angle dual closed-loop control; torque-angle dual closed-loop control is executed, with angle control as the main control in the elastic stage and torque control as the main control in the plastic stage, and the control parameters are adaptively updated based on real-time performance.

[0008] Specifically, in step S1, the superimposed displacement excitation δ is: δ=A d sin(2πf d t)+A h sin(2πf h t), where A d For low-frequency amplitude, A h For high-frequency amplitude, f d For low frequency, f h It is a high frequency.

[0009] Specifically, in step S1, the weights w of the weighted least squares method i The calculation formula is: K dyn,i and C dyn,i C represents the dynamic stiffness and contact damping coefficient at point i, respectively. avg The average value of the contact damping coefficients at all scanning points, max(K) dyn ) is the maximum value of the dynamic stiffness of all scan points.

[0010] Specifically, in step S2, the multi-frequency excitation δ a for: , where f k For multiple different frequencies, A k For the corresponding amplitude, φ k ex This is the preset initial phase.

[0011] Specifically, in step S3, the extended state observer is designed as a third-order nonlinear observer, whose state variables include the ingress angle θ, angular velocity ω, and perturbation d(t), and its form is:

[0012] In this context, adding the symbol ^ at the top of the parameter indicates an estimated value of the parameter, and adding the symbol · at the top of the parameter indicates the derivative with respect to time. Let be the derivative of the estimated angle of rotation with respect to time. Let be the derivative of the estimated angular velocity with respect to time. This is the disturbance estimate. Let ε be the derivative of the disturbance estimate with respect to time, ε be the observer bandwidth, J be the system moment of inertia, and τ be the system moment of inertia. m τ is the motor torque. f For frictional torque, τ thread This refers to the thread engagement torque.

[0013] Specifically, in step S3, the condition for triggering compensation is: the standardized perturbation ζ(t) > ζ. th (t) and dζ / dt>ζ rat Where the standardized disturbance ζ(t) is the magnitude of the disturbance estimate and the reference torque τ ref The ratio, ζ th ζ is an adaptive threshold that increases with increasing axial force and screw-in angle. rat This is the threshold for the rate of change of the disturbance.

[0014] Specifically, in step S4, the correction formula for the angle monitoring threshold is:

[0015] , where Θ nom U is the standard angle threshold. k Φ represents the stiffness uniformity obtained in step S1. m-avg For the average phase margin, Φ target κ1 and κ2 are the target phase margins.

[0016] Specifically, in step S4, in the torque-angle dual closed-loop control, the initial gain of the torque loop PID controller is initialized based on the stiffness and damping parameters at the end of step S2.

[0017] Specifically, in step S4, when torque control is performed during the plastic stage, if the monitoring angle exceeds the maximum allowable plastic angle θ... p-max Then the torque command decreases linearly.

[0018] Beneficial technical effects: The solution of this invention achieves high-precision perception without external sensors. Through end-effector dynamic excitation and frequency domain analysis of contact force, it directly estimates the screw hole normal vector and local contact stiffness, eliminating dependence on vision / probes and improving system adaptability and reliability. It enhances the compliance and adaptability of the screwing process by dynamically adjusting impedance parameters based on real-time phase spectrum identification, enabling the robot to adapt to changing contact states, achieving smoother and more stable insertion, and reducing the risk of impact and jamming. It improves the system's anti-interference and correction capabilities by using an extended state observer (ESO) to estimate disturbances caused by posture deviations, thread anomalies, etc., online and generating compensation force commands in real time, significantly improving the tightening success rate and consistency under non-ideal conditions. It optimizes the final tightening quality and process consistency by integrating multi-dimensional features such as stiffness, phase margin, and friction coefficient perceived in previous steps, and feedforward correcting the control parameters of the torque-angle dual closed loop to achieve active compensation for individual differences in workpieces and changes in working conditions, thereby obtaining a more consistent preload force. Attached Figure Description

[0019] Appendix Figure 1 A schematic diagram of a robot tightening screws.

[0020] Appendix Figure 2 The flowchart is a robot screw-tightening control method based on end-effector excitation and feature fusion according to the present invention. Detailed Implementation

[0021] This invention discloses a control method for robot screw tightening. This method is used for automated screw tightening operations of industrial robots, such as in automotive assembly, electronic product assembly, and aerospace structural component assembly, where high-precision, high-reliability thread tightening tasks are required. Figure 1 As shown, during operation, the screw needs to be precisely screwed into the pre-machined threaded hole on the workpiece. The screw hole opening refers to the opening of the hole into which the screw will be screwed, i.e., the entrance to the pre-machined threaded hole on the workpiece. The robot end effector (screwdriver bit) is a key component installed at the end of the robot, specifically used to drive the screw to rotate and achieve the tightening action. It should be noted that the step of clamping / adsorbing the screw by the robot end effector is omitted before the control steps in this invention. This clamping / adsorbing step can be implemented with reference to other patents or existing technologies. This invention focuses on the control link from hole position sensing to precise tightening.

[0022] like Figure 2 As shown, the robot screw-tightening control method based on end-effector excitation and feature fusion of the present invention specifically includes:

[0023] I. Contact-based active scanning and stiffness-weighted normal vector precise sensing: Without using precise vision or probes, high-precision sensing is achieved through dynamic excitation and stiffness-weighted normal vector estimation to obtain the screw hole axis direction (normal vector) and the local stiffness distribution of the hole opening.

[0024] 1. Scan trajectory

[0025] The robot's end-effector (bit + screw) tip is controlled to contact the outer perimeter of the workpiece surface hole with a constant micro-contact force (usually 2-5N), that is, to perform a circular scanning motion with a radius of R in a horizontal plane above the estimated hole (perpendicular to the roughly estimated normal vector n0 direction), while superimposed with a high-frequency micro-amplitude sinusoidal excitation along the n0 direction.

[0026] The position p of the i-th point on the circular scan i p i =c+Rcosθ i ·u+Rsinθ i ·v, where c is the center of the circle (estimated position of the orifice), u and v are two orthogonal unit vectors in the circumferential plane, perpendicular to n0, and θ i Let be the angle parameter of the i-th point on the circumference, that is, the angle measured counterclockwise from the u-axis.

[0027] A dual-frequency displacement excitation δ=A is superimposed in the n0 direction. d sin(2πf d t)+A h sin(2πf h t), used to simultaneously identify contact stiffness and damping characteristics, where A d For low-frequency amplitude, it can be set to 0.05mm, A h For high-frequency amplitude, it can be set to 0.01mm, f d For low frequencies, such as 10Hz, f h It is a high frequency, such as 50Hz.

[0028] 2. Data Collection

[0029] At each sampling point i, the contact force F is measured by a force sensor. i Extract the normal force component (along the n0 direction) F n,i =F i ·n0.

[0030] 3. Calculation of local dynamic contact stiffness

[0031] Assuming the contact is a linear spring-damped system, the dynamic stiffness and damping coefficient of the system at the excitation frequency can be estimated using the frequency response function. Specifically, for the excitation displacement δ and the normal force response F... n,i For frequency domain analysis, the frequency response function H is defined as: In the formula, the (t) following the parameter has no practical meaning; it only indicates that the parameter is time-varying. F represents the Fourier transform, hence the frequency response function H... i (f) represents the normal force response F. n,i The ratio of the Fourier transform of the excitation displacement δ to that of the excitation displacement δ at frequency f. The dynamic contact stiffness K can then be calculated. dyn,i K dyn,i =|H i (f d )|, i.e., dynamic contact stiffness K dyn,i The frequency response function at low frequencies f d The amplitude at that point is calculated, and the contact damping coefficient is also calculated. Im(·) represents the imaginary part of (·), and the contact damping coefficient C dyn,i It is the damping of the local contact characteristics of the workpiece, which reflects the damping characteristics of the workpiece material or contact interface at point i.

[0032] 4. Stiffness-weighted normal vector estimation

[0033] Collect data from N points, each point having a position p. i (The actual position should theoretically include the excitation displacement, but the excitation displacement is very small, so the position can be approximated as p) i ), normal force F n,i and dynamic stiffness K dyn,i .

[0034] Weighted least squares is used to fit N points, and the normal vector of the fitted plane is the normal vector of the screw hole. The cost function J of the weighted least squares method is... w for: , Where n is the unit normal vector to be determined, d is the distance from the scan point to the plane, and w i It's about weighting, because point contacts with higher stiffness are more reliable, have smaller measurement errors, and therefore have a larger weight, C. avg It is the average of the contact damping coefficients at all scanning points, and its calculation method is the same as that for the average dynamic stiffness. K dyn,i and C dyn,i The dynamic stiffness and contact damping coefficient at point i are respectively, max(K dyn ) is the maximum value of the dynamic stiffness of all scan points, and |(·)| represents the absolute value of (·).

[0035] The screw hole normal vector n can be obtained with high precision using the weighted least squares method. Additionally, the average dynamic stiffness K can be calculated. avg =(∑K dyn,i ) / N, which is used in subsequent control steps. The default subscripts and superscripts of ∑ are i=1 to N, and N is the total number of sampling points collected during the circumferential scanning process. The summation range covers all scanning points.

[0036] II. Adaptive Variable Impedance Guidance Based on Real-Time Phase Spectrum Identification: The robot end effector is smoothly guided along the normal vector n, and the impedance control model parameters are adjusted by real-time phase identification. Adaptive impedance control based on real-time phase identification dynamically adjusts the damping and stiffness of the impedance control.

[0037] 1. Impedance control model

[0038] At the robot's end effector, an impedance controller is designed to make the end effector behave like a second-order mass-damped-spring system. Specifically, the impedance controller is as follows:

[0039] , where e=xx d It is the end position error. The dot sign at the top of the parameter indicates the derivative with respect to time. For speed error, For acceleration error, M d D d K d These are the desired inertia, damping, and stiffness matrices, F, respectively. ext To measure the contact force, F d The desired contact force.

[0040] 2. Dynamic Excitation and Frequency Domain Analysis

[0041] During the import process, multi-frequency excitation δ is continuously applied. a Superimposed on the nominal motion. The excitation signal can be a superposition of multiple sine waves, specifically: Where f k It is a set of frequencies covering the frequency band of interest, such as 5Hz, 15Hz, 30Hz, corresponding to amplitude A. k The diameters can be 0.03, 0.02, and 0.01 mm, respectively, with φ... k ex It is the initial phase of the pre-set excitation signal.

[0042] 3. Real-time phase identification

[0043] For each frequency f k Calculate the excitation displacement δ a Force response F ext Phase difference φ at this frequency k The force response F here ext The measured contact force F in the impedance control model ext For the same physical quantity, namely the instantaneous measurement value of the end force sensor, the optimal parameters can be backfitted from this actual measurement value, and then the phase difference can be calculated. The calculated phase difference is used to adjust the damping, while the original F ext The measured values ​​were then used to calculate the impedance control law. Phase difference φ kThe calculation is as follows: , where T is the integration window, usually taken as an integer multiple of the excitation period, τ is the integration variable, representing the time from tT to t, and t is the current time.

[0044] For a second-order system, phase lag is related to the damping ratio. Near the resonant frequency, the phase lag is highly sensitive to changes in the damping ratio. Therefore, the frequency f, which is most sensitive to phase changes, is chosen. c As a monitoring object. c The system's frequency response can be obtained through initial spectral analysis (e.g., applying a wideband excitation or sweep signal at the beginning of step two), and the frequency at which the phase is most sensitive to damping changes can be selected as f. c Alternatively, a frequency from the multi-frequency excitation in step two can be selected as the dominant frequency.

[0045] Let the phase lag corresponding to the current monitoring frequency be φ. We aim to control the phase lag within a safe range, such as a safe phase threshold φ. safe (e.g., -60 degrees Celsius).

[0046] The damping adjustment law can be designed as: D d =D d0 +ΔD,ΔD=K p (φ-φ safe )+K i ∫(φ-φ safe )dt, where D d0 This is the initial damping, ΔD is the damping adjustment, and K is the initial damping. p For proportional gain, K i For integral gain, ∫ does not specify upper and lower limits for the integral, meaning the upper and lower limits default to 0 to t, which is the integral from the initial control time (t=0) to the current time (t).

[0047] Alternatively, for a faster response, a nonlinear function can be used: ΔD = α D ·max(0,|φ|-|φ safe |), α D This is the gain coefficient.

[0048] Alternatively, for faster response, example control can be used: D d =D d0 +k φ (φ safe -φ), where k φ The proportional gain is used to linearly map the phase lag deviation to the damping adjustment, k φ >0, so when φ becomes more negative, D d Increase.

[0049] Meanwhile, stiffness parameter K dAdjustable based on steady-state force error: K d =K0+β(F d -F lp ), where K0 is the reference stiffness, F lp It is a low-pass filtered contact force, measured in real time by the contact force F. ext F is obtained by low-pass filtering. lp (t)=(1-α K )F lp (t-1)+α K F ext (t), where α K These are the filter coefficients, F d It is the desired contact force.

[0050] Alternatively, to avoid instability caused by excessive adjustments, and to balance fast response and stability, the stiffness parameter K... d Possible forms: Where K0 is the reference stiffness, γ is the adjusted strength coefficient, which can be 0.3, and F d For the desired contact force, F lp This is the contact force for low-pass filtering.

[0051] III. Model Reference Helical Motion and Torque-Attitude Coordinated Disturbance Rejection Control: During the helical motion, disturbances are estimated using an extended state observer, and compliance compensation is achieved using an adaptive impedance controller. The extended state observer estimates disturbances in real time, realizing torque-attitude coordinated disturbance rejection.

[0052] 1. Establish a parameterized spiraling dynamics model:

[0053] Axial force F z for: Where θ is the inclination angle, δ z (θ) represents the axial displacement, indicating the distance the screw advances in the axial direction as the rotation angle θ increases. For a standard thread, one revolution (360 degrees) advances one pitch p, so δ z (θ)=(p / 2π)·θ,K avg To obtain the average contact stiffness, step one involves scanning and measuring the dynamic stiffness at multiple points, then averaging the results to obtain C. axial F is the axial damping coefficient, which describes the damping characteristics of axial motion during screwing in. It can be determined experimentally or estimated based on material properties. Under ideal alignment, F... z It is the theoretical axial force generated by the thread, obtained through theoretical calculation, and it is related to F. ext When the normal components are in the same direction and approximately equal in magnitude, but attitude deviations exist, F ext It includes lateral components; the dot (·) at the top of the parameter indicates the derivative with respect to time. This indicates axial velocity.

[0054] The reference total torque is: τ ref (θ)=τ f (ω)+τ thread (θ), τ f (ω)=τ c ·sgn(ω)+b u ·ω,τ thread (θ)=F z (θ)·p / (2πη(θ)), η(θ)=tanλ / tan(λ+ρ). Among them, τ ref (θ) represents the reference total torque, i.e., the torque predicted by the model at an angle of θ, τ f (ω) is the frictional torque, which is related to the angular velocity ω. c b is the Coulomb friction torque, which is opposite to the direction of velocity. u τ is the coefficient of viscous friction. c and b u It can be calibrated experimentally, for example, by rotating the device at different speeds, measuring the torque, and then fitting the result; ω is the rotational angular velocity, sgn is the sign function, and τ is the torque. thread (θ) is the thread engagement torque, which is related to the screw-in angle θ, F z (θ) represents the axial force at the current angle, calculated from the axial force model; p is the pitch; η(θ) is the screw-in efficiency; λ is the thread helix angle; λ = arctan(p / (π·d)). m )), d m Where ρ is the thread pitch diameter, and ρ is the equivalent friction angle, ρ = arctan(μ / cos(α)). t / 2)), where μ is the friction coefficient of the threaded pair, which can be measured experimentally or estimated online (such as the equivalent friction coefficient μ estimated by the extended state observer in step three). eff ), α t The thread profile angle is given. The friction coefficient μ is the friction coefficient of the thread pair, while the μ estimated in step three is... eff It is the equivalent friction coefficient that integrates various factors (including thread friction, contact surface friction, etc.). The two can be correlated, but are not exactly the same. In the model, we can use the μ estimated in step three. eff This is used to update μ in the model, thereby enabling the model to adapt.

[0055] Based on the above parameterized model, it can be predicted that under ideal alignment, τ will increase with the insertion depth. ref (θ) and axial force F z The model can be initialized based on screw specifications, the average stiffness from step one, and the estimated coefficient of friction.

[0056] 2. Design of Extended State Observer

[0057] Considering the state-space model of the screw-in process, the additional frictional torque caused by attitude deviations, thread anomalies, etc., is regarded as a disturbance d(t). Let the system state be x1=θ (angle), x2=ω (angular velocity), x3=d(t) (disturbance). The system model is:

[0058] The dot at the top of the parameter indicates the derivative with respect to time, hence Let be the derivative of the inflection angle with respect to time. Let be the derivative of angular velocity with respect to time. Let τ be the derivative of the disturbance with respect to time, h(t) be the rate of change of the disturbance (unknown but assumed to be bounded), and τ be the time derivative of the disturbance. m J represents the motor torque, and J represents the system's moment of inertia. The moment of inertia can be identified by experimentally measuring the system's dynamic response.

[0059] Based on the aforementioned system model, the nonlinear ESO can be designed as follows:

[0060] Where ε is the observer bandwidth, and ε=0.01 is chosen to ensure fast convergence. The symbol ^ at the top of the parameter indicates the estimated value of the parameter. This is the disturbance estimate. Let be the derivative of the estimated angle of rotation with respect to time. Let be the derivative of the estimated angular velocity with respect to time. This is the derivative of the disturbance estimate with respect to time.

[0061] Alternatively, to improve computational speed, an ESO can be designed specifically for torque deviation, defining the torque deviation dynamics as: y = Δτ = τ m -τ ref If we consider the disturbance d(t) as a factor affecting the torque deviation, then the model is: y′=-ay+bd(t), where y is the torque deviation, i.e., the deviation between the actual torque and the torque predicted by the model, y′ is the rate of change of the torque deviation, and a and b are constants. Therefore, an ESO can be designed to estimate y and d(t): , where l1 and l2 are the observer gains.

[0062] 3. Triggering and Compensation

[0063] Compensation is triggered when the estimated amplitude of the disturbance exceeds a threshold. The disturbance needs to be standardized; the standardized disturbance ζ(t) is the sum of the amplitude of the estimated disturbance d and τ. ref The ratio of ζ(t) to ζ is triggered by the condition: ζ(t) > ζ. th (t) and dζ / dt>ζ rat , where ζ ratThe threshold value for the rate of change of disturbance can be determined by combining system simulation analysis with experimental calibration; the adaptive threshold value is ζ. th for:

[0064] Where ζ0 is the base threshold, representing the relative magnitude of the perturbation allowed at zero axial force and initial depth, which can usually be set empirically. α is the axial force sensitivity coefficient, controlling the rate at which the threshold increases with increasing axial force. It can be obtained experimentally by testing the maximum perturbation the system can tolerate under different axial forces, and then obtaining F through linear fitting. z (t) represents the real-time axial force, β is the depth sensitivity coefficient, indicating the weight of the influence of the insertion depth on the threshold, which can be determined through experimental calibration or simulation analysis, F z,max The maximum expected axial force can be calculated based on the screw's specifications and material. θ is the real-time screw-in angle, measured by the motor encoder. max The total screw-in angle is the total angle required from the start of screwing in to full screwing in (reaching the designed screw-in depth), which can be calculated based on the screw-in length and pitch.

[0065] Compensation command generation: Assuming the disturbance is caused by attitude deviation, the lateral compensation force F comp The estimated value of the disturbance d can be multiplied by a conversion factor, which converts torque into force. F comp The direction is perpendicular to the axial direction n, but the specific direction needs to be determined based on the disturbance characteristics. Specifically, if the torque fluctuation is periodic during the screwing process, it can be determined as unilateral friction, and the compensation direction is perpendicular to the axial direction and points in the opposite direction of friction.

[0066] Alternatively, a mapping model from perturbation to compensating force can be established:

[0067] , where F comp,x ,F comp,y For the force components (x and y directions) that need to be compensated in a plane perpendicular to the axis, M comp,x M comp,y M represents the torque components that need to be compensated (torques about the x and y axes), and M is a mapping matrix that maps the estimated value of the disturbance d to four compensation quantities, the elements of which can be obtained through theoretical analysis or experimental calibration.

[0068] 4. Compensation is performed using an adaptive impedance controller.

[0069] The compensation force command F comp As the desired contact force input for the adaptive impedance controller in step two, the desired contact force is updated to F. d new =F d +F compThus, the impedance controller generates compliant compensating motion based on the current adaptive parameters: .

[0070] The steady-state disturbance and axial force can be estimated using ESO, and the equivalent friction coefficient μ can be calculated. eff =2πd ss ·cosα t / (p·F z ), where d ss The steady-state value of the estimated perturbation d obtained through ESO can be taken as the average value of the perturbation estimates during the steady-state step of the screw-in process, where p is the pitch and α is the mean. t This refers to the thread profile angle.

[0071] IV. Multi-Feature Feedforward Fusion of Torque-Angle Dual-Closed-Loop Precision Tightening: Step four integrates the sensing information from the first three steps through a multi-feature feedforward fusion mechanism to optimize the control parameters for the final tightening. The dual-closed-loop control using multi-feature feedforward fusion incorporates features such as stiffness, phase, and friction, improving tightening quality. The average dynamic stiffness K obtained in step one... avg It is used to correct the angle monitoring threshold, adapt to differences in workpiece stiffness, and can also calculate the standard deviation of the dynamic stiffness in step one, which is the stiffness uniformity U. k， Used to evaluate stiffness uniformity and assist in quality judgment, Φ obtained in step two m The μ obtained in step three is used to evaluate contact stability and optimize the control switching point. eff Used to compensate for frictional dispersion and improve the consistency of preload.

[0072] 1. Dynamic feedforward initialization of control parameters

[0073] Angle monitoring threshold correction: based on stiffness uniformity U k Adjust the angle window. Let the standard angle threshold be Θ. nom The corrected threshold Θ th =Θ nom ·(1+γ d U k ), where γ d is a coefficient.

[0074] Alternatively, considering stiffness uniformity and contact stability, the corrected threshold can be replaced with , where Φ m-avg The average phase margin, i.e., the average value of the system phase margin during the adaptive impedance control process in step two, reflects the stability of the introduction process. Φ target The target phase margin is defined by κ1 and κ2, which are weighting coefficients that can be set experimentally.

[0075] Torque target value correction: based on the equivalent friction coefficient μ estimated in step three. effCorrect the target torque. Let the nominal target torque be τ. nom This value represents the torque required to achieve the design preload under standard friction coefficients, and is corrected to τ. target =τ nom ·(1+ξ(μ eff -μ nom )), where μ nom ξ is the nominal friction coefficient, and ξ is the friction compensation coefficient.

[0076] Controller gain initialization: Use the damping and stiffness parameters D from the end of step two. d and K d To initialize the gain of the torque loop PID controller. For example, the proportional gain K of the torque loop. p It can be set to be the same as K. d Relatedly, or further, the torque loop PID parameters are initialized as follows:

[0077] Where T2 is the end time of step two, and K d (T2), D d (T2) represents the stiffness and damping parameters of the adaptive impedance controller at the end of step two, respectively. K0 and D0 represent the reference stiffness and initial damping in the adaptive stiffness law of step two, respectively. p0 ,K i0 ,K d0 These are the nominal PID parameters.

[0078] 2. Dual closed-loop control execution

[0079] A torque-angle dual closed-loop control strategy is adopted. During the elastic phase, angle control is primary. Once the contact point is reached, the system enters the plastic zone (plastic phase) and switches to torque control until the target torque is achieved. Throughout the control process, the actual curve is compared in real time with a reference curve based on feature correction for monitoring and alarm functions.

[0080] Specifically, in the elastic zone (angle control), a constant angular velocity screw-in and torque monitoring method is typically used, and the angle command satisfies:

[0081] K θ For the angle loop gain, τ target The corrected target torque is τ, where τ is the current actual torque, and ω is the actual torque. min This is the minimum angular velocity required to ensure the tightening process continues and prevent jamming.

[0082] As the inner loop, the torque loop's output (current or torque command) is:

[0083] Where ∫ does not specify the upper and lower limits of the integral, it means that the upper and lower limits are from 0 to t by default, that is, the integral from the initial control time (t=0) to the current time (t), and the torque error e τ =τ cmd -τ.

[0084] Under angle control, the torque loop's function is to quickly track the torque command τ generated by the angle loop. cmd In the elastic zone, the controller uses angle as the primary control variable, but calculates the torque value in real time. When the torque reaches the engagement torque τ... snug At that time, record the angle θ of the contact point. snug Then it enters the plastic zone. The contact torque τ snug It can be calculated using the proportional method and can be set as the target torque τ. target 0.25-0.35 times. The engagement torque marks the beginning of bolt stretching (i.e., the engagement point), which occurs at the end of the elastic zone and the beginning of the plastic zone. The corrected torque target value is the final torque value set to achieve the design preload, i.e., the torque expected at the end of the plastic zone.

[0085] In the plastic zone (torque-controlled), torque is the primary control factor. While maintaining the target torque, the angle is monitored. If the angle does not exceed the limit, the target torque is maintained; if the angle exceeds the maximum allowable plastic angle θ... p-max At this time, the torque is reduced linearly to prevent thread damage. The specific torque command is as follows:

[0086] τ cmd =τ target +Δτ(t),

[0087] , where q is the angle compensation coefficient.

[0088] 3. Adaptive parameter update

[0089] Update PID parameters based on real-time performance:

[0090] Where ∫ does not specify the upper and lower limits of the integral, it means that the upper and lower limits are from 0 to t by default, that is, the integral from the initial control time (t=0) to the current time (t), e τ,max For the maximum permissible torque error, η p η i η d These are the proportional learning rate, integral learning rate, and differential learning rate, which can be set experimentally. i is the integration time constant, i.e., the normalized time scale of the integral term, and sat is the saturation function.

Claims

1. A robot screw-tightening control method based on end-effector excitation and feature fusion, characterized in that, Includes the following steps: S1. Contact-based active scanning and stiffness-weighted normal vector perception: The robot's end-effector performs a circular scanning motion above the estimated screw hole, while simultaneously superimposing a high-frequency displacement excitation along the tool axis; contact force data at the scanning points are collected by force sensors, and the dynamic contact stiffness and damping of each point are calculated based on frequency domain analysis of the excitation and response; a weighted least squares method is used to fit the scanning point plane, where the weights are calculated based on the dynamic stiffness and damping of each point, resulting in a high-precision screw hole axis normal vector; S2. Adaptive variable impedance guidance based on real-time phase identification: a guidance tool is used along the normal vector direction obtained in step S1, and multi-frequency excitation is continuously superimposed on the end during the guidance process; The phase difference between the excitation displacement and the contact force response at a selected frequency is calculated in real time, and the damping parameters of the impedance controller are dynamically adjusted based on this phase difference. At the same time, the stiffness parameters are adjusted according to the steady-state error of the contact force. S3. Model Reference Helical Motion and Torque-Attitude Cooperative Disturbance Rejection Control: Establish a parameterized helical dynamics model to predict the torque and axial force curves under ideal alignment conditions; An extended state observer is designed to treat attitude deviations and thread anomalies during the screwing process as disturbances and estimate them in real time. When the estimated disturbance exceeds the adaptive threshold, a compensation force or torque command is triggered and generated, and compliance compensation is performed through an adaptive impedance controller. S4. Multi-feature feedforward fusion of torque-angle dual closed-loop precision tightening: The average dynamic stiffness and uniformity obtained in step S1, the system phase margin obtained in step S2, and the equivalent friction coefficient obtained in step S3 are fused to correct the monitoring threshold, target torque, and controller gain of torque-angle dual closed-loop control; torque-angle dual closed-loop control is executed, with angle control as the main control in the elastic stage and torque control as the main control in the plastic stage, and the control parameters are adaptively updated based on real-time performance.

2. The method according to claim 1, characterized in that, The displacement excitation δ superimposed in step S1 is: δ = A d sin(2πf d t) + A h sin(2πf h t), where A d is the low frequency amplitude, A h is the high frequency amplitude, f d is the low frequency, and f h is the high frequency.

3. The method according to claim 1, characterized in that, In step S1, the weights w of the weighted least squares method i The calculation formula is: K dyn,i and C dyn,i C represents the dynamic stiffness and contact damping coefficient at point i, respectively. avg The average value of the contact damping coefficients at all scanning points, max(K) dyn ) is the maximum value of the dynamic stiffness of all scan points.

4. The method according to claim 1, characterized in that, In step S2, the multi-frequency excitation δ a for: , where f k For multiple different frequencies, A k For the corresponding amplitude, φ k ex This is the preset initial phase.

5. The method according to claim 1, characterized in that, In step S3, the extended state observer is designed as a third-order nonlinear observer, whose state variables include the ingress angle θ, angular velocity ω, and perturbation d(t), and its form is: In this context, adding the symbol ^ at the top of the parameter indicates an estimated value of the parameter, and adding the symbol · at the top of the parameter indicates the derivative with respect to time. Let be the derivative of the estimated angle of rotation with respect to time. Let be the derivative of the estimated angular velocity with respect to time. This is the disturbance estimate. Let ε be the derivative of the disturbance estimate with respect to time, ε be the observer bandwidth, J be the system moment of inertia, and τ be the system moment of inertia. m τ is the motor torque. f For frictional torque, τ thread This refers to the thread engagement torque.

6. The method according to claim 1, characterized in that, In step S3, the condition for triggering compensation is: the normalized perturbation ζ(t) > ζ. th (t) and dζ / dt>ζ rat Where the standardized disturbance ζ(t) is the magnitude of the disturbance estimate and the reference torque τ ref The ratio, ζ th ζ is an adaptive threshold that increases with increasing axial force and screw-in angle. rat This is the threshold for the rate of change of the disturbance.

7. The method according to claim 1, characterized in that, In step S4, the correction formula for the angle monitoring threshold is: , where Θ nom U is the standard angle threshold. k Φ represents the stiffness uniformity obtained in step S1. m-avg For the average phase margin, Φ target κ1 and κ2 are the target phase margins.

8. The method according to claim 1, characterized in that, In step S4, in the torque-angle dual closed-loop control, the initial gain of the torque loop PID controller is initialized based on the stiffness and damping parameters at the end of step S2.

9. The method according to claim 1, characterized in that, In step S4, during torque control in the plastic stage, if the monitoring angle exceeds the maximum allowable plastic angle θ... p-max Then the torque command decreases linearly.

Images