Search

CN-121973024-A - Six-axis robot self-adaptive polishing method based on double observers and curvature estimation

CN121973024ACN 121973024 ACN121973024 ACN 121973024ACN-121973024-A

Abstract

The invention discloses a six-axis robot self-adaptive polishing method based on double observers and curvature estimation, which comprises the steps of firstly, establishing a double observer framework, wherein an inner ring observer estimates and compensates nonlinear friction and model uncertainty in a joint in real time aiming at robot joint dynamics, and an outer ring observer outputs a contact force differential signal while filtering high-frequency noise aiming at a terminal six-dimensional force sensor signal. Secondly, a blind state curvature real-time estimation algorithm is provided, and the normal vector change rate and curvature characteristics of the contact surface are calculated in real time based on a differential geometry principle by utilizing a force differential vector output by an outer ring observer and the tangential speed of the tail end of the robot. And finally, designing a curvature self-adaptive variable impedance controller, constructing a nonlinear mapping model of rigidity and damping parameters changing along with the curvature, automatically reducing the rigidity to enhance the flexibility when the high curvature characteristic is detected, simultaneously increasing the damping ratio to inhibit the contact flutter, and generating a correction position instruction of the tail end of the robot through a discretization control law.

Inventors

  • HUANG XINJIANG

Assignees

  • 绿研智能装备(江苏)有限公司

Dates

Publication Date
20260505
Application Date
20251212

Claims (5)

  1. 1. The six-axis robot self-adaptive polishing method based on the double observers and curvature estimation comprises the following specific steps of: The method comprises the steps of 1, setting up a hardware platform, communicating with an upper computer through a TCP/IP interface in a six-axis industrial robot, and installing a six-dimensional torque sensor between a robot end flange and a grinding head, wherein an electric spindle is selected as a polishing execution end spindle and has a constant speed control function; Step 2, dynamics identification, namely establishing a physical measurement equation, and controlling the robot to perform multi-gesture slow motion under the condition that the tail end of the robot is not contacted with a workpiece; Step3, establishing a double-expansion-state observer, stripping the friction interference of the robot and the contact force of the external environment, respectively performing independent observation and treatment, further eliminating friction moment interference, and observing the contact force; step 4, estimating blind state curvature in real time, updating the normal vector of the contact surface in real time by using the estimated contact force value output by ESO-2, and calculating the curvature in real time; And 5, establishing a rigidity adjusting and damping adjusting formula by the curvature self-adaptive variable impedance controller, and controlling the contact force of the robot according to the rigidity adjusting value and the damping adjusting value.
  2. 2. The six-axis robot adaptive polishing method based on the dual observer and the curvature estimation according to claim 1, wherein the dynamics recognition in the step 2 can be expressed as: Step 2.1, establishing a load physical model; Let the six-dimensional force sensor coordinate system be The world coordinate system is According to the Newton Euler kinetic equation, the measurement equation can be expressed as: ; Wherein, the The original force measuring vector is the coordinate system of the sensor A measured triaxial resultant force reading; the original moment measurement vector is a triaxial resultant moment reading measured by the sensor under the coordinate system of the sensor; the external contact force is an acting force generated by the physical contact between the grinding head and the surface of the workpiece after eliminating the interference of gravity, inertial force and sensor zero drift; The external contact moment is the reaction moment generated in the contact process; the vector projection under the world coordinate system is transformed to the sensor coordinate system by the transposition of the rotation matrix; Is a gravity load vector in the world coordinate system, Wherein Is the load mass, g is the gravitational acceleration; The center of mass position vector is the position offset of the center of gravity of the grinding head tool relative to the measuring origin of the sensor; For the zero-point offset of the force, The zero offset of the moment is the output value of the sensor in an ideal state without load and gravity; Is the load mass, i.e., the total mass of the grater tool; The linear acceleration is the linear acceleration vector, and the linear acceleration of the origin of a coordinate system of a robot end sensor; is an inertial tensor matrix; As a vector of the angular velocity of the wheel, Is an angular acceleration vector; as a gyro moment item, when the robot simultaneously rotates at a high speed and changes the gesture, an additional moment is generated due to conservation of angular momentum; Step 2.2, designing an excitation track to solve a physical model; to solve the above parameters, the robot tip is made to not contact any workpiece so that And 0, Then making the robot make specific action, collecting data of specific action, making key gesture point of robot contain 20 labels, ensuring that the gravity vector has projection in the positive and negative directions of three axes of sensor coordinate system respectively, controlling robot to move to each discrete gesture point and make it stand still for 0.5 seconds, recording joint gesture of robot and sensor reading at that moment 、 Solving the load physical model, wherein the load model can be expressed as: ; the parameters to be identified in the system comprise load quality Centroid position vector Zero bias of force Zero offset of moment Finally, the physical model is rewritten into a linear regression form, and the weighted least square method is used for solving 、 、 、 ; Step 2.3, calculating external contact force after eliminating interference of gravity, inertial force and sensor zero drift; collecting joint gestures and sensor readings in real time at a sensor 、 Then, calculating the external contact force without the interference of gravity, inertia force and sensor zero drift through the load physical model and the identified system parameters And external contact moment 。
  3. 3. The six-axis robot adaptive polishing method based on the double observer and the curvature estimation according to claim 1, wherein the establishment of the double extended state observer in the step 3 can be represented as follows: step 3.1, deploying an inner ring observer ESO-1, namely observing the nonlinear friction moment of the joint; The inner ring observer is used for eliminating nonlinear friction moment and model uncertainty in the robot joint, and ensuring that a moment instruction can be accurately transmitted to the tail end; for a single joint kinetic equation of the robot, expressed as: ; Wherein, the The joint position is fed back by a motor encoder; the joint angular velocity is fed back by a motor encoder; the joint angular acceleration is fed back by a motor encoder; is the moment of inertia; is a complex nonlinear friction torque; Is an external load moment; The electromagnetic torque actually output by the motor; ESO-1 will contain friction And the total disturbance of the unmodeled dynamics is defined as a disturbance estimation state Constructing an ESO-1 observer formula: gain through observer , , Real-time joint nonlinear friction moment : ; Wherein, the Is an observation error; an observer estimate for joint position q; is the angular velocity of the joint Is a model of the observer estimate; Is a disturbance estimated value; The inertia matrix is calculated in real time through robot rigid body dynamics; 、 、 Respectively are 、 、 Is a derivative of (2); 、 、 gain measured by the joints respectively, and the input of the ESO-1 observer is , , Output as estimated total disturbance After the total disturbance is estimated, the electromagnetic moment actually output by the motor Compensating the output , To compensate for moment The moment is used for counteracting the disturbance moment, and can logically linearize the nonlinear robot system and eliminate the force control hysteresis caused by friction; step 3.2, deploying an outer ring observer ESO-2, namely filtering force signals and extracting differentiation; the contact force after the gravity compensation in the step 2 is processed by the ESO-2 of the outer ring observer And differential signal of contact force Signal the contact force Converting to a world coordinate system, and constructing a second-order tracking differentiator as a tracking target position, wherein the formula of the ESO-2 observer is as follows: ; Wherein, the For signals of contact force in the sensor coordinate system Converting the contact force to a world coordinate system; Is that Is a function of the estimated value of (2); Is that An estimate of the derivative; Error is observed for force; And Gain for force observer; 、 Respectively is 、 Is a derivative of (2); The input of the outer ring observer ESO-2 is the original force signal after the gravity compensation Output is the filtered contact force estimated value And differential contact force estimation 。
  4. 4. The six-axis robot self-adaptive polishing method based on the double observer and the curvature estimation according to claim 1, wherein the real-time blind state curvature estimation in the step 4 is represented as follows: Step 4.1, normal vector calculation of the contact point; In an ideal polishing contact, the robot tip applies the main force direction and contact force Although tangential friction exists, since joint friction is already compensated in ESO-1 of step 3 and polishing generally requires a constant normal force, the direction of the contact force can be approximated as the direction of the normal vector of the surface; obtaining unit normal vector by using the contact force vector output by ESO-2 in step 3 : ; Wherein, the The unit normal vector is the time t; A force differential estimated value output by ESO-2 at the time t; A force estimation value output by ESO-2 at the time t; is the L2 norm; Step 4.2, calculating curvature; To prevent division by zero and to enhance robustness, the curvature calculation formula is as follows: ; Wherein, the Is that Is a derivative of (2); as regularization parameters, preventing denominator from being 0; the tangential linear velocity of the robot: ; Wherein, the Is the cartesian velocity of the robot tip.
  5. 5. The six-axis robot adaptive polishing method based on the dual observer and the curvature estimation according to claim 1, wherein the curvature adaptive variable impedance controller in the step 5 is represented as follows: Step 5.1, establishing a rigidity adjusting function; When the curvature is When the curved surface belongs to a flat area, the robot needs to keep high rigidity so as to resist cutting force interference and ensure machining efficiency and surface flatness; When the curvature is When the robot increases, the curved surface enters a high curvature area, if the robot keeps high rigidity, over-cutting or force control overshoot is easy to occur, so that the rigidity needs to be reduced, and the shape of the curved surface is conformed; the stiffness adjustment function is a smooth nonlinear decay function: ; Wherein, the Maximum stiffness in planar processing; Minimum stiffness in planar processing; For adjusting the coefficient; is a rigidity adjusting value; step 5.2, establishing a damping adjustment function; When the curvature is In order to restrain vibration, a critical damping or over-damping strategy is adopted to increase the dissipation capacity of the system; firstly, calculating a target damping ratio: ; Wherein, the Is a command damping ratio; is the base damping ratio; is a damping growth coefficient; and then calculating a physical damping coefficient: ; Wherein, the Is a command damping coefficient; is the target inertia; Curvature calculated for step 4.2; And 5.3, finally issuing a control instruction: converting the equation into a corrected position instruction of the robot bottom layer The method comprises the following steps: ; Wherein, the The target position is issued to the servo motor; is an ideal polishing path; the deviation between the actual position of the tail end of the robot and the set value; The differential of the deviation of the actual value and the set value of the end position of the robot is obtained.

Description

Six-axis robot self-adaptive polishing method based on double observers and curvature estimation Technical Field The invention belongs to the technical field of industrial robots, and particularly relates to a six-axis robot self-adaptive polishing method based on a double observer and curvature estimation. Background Along with transformation and upgrading of modern manufacturing industry, industrial robots are widely applied to the field of polishing and grinding of complex curved surfaces of aviation blades, automobile panel molds, 3C electronic product shells and the like. Compared with the traditional manual polishing, the robot polishing has the advantages of high efficiency, good consistency, low labor intensity and the like. In existing robotic polishing techniques, constant impedance control or force/bit hybrid control is the most common control strategy. Such methods typically require an operator to empirically set fixed stiffness, damping, and inertia parameters in advance. However, in a practical industrial scenario, the prior art has the following significant limitations and technical bottlenecks in facing complex workpieces of unknown geometry or drastically changing curvature: 1. Lacking the ability to adapt to complex geometric features, existing robotic polishing systems typically rely on accurate offline CAD models for path planning, or expensive 3D vision systems for online scanning. The defect of dependence on CAD model is that the actual workpiece often has larger dimensional tolerance, which causes the discrepancy between the actual shape and the theoretical model, and the abrupt change, the over-cutting or the insufficient polishing of the contact force are easily caused. The defect of depending on a vision system is that a polishing site is generally full of dust, water mist and illumination interference, the accuracy of a vision sensor is seriously affected, and the vision system is complex in calibration and high in cost and is difficult to work in a narrow space or a shielding area. At present, a 'blind state' sensing technology capable of deducing the curvature change of the surface of a workpiece in real time only by means of force sense sensing is lacking. 2. The contradiction between constant control parameters and variable curvature workpieces, the requirements of the robot for environmental compliance are dynamically changing while processing variable curvature surfaces. In a flat area, high rigidity is required to ensure the removal efficiency and the track precision, in an area with high curvature, if the high rigidity is maintained, the workpiece is extremely easy to wear out or cause instability of a system due to excessive contact force, and if the low rigidity is adopted in the whole process for safety, the processing efficiency of the flat area is low and the precision is difficult to ensure. In the prior art, single fixed impedance parameters are adopted, so that the machining quality and the safety under different geometric characteristics cannot be considered, and an effective flutter suppression mechanism is lacked. 3. The interference of signal noise and joint friction on force control accuracy, high-accuracy force control depends on pure force signals and accurate actuating mechanisms. Signal differential noise-to the rate of change of force that must be obtained in order to perceive the surface curvature by the force signal. However, six-dimensional force sensors in industrial sites are ubiquitous in high frequency noise, and conventional differential algorithms can infinitely amplify the noise, resulting in the inability to use the calculated geometric features. Joint nonlinear friction, namely complex nonlinear friction and backlash exist inside the joint of the industrial robot. These internal disturbances, if not effectively compensated, can seriously affect the linearity and response speed of the tip force output, resulting in a robot with a "feel" that is slow, and difficult to implement fine force tracking control. In summary, how to eliminate the interference of sensor noise and joint friction and sense the curvature characteristics of the workpiece surface in real time under the condition that the geometric model is unknown and no visual assistance is available, and dynamically adjust the compliance of the robot according to the curvature characteristics, is a key technical problem to be solved in the field of complex curved surface polishing of the current robot. Disclosure of Invention In order to solve the problems, the invention provides a six-axis robot self-adaptive polishing method based on a double observer and curvature estimation, which comprises the following specific steps: The method comprises the steps of 1, setting up a hardware platform, communicating with an upper computer through a TCP/IP interface in a six-axis industrial robot, and installing a six-dimensional torque sensor between a robot end flange and a grinding head, wherein an