Friction model based industrial robot force-torque sensor free teaching method

Abstract

The application relates to a friction model-based industrial robot torque-free sensor teaching method, which refers to a joint friction model, judges the starting intention of a joint in a static state by using the internal variable of elastic friction, and increases the driving torque of the joint by temporarily increasing the estimated value of the friction force of the joint, so that the joint can be started instantaneously. The application has the beneficial effect that the control effect of the drag teaching in the motion and static states is adjusted by modifying the elastic friction model and adopting the variable parameter method, so that the industrial robot torque-free sensor teaching is realized.

Description

Technical Field

[0001] This invention relates to robotic arm drag teaching technology and the establishment of joint friction models, and particularly to a torque sensor-free teaching method for industrial robots based on friction models. Background Technology

[0002] Based on a generalized momentum-based external force observer and admittance control scheme, this paper uses a joint elastic friction model to estimate joint friction force, estimates and plans the friction parameters existing in the joint initiation phase, and realizes online teaching of the robot with zero force balance and torque control. Using an admittance control scheme, the joint motion trajectory is generated based on the observed external force, realizing drag teaching of the industrial robot. The elastic friction model is used to model joint friction, realizing friction force estimation in low-speed and stationary states. To solve the problem of difficult dragging of the joint in a stationary state, the friction force estimation in the joint initiation phase is planned. By briefly increasing the estimated value of joint friction, the joint driving torque is increased, thereby realizing regular dragging of the joint, and the initiation planning algorithm does not affect other motion phases of the robot joint. Currently, industrial robots are generally taught using teach pendants, planning complex motion trajectories through programming. To shorten the robot teaching time and achieve its localization and complex trajectory generation, manual guidance of robot movement can be used, i.e., drag-based online teaching. Drag-based online teaching makes robot use more convenient for operators and allows for easy avoidance of environmental obstacles. Driving teaching of industrial robots typically uses force sensors; however, these sensors are expensive, hindering their widespread adoption in industrial applications. Therefore, driving teaching without torque sensors is the preferred method. Besides the nonlinearity of joint friction, internal transmission mechanisms introduce more complex friction into joint motion. Friction exists in structures such as reducers, bearings, and input / output shafts within the joint, including both viscous and dry friction. This makes the friction of the joint more complex in a stationary state, often requiring greater external force to start the joint from a standstill during driving teaching. To address the difficulties in the joint's starting phase, the joint friction model is modified, employing a variable parameter method to adjust the control effect of driving teaching in both moving and stationary states. Summary of the Invention

[0003] To address the problems existing in the current technology, an algorithm for starting planning is proposed by establishing a joint friction model for industrial robots. This algorithm uses the intrinsic variables of the elastic model to determine the starting intention of the joint in a static state, and increases the driving torque of the joint by briefly increasing the estimated value of the joint friction force, so that the joint can start easily.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for sensorless drag teaching of a robot based on a joint friction model, comprising the following steps:

[0005] Step 1: Based on the robot's dynamic and kinematic models, use an external force estimation method to convert the joint forces applied to the robot by the operator into motion commands for the robot.

[0006] Step 2: Based on the external force estimation method used in Step 1, when the joint angular acceleration information value is missing, the parameters are estimated according to the externally applied external force. The admittance control scheme is used to generate motion commands for the external force, and the update equation of the joint target position in discrete state is derived, thereby realizing the teaching of the industrial robot without torque sensor.

[0007] Step 3: Using the external force estimation method based on generalized momentum, and leveraging the robot's dynamic model and motion state, we can determine whether there are signs of joint activation.

[0008] Step 4: Based on the motion model established in Step 3, solve the inertia matrix of the robotic arm through experiments. Measure the mass, inertia, length, cross-sectional area, and other parameters of each joint in the robotic arm. Then, substitute these parameters into the calculation formula of the inertia matrix to obtain the inertia matrix of the robotic arm. Calculate the required torque through inverse dynamics.

[0009] Step 5: Use a joint friction model to estimate the friction force during the start-up of the robotic arm, where the friction element satisfies:

[0010] τ fc =Δq*K c

[0011] Δq=qq c

[0012] Where τ fc For dry friction, K c Let q be the stiffness of the spring, and q be the position of the friction element. c Let τ be the relative position of the joint arms; the viscous friction force on the joint arms can be expressed as: τ fs =μ*q c The formula for the coefficient of viscous friction is: μ = F / N;

[0013] Step Six: Based on the joint position determined in Step Five, the final dry friction value τ is obtained. fc With viscous friction τ fs The sum of the two values ​​yields an estimate of the friction at startup of the robotic arm:

[0014] Step Seven: Set a measurement value a, where a is an extremely small positive value, and determine whether the articulated arm is stationary based on a:

[0015] When the joint is stationary, due to the influence of the motion controller noise, the joint angular velocity q is not zero. Therefore, it is defined that the joint is stationary when / q / < a; define G d and G u , when G d ∈[0,1], it is a geometric decreasing speed, and when G u ∈[1, +∞], it is a geometric increasing speed; define Δq th as a positive value. Δq th is used to judge whether there is an intention to start the joint together with / Δq c / . When it is detected that there is an intention to start, set Δτ f as the friction estimation compensation value during the start-up phase of the joint; Judgment of the start-up intention: Δτ f = k τ F c sign(Δq c (k)), where k τ is the friction coefficient, F C is equivalent to the Coulomb friction force of the joint in the rotating state, and sign(Δq c (k)) is the sign function of a joint position;

[0016] Step Eight: Judge the position q of the friction unit and the position q c of the articulated arm, and take the difference between the two to obtain Δq;

[0017] Step Nine: Update the state of Δq th . When / q / < a, it is stationary, and Δq th continually decreases to G d or vice versa increases to G u ;

[0018] Step Ten: Compare the values of / Δq c / and Δq th to judge whether each joint has an intention to start. When / Δq c / > Δq th , the joint has an intention to start; Judge whether the articulated arm is in a stationary state. If / q /

[0019] Step Eleven: Sum up the friction estimation value mentioned in Step Six and the friction estimation compensation value Δτ f judged in Step Seven to complete the friction estimation of the joint, and thus obtain the elastic friction model and substitute it into the robot to complete the teaching.

[0020] Based on the above technical solution, the present invention can be further improved as follows:

[0021] Furthermore, the external force estimation method based on generalized momentum used in step one is: Fαt=Mv.

[0022] Furthermore, the robot dynamics model in step three is as follows:

[0023] Furthermore, the method for calculating the required torque using inverse dynamics in step four is as follows: The required joint acceleration is calculated using the end-effector pose and desired acceleration, and then the required torque is calculated based on the dynamic model. The calculation formula includes inertial force terms, Coriolis force and centrifugal force terms, gravity terms, and friction terms. The friction term can be decomposed into viscous friction terms, Coulomb friction terms, and compensation terms, i.e.:

[0024] The beneficial effects of this invention are as follows: by modifying the joint friction model, the control effect of drag online teaching in motion and static states is adjusted by using a variable parameter method, thereby realizing sensorless drag teaching of industrial robots. Attached Figure Description

[0025] Figure 1 This diagram illustrates the estimation of joint forces applied to an industrial robot using a joint friction model.

[0026] Figure 2 This is a schematic diagram showing how the position of the joint is obtained by subtracting the position of the elastic friction element from that of the elastic friction model.

[0027] Figure 3 A mind map designed to determine whether a joint has the intention to activate;

[0028] Figure 4 This is a schematic diagram of an industrial robot structure.

[0029] Figure 5 The coordinates are in the user coordinate system (the coordinates of the TCP in the tool coordinate system 1 of the current robot in the user coordinate system 0).

[0030] Figure 6 This is a schematic diagram of the calibration process for an industrial robot.

[0031] Figure 7 A schematic diagram of a control scheme for online teaching of an industrial robot to combine external force estimation and admittance control;

[0032] Figure 8 A parameter diagram of the start-up planning algorithm used in the dragging experiment of the joints of an industrial robot;

[0033] Figure 9 The parameter diagram for dragging experiments on industrial robot joints without using the start-up planning algorithm. Detailed Implementation

[0034] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0035] A method for teaching a robot to be driven without torque sensors based on an elastic friction model includes the following steps:

[0036] Step 1: Based on the robot's dynamic model and motion state, estimate the joint forces applied by the operator to the robot using a generalized momentum-based external force estimation method, and convert the estimated external forces into robot ( Figure 4 The diagram shows the robot's structure, where a is the connecting flange, b is the wrist joint, c is the forearm, d is the servo motor, e is the reducer, f is the elbow joint, g is the upper arm, h is the waist joint, and i is the base. The motion command is based on the external force estimation method using generalized momentum: Fαt = Mv.

[0037] Step Two: Based on the external force estimation method used in Step One, estimation is performed even without joint angular acceleration information. Based on the estimated externally applied force, an admittance control scheme is used to generate motion commands from the external force, and the update equation for the joint target position in discrete states is derived, thereby achieving torque sensor-free drag teaching.

[0038] Step 3: Employing a generalized momentum-based external force estimation method, and utilizing the robot's dynamic model and motion state, the external force estimation lays the foundation for subsequent dragging teaching and is used to determine if there are signs of joint activation. The robot's dynamic model is as follows:

[0039]

[0040] Step 4: Based on the motion model established in Step 3, the inertia matrix of the robotic arm can be solved experimentally. Parameters such as the mass, inertia, length, and cross-sectional area of ​​each joint in the robotic arm are measured. These parameters are then substituted into the inertia matrix calculation formula to obtain the inertia matrix of the robotic arm. The required torque is calculated through inverse dynamics, that is, the required joint acceleration is calculated based on the end-effector pose and desired acceleration, and then the required torque is calculated according to the dynamic model. The calculation formula includes inertial force terms, Coriolis force and centrifugal force terms, gravity terms, and friction terms. The friction term can be decomposed into viscous friction terms, Coulomb friction terms, and compensation, i.e.:

[0041] Step 5: As Figure 1 and 2 As shown, Figure 1To facilitate the observation and identification of static friction, a joint friction model is used to estimate the external joint forces applied to the robot, so as to improve the accuracy of estimating joint friction. Figure 2 The elastic friction model is established according to the above idea by taking the difference between the positions of the joint friction unit and the joint friction model to obtain the position moved by the joint. The elastic friction model is used to estimate the friction force during the startup of the robotic arm, where the friction unit satisfies:

[0042] τ fc =Δq*K c

[0043] Δq = q - q c

[0044] where τ fc is dry friction. Dry friction is the frictional force between two objects when they are in contact, without considering the influence of lubricants. It is a type of static friction, and its magnitude depends on the roughness of the contact surface and the hardness of the materials. K c is the stiffness of the spring. (The joint friction model defines an elastic friction unit that has an elastic connection with the joint arm, and its stiffness is K c ), q is the position of the friction unit, q c is the relative position of the joint arm. The viscous friction force for the joint arm can be expressed as: τ fs =μ*q c , and the formula for the viscous friction coefficient is: μ = F / N.

[0045] Step Six: Based on the joint position determined in Step Five, the finally obtained dry friction value τ fc and the viscous friction force τ fs are summed up to obtain the friction estimation value during the startup of the robotic arm. Therefore, we have:

[0046] Step Seven: As Figure 3 shown, Figure 3 is a flowchart established to determine whether the joint has an intention to start, which combines the joint friction model and the startup planning algorithm. Set a measurement value a, where a is a very small positive value, and determine whether it is stationary. When the joint is stationary, due to the influence of the servo controller noise, the joint angular velocity q is not zero. Therefore, it is defined that the joint is stationary when / q / < a. Define G d and G u . When G d ∈[0,1], it is the geometric descent rate, and when G u v[1, +∞], it is the geometric ascent rate. Define Δq th as a positive value. Δq th is used to compare with / Δq c / Determine whether there is an intention to start the joint. When an intention to start is detected, Δτ is set in the start-up phase of the joint f is the friction estimation compensation value. Judgment of the start intention: Δτ f = k τ F c sign(Δq c (k)). Where k τ is the friction coefficient, and F C is equivalent to the Coulomb friction force of the joint in the rotating state. sign(Δq c (k)) is the sign function of a joint position.

[0047] Step Eight: Determine the position q of the friction unit and the position q c of the joint arm, and take the difference between the two to obtain Δq;

[0048] Step Nine: Update the state of Δq th , when / q / < a, it is stationary, and Δq th continually drops to G d or rises to G u conversely;

[0049] Step Ten: Compare / Δq c / with the value of Δq th to determine whether each joint has an intention to start. When / Δq c / > Δq th , the joint has an intention to start; determine whether the joint arm is in a stationary state. If / q /

[0050] Step Eleven: Sum the friction estimation value mentioned in Step Six and the friction estimation compensation value Δτ f judged in Step Seven to complete the friction estimation of the joint, thereby obtaining the elastic friction model and substituting it into the robot to complete the teaching. As Figure 7 shown, it is a control scheme for robot drag teaching by combining external force estimation and admittance control. The identification of joint friction adopts a joint friction model combined with starting planning. By adding the compensation value Δτ f in the joint starting phase, the external force estimation of the joint can be briefly improved and further improve the driving torque τ m of the joint, enabling the joint to start smoothly.

[0051] The following illustrates the beneficial effects achieved by the present invention through specific experiments:

[0052] A FANUC LR Mate 200iD / 4s six-axis robot with a 4kg payload, 717mm reach, and 20kg weight was used for the towing teaching experiment. An R-30 / B Mate controller was employed; all components in the control cabinet plus the teach pendant constituted the controller. Based on the robot's operational instructions and signals from sensor feedback, the controller directed the robot's actuators to complete the prescribed movements and functions. This involved both controlling the robot's own movement and coordinating its movement with surrounding equipment.

[0053] like Figure 6 As shown, using a teach pendant, the initial position of the robotic arm is set, and its world coordinates are zeroed to calibrate it. The teach pendant is then used to move the articulated arm to the predetermined position, recording the required time, speed, and position. A dragging teaching experiment is then conducted using the robot's last three axes (only axes 4, 5, and 6 are used; the other axes remain unchanged). By identifying the robot parameters and its motion state, estimated values ​​for various dynamic terms of the robot, such as G, can be calculated. The estimated external force can be obtained by using the external force estimation method based on generalized momentum. Joint initiation experiments of the robot were conducted using an elastic friction model and an initiation planning algorithm. The parameter settings for the last three joints of the robot are as follows: Figure 5 As shown.

[0054] Determine the value of 'a' during joint movement and analyze whether the joint arm intends to initiate movement. To improve movement stability, K... c The set values ​​are all lower than the measured friction stiffness, which makes it easy for the joint to slow down to a standstill when it is moving at low speed.

[0055] Data was recorded for the starting and dragging experiments of the last three joints according to the starting planning algorithm. Only the elastic friction model was used. Record the experimental data. The numerical settings for the last three joints of the robot are obtained from Table 1:

[0056] <![CDATA[K c / (N·m·rad -1 )]]> α <![CDATA[G d ]]> <![CDATA[G u ]]> Joint 4 350000 10 0.99 1.04 Joint 5 350000 10 0.99 1.06 Joint 6 250000 15 0.99 1.04

[0057]

[0058] like Figure 8 and 9 As shown, experiments were conducted on joints 4, 5, and 6 using the startup planning algorithm. Comparing the results of the same joints in two experiments, the planning algorithm produced a larger estimated torque at the moment of startup. An estimated torque of 31.09 N·m was generated during the start-up process, while the highest torque generated by using only the elastic friction model was 2.521 N·m.

[0059] The maximum torque generated during the startup phase using the planning algorithm is 26.05 N·m, while the maximum torque generated using only the elastic friction model is only 5.881 N·m.

[0060] During the robot's startup planning phase, the joint's Δq c It can quickly reach its maximum value in a short time. In experiments using the planning algorithm, Δq c The time required to rise from 0 to the maximum value is 47ms, while in the experiment using only the joint friction model, it takes 161ms, a reduction of 70.81% compared to the former.

[0061] In the experiment using the planning algorithm, the time required for joint start-up was 66ms, while in the experiment using only the joint friction model, it was 263ms, a reduction of 74.90%.

[0062] The friction estimation compensation generated during the start-up planning phase is very short-lived. The operator needs to continuously apply external force during the joint's start-up phase to bring it into a high-speed motion state; otherwise, the joint easily decelerates and returns to a stationary state. This characteristic gives the joint a certain degree of resistance to interference during the start-up phase.

[0063] Experiments show that starting planning can effectively increase the estimated external force value of the joints, with the estimated torque reaching over 26 N·m after planning. The planning algorithm can shorten the joint start-up time, reducing it by 70.81% for joint 6 and 74.90% for joint 4. Simultaneously, the robot's drag-and-teach scheme gives the joints a certain degree of anti-interference capability during the start-up phase.

[0064] In summary, this invention relates to the field of torque sensor-free drag teaching of robots, which is commonly used in various fields such as drag teaching of industrial robots and conventional manufacturing. Dragging teaching of industrial robots typically uses force sensors; however, these sensors are expensive, hindering their widespread adoption in industrial applications. Therefore, drag teaching without torque sensors is a preferred method. This invention, specifically for robot teaching, demonstrates better performance in manual robot teaching when applied to drag teaching.

[0065] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for teaching an industrial robot without torque sensors based on a friction model, characterized in that, Includes the following steps: Step 1: Based on the robot's dynamic model and motion state, estimate the joint forces applied by the operator to the industrial robot using the external force estimation method based on generalized momentum, and convert the estimated external forces into motion commands for the industrial robot; Step 2: Based on the external force estimation method used in Step 1, estimation is performed in the absence of joint angular acceleration information; based on the externally applied external force estimation, an admittance control scheme is used to generate motion commands for the external force, and the update equation for the joint target position in discrete state is derived, thereby realizing drag teaching without torque sensor; Step 3: Using the external force estimation method based on generalized momentum, the robot's dynamic model and motion state are used to estimate the external force, which lays the foundation for subsequent drag teaching and is used to determine whether the joints show signs of starting. Step 4: Based on the motion model established in Step 3, solve the inertia matrix of the robotic arm through experiments, and measure the mass, inertia, length, and cross-sectional area parameters of each joint in the robotic arm; then, substitute these parameters into the calculation formula of the inertia matrix to obtain the inertia matrix of the robotic arm, and calculate the required torque through inverse dynamics. Step 5: Use a joint friction model to estimate the friction force during the start-up of the robotic arm, where the friction element satisfies: fc =Δq*K c Δq=q-q c in fc For dry friction, K c Let q be the stiffness of the spring, and q be the position of the friction element. c Let be the relative position of the joint arms; the viscous friction force of the joint arms can be expressed as: fs =μ*q c , The formula for the viscous friction coefficient is: μ = F / N, where μ is the viscous friction coefficient, F is the maximum static friction force, and N is the normal force. Step Six: Determine the joint position based on Step Five, and finally obtain the dry friction value. fc With viscous friction fs The sum of the two values ​​yields an estimate of the friction at startup of the robotic arm: F = fc + fs； Step 7: Set a measurement value 'a', where 'a' is a very small positive value, and use 'a' to determine whether the articulated arm is stationary. When the joint is stationary, affected by the noise of the servo controller, the joint angular velocity q is not zero. Therefore, it is defined that the joint is stationary when |q| < a; define G d and G u such that when G d ∈ [0,1], it is a geometric decreasing speed, and when G u ∈ [1, +∞], it is a geometric increasing speed; define Δq th as a positive value, and Δq th is used to judge whether there is an intention to start the joint with |Δq c |; when an intention to start is detected, is set as the friction estimation compensation value during the start-up phase of the joint; Judgment of the start-up intention: , where is the friction coefficient, F C is equivalent to the Coulomb friction force of the joint in the rotating state, is the sign function of a joint position; Step 8: Determine the position q of the friction unit and the position q of the joint arm. c The difference between the two yields Δq; Step Nine: Update Δq th When the state of / q / < a, it is stationary, and Δq th continually drops to G d or conversely rises to G u ; Step 10: Compare / Δq c / and Δq th The value is used to determine whether each joint has the intention to start; when / Δq c / >Δq th The joint shows an intention to initiate; determine if the joint arm is in a stationary state, if / q / Step 11: Calculate the friction estimate mentioned in Step 6. F Compared with the friction estimate compensation value determined in step seven The frictional forces of the joints are estimated by summing the results, thus obtaining an elastic friction model which is then substituted into the robot to complete the teaching process.

2. The method for teaching an industrial robot without torque sensors based on a friction model according to claim 1, characterized in that, In step one, the external force estimation method based on generalized momentum is: Fαt = Mv, where Let M be the generalized momentum force, M be the robot joint mass matrix, and v be the joint angular velocity vector.

3. The method for teaching an industrial robot without torque sensors based on a friction model according to claim 1, characterized in that, The dynamic model of the industrial robot in step three is as follows: ,in Let the joint rotation vector be... For joint velocity vectors, The joint acceleration vector, For the inertial mass of the robot joints, This is the velocity term matrix related to centrifugal force and Coriolis force. For gravity, For friction, For the motor output torque, This is the vector of external torques.

4. The method for teaching an industrial robot without torque sensors based on a friction model according to any one of claims 1 to 3, characterized in that, The method for calculating the required torque using inverse dynamics in step four is as follows: The required joint acceleration is calculated using the end-effector pose and desired acceleration. Then, the required torque is calculated based on the dynamic model. The calculation formula includes inertial force terms, Coriolis force and centrifugal force terms, gravity terms, and friction terms. The friction term can be decomposed into viscous friction terms, Coulomb friction terms, and compensation terms, i.e.: ,in For joint friction, For joint velocity vectors, This refers to the viscous friction of the joint. Coulomb friction, Let be the sign function of the joint velocity. This is a joint compensation item.

Images