A Flexible Control Method for Dual-Robot Assembly of Fuze and Detonating Tube Based on Neural Networks

Abstract

This invention discloses a flexible control method for dual-robot assembly of a fuse and detonator based on a neural network, comprising the following steps: First, the estimated desired contact force is obtained experimentally on a force sensor, and the estimated position of the robotic arm's end effector is derived based on the contact position after trajectory planning; the kinematic model of the dual robots is obtained, and an adaptive impedance controller is designed based on the desired contact force, combining the impedance control relationship with a PID control strategy; the actual contact force between the two workpieces generated during each dual-robot collaborative assembly is collected, and the difference between the average value of the actual contact force and the desired contact force is calculated, and the difference is imported into the adaptive impedance controller; a neural network is used to fit the relationship between the actual posture, the actual driving quantity, and the difference between the theoretical driving quantity, and the trained neural network corrects the trajectory online. This invention can overcome the interference of external environment and uncertain force fields, making impedance control more accurate.

Description

Technical Field

[0001] This invention belongs to the field of force control and robot collaborative control technology, specifically relating to a flexible control method for dual-robot assembly of a fuse and a detonating tube based on neural networks. Background Technology

[0002] Traditional methods of assembling two parts using dual-robot collaborative assembly suffer from unsatisfactory assembly results and numerous safety hazards due to the inability to accurately meet constraint forces and the interference of external environmental factors and contact forces. Therefore, improved compliant control for the assembly of two parts has become a key research focus in the field of dual-robot collaborative assembly.

[0003] To address these issues, common methods include force / position hybrid control and impedance control combined with a six-dimensional force sensor for force detection and control. Impedance control offers better control performance and is easier to implement than force / position hybrid control, making it more prevalent in collaborative robot engineering.

[0004] However, traditional impedance control is difficult to accurately obtain the reference trajectory of the robot's end effector in practical applications due to environmental interference factors, and it cannot solve time-varying interference. Today's dual-robot processing environment is more complex and special. How robots can adapt to environmental disturbances and overcome the interference of disturbances to flexibly assemble objects has become a major challenge in current industrial production. Summary of the Invention

[0005] To address the aforementioned problems in existing technologies, this invention provides a flexible control method for the assembly of a fuse and detonating tube using a dual-robot system based on a neural network. This method can overcome interference from the external environment and uncertain force fields, making impedance control more precise, improving the working efficiency of the dual robots, and ultimately achieving flexible assembly of the fuse and detonating tube.

[0006] The objective of this invention is achieved through the following technical solution:

[0007] A flexible control method for assembling a fuse and detonator using a dual-robot assembly based on a neural network includes the following steps:

[0008] S1: After the dual robots clamp the fuse and detonator to the expected position through the planned path, the estimated expected contact force is obtained on the force sensor through experiment, and the position estimate of the end position of the robotic arm is obtained according to the contact position after the trajectory planning of the dual robots.

[0009] S2: Obtain the kinematic model of the two robots, design an adaptive impedance controller based on the desired contact force, combine the impedance control relationship with the PID control strategy, so that the actual force between the robots tracks the desired contact force, and convert the actual contact force into the control of the actual position.

[0010] S3: Collect the actual contact force between the two workpieces generated during each dual-robot collaborative assembly, calculate the difference between the average value of the actual contact force and the expected contact force, and import the difference into the adaptive impedance controller to reduce the error of the expected trajectory;

[0011] S4: Use a neural network to fit the relationship between the actual pose, the actual driving force and the difference between the theoretical driving force, and then correct the trajectory online using the trained neural network.

[0012] The beneficial effects of this invention are as follows:

[0013] This invention provides a flexible control method for the assembly of a fuse and detonating tube using dual robots based on neural networks. By considering the actual motion trajectory and environmental influences, it overcomes uncertainties and interferences, ensuring that the tightening force of the two robots in cooperation is appropriate. An impedance controller is constructed by combining a dynamic model and an RBFNN neural network. This method can achieve a good fit to the nonlinear model, allowing the robotic arms to operate simultaneously in both free space and contact space, and simultaneously track both constant and time-varying forces. When the two robotic arms are in contact and docking, the contact force at the end caps can be tracked. In the free space before docking, position and velocity can be tracked simultaneously. The improved impedance relationship, through the selected PID control, effectively reduces force overshoot and improves response speed. This invention can be applied to position and force control in environments with uncertainties and slight disturbances. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the dual-robot collaborative platform system structure used in an embodiment of the present invention;

[0015] Figure 2 This is a schematic diagram of the flexible control method for assembling a fuze and detonating tube using a dual-robot based on a neural network, as described in this invention.

[0016] Figure 3 This is a schematic diagram of the PID position / force impedance control closed-loop system based on adaptive method and neural network as described in an embodiment of the present invention.

[0017] Figure 4 Schematic diagram of RBFNN neural network structure Detailed Implementation

[0018] To enable those skilled in the art to better understand the present invention, the invention will be further described in detail below with reference to the accompanying drawings and specific implementation methods. The meanings of the specific parameters are explained in detail in the invention description.

[0019] This invention provides a neural network-based flexible control method for the assembly of a fuse and detonator using two robots. First, when the two robots pick up the fuse and detonator for flexible assembly, the actual contact force is obtained through a force sensor. Then, the desired force required for compliant assembly of the fuse and detonator is determined experimentally. Next, the dynamic model is analyzed. Due to significant errors in the movement of the two robotic arms, the dynamic model needs improvement. An impedance controller is designed to manage position and force in a unified manner, and a PID controller is introduced for contact force compensation. This allows for rapid convergence to the desired value and effectively reduces contact force overshoot, thus improving the control performance of the robotic arms. The difference between the average actual contact force and the desired force is calculated and fed into the impedance controller to reduce the error of the desired trajectory. The designed position tracking is used as the inner loop of the control system to estimate the relative velocity and contact torque at the end of the robotic arm. Finally, an RBFNN neural network is used to correct the uncertainties brought about by the environment and unknown forces online, enabling the fuse and detonator to flexibly dock under the cooperation of the two robots.

[0020] Example

[0021] This embodiment presents a flexible control method for the assembly of a fuse and detonating tube using a dual-robot assembly based on a neural network. It achieves this through methods such as... Figure 1 The illustrated dual-robot collaborative platform system includes two robots (two UR10 robots), a robot controller, two six-dimensional force / torque sensors, and two pneumatic grippers. The two pneumatic grippers are respectively mounted on the end caps of the two UR10 robot arms, and the two six-dimensional force / torque sensors are respectively mounted on the two pneumatic grippers. Each UR10 robot includes a robot body, a servo driver, and a servo motor.

[0022] like Figure 2 As shown, the flexible control method for assembling a fuse and detonator using a dual-robot based on a neural network includes the following steps:

[0023] S1: After the dual robots clamp the fuse and detonator to the expected position through the planned path, the magnitude of the expected contact force is first obtained through experiments on the force sensor. Then, based on the contact position after the trajectory planning of the dual robots, the estimated position of the end of the robotic arm is obtained, which prepares for the error of the subsequent dynamic model of the robot.

[0024] S2: Obtain the kinematic model of the two robots, design an improved adaptive impedance controller based on the magnitude of the desired force, combine the impedance control relationship with the PID control strategy, so that the actual force between the robots tracks the desired contact force, and convert the actual contact force into actual position control.

[0025] The adaptive impedance controller described in step S2 is established by the following steps:

[0026] S201: First, establish a coordinate system for the two UR10 robots that hold the two workpieces (detonating tube and fuse) to determine the position of the center of gravity of the two workpieces, so that the two robots are established in the same coordinate system, and model the cooperation of the two robots based on the master-slave framework.

[0027] S202: Deriving the robot's dynamic model using the Lagrange method:

[0028]

[0029] In the formula, q, Let M0(q) represent the joint angular position, velocity, and acceleration vectors of the robotic arm, respectively; M0(q) is a positive definite symmetric inertia matrix. G0(q) represents the Coriolis force and centripetal torque; G0(q) represents the gravity vector; τ represents the input joint torque of the robotic arm.

[0030] The external environment will affect the dynamic model during the modeling process, so its influence cannot be ignored. The complete dynamic model should be expressed as:

[0031]

[0032] τ e =J T (q)F e

[0033] The above expression can also be written as the following expression:

[0034]

[0035] in, This is denoted as an uncertainty term, i.e.

[0036]

[0037] In the formula, M(q), G(q) corresponds to the actual value of the robotic arm's dynamics model. The effect of joint friction on the robotic arm; τ d τ represents a bounded unknown disturbance, including unknown disturbances in the environment and the load at the end effector of the robotic arm; eF represents the contact torque when the robotic arm makes contact. e This is the end contact force when the robotic arm makes contact.

[0038] M(q), G(q) is represented separately as M0(q). G0(q) and the uncertain part ΔM(q), ΔG(q), these variables satisfy the following equation:

[0039] M(q) = M0(q) + ΔM(q)

[0040]

[0041] G(q) = G0(q) + ΔG(q)

[0042] The above expression can also be written as the following expression:

[0043]

[0044] in, This is denoted as an uncertainty term, i.e.

[0045]

[0046] S203: Construct the corresponding dynamic model of the robotic arm in the Cartesian coordinate system, as shown below:

[0047]

[0048] In the formula, X, These are respectively represented as the position vector, velocity vector, and acceleration vector of the robotic arm's end effector in the task space;

[0049]

[0050]

[0051] S204: Improve the dynamic model of the robotic arm by designing an impedance controller to unify the management of position and force. The expression for the impedance controller is:

[0052]

[0053] Among them, X r , M is represented by the reference position trajectory vector, reference velocity trajectory vector, and reference acceleration trajectory vector at the end of the robotic arm. d B d ,K dLet B represent the desired inertia parameter matrix, damping matrix, and stiffness parameter matrix, respectively. d ,K d These are usually unknown, so we use M, B, and K to represent them.

[0054] The desired contact force is incorporated into the impedance relationship, as shown in the following equation:

[0055]

[0056] This formula will include the contact force error E. f =F e -F d As the driving force of the impedance controller, it enables force tracking. When not in contact with the environment, F... e ,F d Both are 0. When the two objects are connected, the impedance controller will adjust according to the driving quantity E. f This is used to correct the motion of the robot's end effector, thereby enabling the tracking and control of contact forces.

[0057] S205: Combine the impedance controller designed in step S204 with the PID control strategy to obtain an adaptive impedance controller, the expression of which is:

[0058]

[0059] Among them, K P ,K i ,K d It is a diagonal positive definite parameter matrix.

[0060] Introducing PID control for contact force compensation can quickly converge to the desired value and effectively improve the phenomenon of contact force overshoot, thereby improving the control performance of the robotic arm.

[0061] S3: Collect and record the actual contact force between the two workpieces generated during each dual-robot collaborative assembly, calculate the difference between the average value of the actual contact force and the expected contact force, import the difference into the adaptive impedance controller to reduce the error of the expected trajectory, and use the designed position tracking equation as the inner loop of the control system to estimate the relative speed and contact torque at the end of the robotic arm.

[0062] Step S3 specifically includes:

[0063] S301: The end contact force is first studied in one direction to obtain the improved impedance control equation:

[0064]

[0065] The end contact force model can be expressed as:

[0066]

[0067] Error e f =f d -f e ,

[0068] In the formula, x r ,x,x e ,f e ,f d ,m,b,k,k p ,k i ,k d , to represent variables and parameters in a single direction.

[0069] Performing a Laplace transform on the improved impedance control equation proposed in step S301, we obtain:

[0070]

[0071]

[0072] Wherein, the expression in the formula

[0073]

[0074] Among them, T s (s)=(k d s 2 +k p s+k i (b) e s+k e )+(ms 3 +bs 2 +ks),

[0075] The steady-state tracking error should be expressed as:

[0076]

[0077] By converging the steady-state error at equilibrium to zero, a position tracking equation is established, and the reference position trajectory, after correction, can be designed as follows:

[0078]

[0079] The equation contains the desired force f d (t), environmental location x e Environmental stiffness k e and damping b e Impedance parameters m, b, k and k p k i k dBy inputting the magnitude of the desired force and other dynamic parameters, the reference position can be obtained.

[0080] S302: The designed position tracking equation is used as the inner loop of the control system to maximize the advantages of the position controller. A position vector X is defined. c As input to the inner loop of the location:

[0081]

[0082] Where Z = X r -X represents the trajectory correction error. The relationship between the correction error and contact force in the improved impedance controller design described above is as follows:

[0083]

[0084] Performing a Laplace transform on the above equation, we get

[0085]

[0086] The error data of the impedance controller for trajectory correction were obtained, and the error was compensated by the RBFNN neural network designed below.

[0087] S4: Combining the RBFNN neural network method, the neural network is used to fit the relationship between the difference between the real posture, the real driving force and the theoretical driving force. The trained neural network provides an online training method to correct the trajectory online, ensuring that the neural network can correct the uncertainty brought about by the environment and unknown forces online, and can compensate for new error sources of the robot.

[0088] Step S4 specifically includes:

[0089] S401: Design an adaptive neural network, using the RBFNN algorithm, to compensate for the uncertainties of the robot system and correct for uncertainties.

[0090] S402: Define ε = [ε1,...,ε] l ] T Let J(q) be the parameter vector in the Jacobian matrix, representing the relative velocity of the two objects at the end of the dual robotic arms and the contact torque τ of the dual robotic arms. e They can be represented as:

[0091]

[0092] τ e =J T (q)F e =Y f (q,F e )ε

[0093] in, and Y f (q,F e Let these be the end-effector velocity regression matrix and the joint torque regression matrix, respectively. Since kinematic parameters often have uncertainties, the Jacobian matrix is ​​not a fixed value. Therefore, we use the estimated Jacobian matrix. The spatial velocity in the formula The estimated value of the contact torque τ e Estimate The following formula represents:

[0094]

[0095]

[0096] in, This is the estimated parameter vector.

[0097] S403: Define a vector υ as follows:

[0098]

[0099] Where Λ=Λ T >0 is a positive definite matrix, E x =X c -X represents the position tracking error of the end effector.

[0100]

[0101] in, This represents the velocity tracking error at the end effector of the robotic arm in the task space.

[0102] S404: Defines a composite tracking error in the operating space.

[0103]

[0104] Combining this with the formula above, we get:

[0105]

[0106] Differentiating the above equation, we get:

[0107]

[0108] Then, define a composite tracking error for the space within each joint:

[0109]

[0110] S405: Define a virtual joint velocity as follows:

[0111]

[0112] Differentiate the above expression and redefine it

[0113]

[0114] The fusion derivation is obtained from the above formula.

[0115] S406: Define the system's state x1 = q and get

[0116]

[0117] in, Let be the input vector of RBFNN, and have

[0118]

[0119] Among them, the dynamics are M(x1), C(x1,x2), G(x1) and τ f (x2) are all unknown. M(x1), C(x1,x2), G(x1) and τ f Since (x2) is bounded and the Jacobian matrix J(x1) is also bounded, the unknown nonlinear function H(x) is also bounded. Therefore, the function can be approximated by RBFNN.

[0120] The position tracking algorithm of RBFNN designed based on the above function is as follows:

[0121]

[0122] Where L p >0 and L v >0 represents the position and velocity gain of the controller, u c This is denoted as the robust compensation term of the controller, used to compensate for external disturbances and reconstruction errors of the RBFNN.

[0123] Parameters of the Jacobian matrix and RBFNN weights The adaptive update rate is as follows:

[0124]

[0125]

[0126] Among them, Γ ε and Γ θ All are positive definite matrices. Therefore, the dynamic model can be compensated for, and the contact force F... e It can converge to the desired force F d .

[0127] In practical use, the controller needs to communicate with the six-dimensional force / torque sensor to obtain real-time contact force data. The implementation of the impedance control algorithm on the controller side involves programming the impedance formula in C++, and then compiling and importing the RBFNN neural network into the control system. The input of the impedance control algorithm is the desired motion trajectory and the external force after compensation by the neural network, and the output is the pose offset relative to the reference coordinate system.

[0128] The above are embodiments of the present invention, outlining the purpose and technical solutions of the invention. However, the scope of protection of the present invention is not limited thereto. Any technical solution of the present invention and similar substitutions to the inventive concept are within the scope of protection of the present invention.

Claims

1. A neural network-based flexible control method for fuze and booster dual-robot assembly, characterized in that, Includes the following steps: S1: After the dual robots clamp the fuse and detonator to the expected position through the planned path, the expected contact force is first estimated on the force sensor through experiment, and the position estimate of the end position of the robotic arm is obtained according to the contact position after the trajectory planning of the dual robots. S2: Obtain the kinematic model of the dual robots, design an adaptive impedance controller based on the desired contact force, combine the impedance control relationship with the PID control strategy, make the actual contact force between the two workpieces track the desired contact force, and convert the actual contact force into actual position control. S3: Collect the actual contact force between the two workpieces generated during each dual-robot collaborative assembly, calculate the difference between the average value of the actual contact force and the expected contact force, and import the difference into the adaptive impedance controller to reduce the error of the expected trajectory; S4: Use a neural network to fit the relationship between the actual pose, the actual driving force and the difference between the theoretical driving force, and then correct the trajectory online using the trained neural network.

2. The flexible control method for dual-robot assembly of fuse and detonator based on neural network as described in claim 1, characterized in that, The adaptive impedance controller in step S2 is established by the following steps: S21: Establish a coordinate system for the two robots holding the two workpieces to determine the position of the center of gravity of the two workpieces, so that the two robots are established in the same coordinate system, and model the cooperation of the two robots based on the master-slave framework. S22: Derive the dynamic model of the robot using the Lagrange method; S23: Construct the corresponding dynamic model of the robotic arm in the Cartesian coordinate system; S24: Improve the dynamic model of the robotic arm, design an impedance controller, and manage position and force in a unified manner; S25: Combine the impedance controller designed in step S24 with the PID control strategy to obtain an adaptive impedance controller.

3. The flexible control method for dual-robot assembly of fuse and detonator based on neural network as described in claim 2, characterized in that, Step S22 includes: The dynamic model of the robot is derived using the Lagrange method. The complete dynamic model should be expressed as: ； In the formula, These represent the joint angular position, velocity, and acceleration vectors of the robotic arm, respectively. It is a positive definite symmetric inertia matrix; Represents the Coriolis force and the centripetal torque; Represents the gravity vector; This represents the input joint torque of the robotic arm; The above expression can be written as: ； wherein denoted as the uncertainty term, i.e. ； In the formula, Corresponding to the actual values ​​of the robotic arm's dynamics model, The effect of joint friction on the robotic arm; This represents a bounded unknown disturbance, including unknown disturbances in the environment and the load at the end of the robotic arm; This represents the contact torque when the robotic arm makes contact. This is the end contact force when the robotic arm makes contact.

4. The neural network-based flexible control method for fuze and booster dual-robot assembly according to claim 3, characterized in that, The robotic arm dynamics model constructed in the Cartesian coordinate system in step S23 is as follows: ； In the formula, respectively represent the position vector, velocity vector and acceleration vector of the end effector of the robot arm in the task space. 。

Images