Parallel robot system and material sorting method

Abstract

The invention belongs to the technical field of robots, and provides a parallel robot system and a material sorting method. The camera module is configured to identify the material in real time to determine the position information of the material so as to obtain the center point coordinate data of the material; the controller is provided with a sensing layer and a motion control layer, the sensing layer is used for storing center point coordinate data of the materials, and the motion control layer is configured to determine the displacement of each joint of the parallel robot body according to the center point coordinate data; and the driver is configured to output the driving force of each joint according to the displacement generated by the controller so as to drive the parallel robot body to move to grab the materials. According to the parallel robot system provided by the invention, three neural networks are introduced according to a neural network online learning control algorithm and are used for respectively approaching an inertia force term, a Coriolis force and centripetal force term and a gravity term of the robot to ensure asymptotic stability of the control system.

Description

Technical Field

[0001] The present invention relates to the field of robot technology, and in particular to a parallel robot system and a material sorting method. Background Art

[0002] With the continuous development of industrial automation, parallel robots are increasingly used in efficient sorting tasks. This type of robot has high rigidity, high precision and fast movement speed, which is very suitable for complex production environments. Today, parallel robots usually use four degrees of freedom. They are composed of multiple kinematic chains and adopt a closed-loop structure, which enables them to achieve high load capacity while maintaining high motion accuracy. Its basic structure includes a fixed base, a movable platform and multiple drive mechanisms. Compared with serial robots, four-degree-of-freedom parallel robots have better rigidity and stability, and are suitable for precision grasping and sorting.

[0003] At the same time, in terms of the integration of visual technology, many researchers have begun to use deep learning algorithms to improve object recognition capabilities. Domestic and foreign research teams have combined convolutional neural networks (CNNs) with their four-degree-of-freedom robots to achieve rapid recognition of objects of different shapes and colors, significantly improving the accuracy and efficiency of sorting. For example, in the production line of electronic components, parallel robots are used to quickly grasp and place small components. By introducing this robot, production efficiency has been successfully increased by 30% and the error rate caused by manual operation has been reduced.

[0004] Although the four-degree-of-freedom parallel robot has shown strong advantages in the field of sorting, it still faces some challenges: accurate modeling is crucial for efficient control, but due to the complex structure, the model establishment and verification process is often difficult; how to maintain high-precision positioning of the robot during high-speed operation is a key issue, especially when dealing with objects of different sizes and shapes. The system needs to process data from sensors in real time to ensure that the robot can respond quickly to environmental changes, which places high demands on computing power and algorithm efficiency; the effective integration of the robot with the visual system, controller and software platform is a prerequisite for achieving efficient sorting, but the compatibility and real-time performance between different programming languages ​​​​and systems are difficult to meet the requirements of high-speed sorting. Summary of the invention

[0005] The object of the present invention is to provide a parallel robot system and a material sorting method to solve the above problems.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is:

[0007] A parallel robot system, comprising:

[0008] Parallel robot body;

[0009] A camera module is configured to identify the material in real time and determine the location information of the material to obtain the coordinate data of the center point of the material;

[0010] A controller, wherein a sensing layer and a motion control layer are configured, wherein the sensing layer is used to store the center point coordinate data of the material, and the motion control layer is configured to determine the displacement of each joint of the parallel robot body according to the center point coordinate data;

[0011] A driver configured to output a driving force of each joint according to a displacement generated by the controller, so as to drive the parallel robot body to move and grasp the material;

[0012] The displacement is obtained by solving the pre-built kinematic model, and the displacement of each joint is adjusted in real time through multiple neural networks.

[0013] Preferably, the kinematic analysis of the parallel robot body is performed using a closed-loop vector law to obtain its kinematic inverse solution as its kinematic model.

[0014] Preferably, the velocity Jacobian matrix of the parallel robot system is calculated based on the inverse kinematic solution to obtain a set of singular positions, and the set of singular positions forms an invalid working space of the parallel robot system.

[0015] Preferably, a dynamic model of the parallel robot body is formed by a kinematic model, and the controller generates the displacement of each joint of the parallel robot body according to the dynamic model.

[0016] Preferably, a plurality of neural networks are formed by pre-training a dynamics model.

[0017] Preferably, there are three neural networks, which are used to approximate the inertial force term, centripetal force, Coriolis force term and gravity term of the parallel robot respectively, and the online learning update law of the neural network weights is configured to use the neural network to perform online learning of the centripetal force, Coriolis force term and gravity term of the parallel robot.

[0018] Preferably, a position encoder is configured on each joint, and the joint position obtained by the position encoder is determined to be in a safe state based on a set of singular positions; when the joint position is not in a safe state, the motion control layer is configured to switch the operating mode of the motor corresponding to the joint from a synchronous periodic torque mode to a synchronous periodic position mode.

[0019] Preferably, the motion control layer is configured to send a position signal of the current joint position to the driver while changing the operation mode to stop the driver from outputting driving force.

[0020] The present invention also provides a parallel robot material sorting method, comprising the steps of:

[0021] Identify materials in real time and obtain the coordinate data of the center point of the materials;

[0022] The displacement of each joint of the parallel robot is determined according to the coordinate data of the center point; the displacement is obtained by solving the pre-constructed kinematic model, and the displacement of each joint is adjusted in real time through multiple neural networks; the dynamic model of the parallel robot is formed through the kinematic model, and the displacement of each joint of the parallel robot is generated according to the dynamic model;

[0023] The driving force of each joint is output according to the generated displacement to drive the parallel robot to move and grab the material;

[0024] Preferably, there are three neural networks, which are used to approximate the inertial force term, centripetal force, Coriolis force term and gravity term of the parallel robot respectively, and an online learning update law for the neural network weights is configured to utilize the neural network to perform online learning on the centripetal force, Coriolis force term and gravity term of the parallel robot; the neural network is formed by pre-training of a dynamic model.

[0025] Compared with the prior art, the present invention has at least the following beneficial effects:

[0026] (1) Three neural networks are introduced into the online learning control algorithm based on the neural network pre-trained by the dynamic model. They are used to approximate the robot's inertial force, centripetal force, Coriolis force, and gravity term respectively. An adaptive control algorithm is used to ensure the asymptotic stability of the control system.

[0027] (2) The dynamic model of the parallel robot is formed by combining the principle of virtual work with its kinematic model. In the dynamic model parameters, the inertial dynamics, which accounts for a large proportion of the dynamics, is directly introduced into the control algorithm. The rest is learned online through a neural network, thereby improving the dynamic response capability of the system while meeting the computational efficiency.

[0028] (3) At the same time, a sliding mode control law is designed to handle unknown disturbances in the system and improve the robustness of the system. A continuous integral robust control algorithm is introduced to avoid sudden changes in the signal, further improving the robustness of the control system and meeting the high-precision response requirements of the parallel robot under high dynamic motion.

[0029] (4) The safety protection function is realized based on the cooperation between hardware and software. The position encoders equipped on the four active joints can provide real-time feedback on the motor position. At the same time, by changing the operation mode of the active joint motor and converting the torque signal into a position signal to the driver, the active joint can make an emergency stop operation to avoid over-limit and "breakdown". BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0031] Figure 1 Schematic diagram of the framework of the parallel robot system in the embodiment provided by the present invention;

[0032] Figure 2 is a schematic structural diagram of a parallel robot body in an embodiment provided by the present invention;

[0033] Figure 3 is a schematic diagram of a kinematic model of a parallel robot system in an embodiment provided by the present invention;

[0034] Figure 4 It is a flow chart of the safety protection function of the parallel robot system in the embodiment provided by the present invention. DETAILED DESCRIPTION

[0035] It should be noted that all directional indications in the embodiments of the present invention (such as up, down, left, right, front, back, etc.) are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0036] In addition, the technical solutions between the various embodiments of the present invention can be combined with each other, but it must be based on the fact that ordinary technicians in the field can implement it. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0037] The following are specific embodiments of the present invention, and the technical solutions of the present invention are further described in conjunction with the accompanying drawings, but the present invention is not limited to these embodiments.

[0038] like Figure 1 and Figure 2 As shown, this embodiment provides a parallel robot, which includes a parallel robot body 10, a controller 20, a driver 30 and a camera module 40;

[0039] The parallel robot body includes a basic static platform 110 and a dynamic platform 120, wherein the dynamic platform 120 has four degrees of freedom, namely, three translations (along the X, Y, and Z axes) and one rotation (around the Z axis). The dynamic platform is driven by the controller 20 and the driver 30 to complete the material grabbing operation, thereby realizing the material sorting operation.

[0040] During the material sorting work, the camera module 40 is used to identify the materials. When a material passes by, the camera module takes real-time photos for identification, thereby determining the position of the material and obtaining the coordinate data of the center point of the material, that is, the end coordinate data of the moving platform 120. The coordinate data of the center point of the material will then be sent to the controller 20.

[0041] The controller 20 is configured with a perception layer and a motion control layer. The perception layer is used to store the center point coordinate data of the material, and the motion control layer is configured to determine the displacement of each joint of the parallel robot body according to the center point coordinate data. The center point coordinate data stored in the perception layer will adopt the ADS communication protocol and be transmitted to the motion control layer in the form of variable mapping.

[0042] Then, the driver can output the driving force of each joint according to the displacement generated by the controller to drive the parallel robot body to move and grab the material. The driving force is sent by the driver to the motors corresponding to each joint on the parallel robot body in the form of current value, thereby causing each joint position to move.

[0043] The displacement generated by the controller is obtained by solving the pre-built kinematic model. Figure 3 As shown, taking the M-type "4PPA-2PAR" four-degree-of-freedom special "4-2-1" geometric configuration parallel robot body as an example, after determining the end coordinate data of the moving platform, the parallel robot kinematic model is established through the closed-loop vector law. The established parallel robot kinematic model is an inverse kinematic model, which can be expressed as:

[0044]

[0045] The meanings of the various parameters are as follows: (x, y, z, θ) represent the translation of the moving platform along the X, Y, and Z axes of the world coordinate system and the rotation around the Z axis, respectively; (a, b, lu, l d , h1, h2) represent the length of the moving platform, the distance between the guide rails, the length of the proximal parallelogram rod, the length of the distal parallelogram rod, the distance between points C and E, and the distance between points H and P, respectively.

[0046] After obtaining the inverse kinematic solution as its kinematic model, the velocity Jacobian matrix of the parallel robot system is calculated simultaneously to obtain a set of singular positions, which forms the invalid working space of the parallel robot system.

[0047] It uses the closed-loop vector method for velocity modeling, eliminates the passive joint velocity, and obtains the velocity mapping relationship between the active joint and the moving platform as follows:

[0048]

[0049] in, Represents the displacement of the four active joints along the Y-axis in the world coordinate system {B}; Represents the position vector of the two linear guides along the X-axis in the world coordinate system {B}; and Respectively and The unit direction vector of ; Represents the distance from the origin of the local coordinate system {E} to P in the global coordinate system {B} j The position vector of the point;

[0050] By differentiating the resulting closed-loop position equation with respect to time:

[0051]

[0052] in They represent the translational velocities of the active joints, and Respectively represent the connecting rod l u , l d and the angular velocity of the moving platform, and The coefficient matrices represent the active joints, the connecting rods, and u , l d and the mapping relationship between the velocity of the moving platform and the velocity of the origin of the local coordinate system {E};

[0053] The expression of the positive Jacobian matrix obtained by calculation is:

[0054]

[0055] in,

[0056]

[0057] The inverse Jacobian matrix expression is:

[0058]

[0059] Singular positions include forward singularities and reverse singularities. Forward singularity is a special configuration in which a parallel mechanism instantaneously obtains one or more degrees of freedom during motion. Its geometric meaning is that when the active joint is completely locked, the end moving platform can still produce instantaneous motion. When the positive Jacobian matrix G is not full rank, that is, det(G) = 0, the parallel robot body has four forward singular configurations. Reverse singularity is a special configuration in which a parallel mechanism instantaneously loses one or more degrees of freedom during motion, which usually occurs at the boundary of the parallel mechanism's workspace. When the matrix J q Not full rank, that is, det(Jq )=0, the parallel robot body has two reverse singular configurations.

[0060] According to the establishment of the kinematic model and the use of the velocity Jacobian matrix solved by kinematics, the set of singular positions of the parallel robot body is statistically calculated to avoid the singular position of picking materials for sorting operations, thereby avoiding interference between the components of the parallel robot body during movement.

[0061] The kinematic model is used to form the dynamic model of the parallel robot body. The controller generates the displacement of each joint of the parallel robot body according to the dynamic model, and adjusts the displacement of each joint in real time through three neural networks.

[0062] In this embodiment, the velocity parameters obtained by derivation of the closed-loop vector law are used as the velocity, acceleration in the centripetal force and Coriolis force terms in the dynamic model, and the velocity parameters in the force-Jacob ratio, thereby forming a dynamic model of the parallel robot.

[0063] The derived dynamic model is then used to pre-train the neural network, thereby introducing dynamic information and improving the transient response accuracy of the robot.

[0064] In the neural network online learning control algorithm pre-trained according to the dynamic model adopted in this embodiment, three neural networks are introduced, which are used to approximate the robot's inertial force term, centripetal force and Coriolis force term, and gravity term respectively. An adaptive control algorithm is used to ensure the progressive stability of the control system.

[0065] Furthermore, an online learning update law for the neural network weights is configured in the algorithm to utilize the neural network to perform online learning on the centripetal force, Coriolis force term and gravity term of the parallel robot, and to directly introduce the inertial force term with a small computational burden and a large proportion in the dynamic model into the control algorithm, thereby significantly improving the steady-state tracking performance.

[0066] At the same time, a sliding mode control law was designed to deal with unknown disturbances in the system and improve the robustness of the system. A continuous integral robust control algorithm was introduced to avoid sudden changes in the signal, further improving the robustness of the control system and meeting the high-precision response requirements of the parallel robot under high dynamic motion.

[0067] In addition, the parallel system robot system also has a safety protection function, which is mainly achieved through the cooperation between hardware and software. The sensors equipped on the four active joints are absolute position encoders, which can provide real-time feedback on the position of the motor, thereby achieving software-level safety protection.

[0068] At the software level, it is based on the set obtained by solving the singular position of the parallel robot, so as to first perform the safety judgment of the four active joint positions in the control loop execution function of the loop execution, and then perform subsequent operations after judging that it is in a safe state. If it is judged that the joint position exceeds the limit or the distance is too close, the motor mode will be switched and the motor will be stopped urgently.

[0069] Its specific physical meaning is that the active joint is always within the travel limit, and there is no collision between the two active joints on the same side guide rail. The specific collision gap threshold can be adjusted according to actual conditions.

[0070] The realization of this function mainly depends on how to perform the emergency stop operation of the motor. When the position of the active joint motor exceeds the limit or is abnormal, the usual solution is to remove the power or enable the motor. However, for the parallel robot system of this embodiment, when the joint position exceeds the limit, the entire robot will still be in motion. At this time, after the power is removed, the robot still has a certain inertia and will continue to move in the direction of the exceeded limit. This unsafe cut-off does not meet the requirements of safe production.

[0071] In particular, when the active joints of the parallel robot are installed with, for example, magnetic levitation linear motors, they do not have a mechanical self-locking function. Therefore, if the joints are de-energized when an over-limit occurs, the entire mechanism will "break down".

[0072] Therefore, this embodiment proposes a method of switching the active joint motor operation mode from a synchronous periodic torque mode (CST) to a synchronous periodic position mode (CSP), and at the same time converting the torque signal sent during normal control into a position signal. At this time, only the zero position of the motor position needs to be sent to the downstream driver, and the active joint will perform an emergency stop operation, and there will be no over-limit or "breakdown" situation.

[0073] Therefore, the parallel robot provided in this embodiment, in addition to building the control system hardware including the parallel robot body, driver, and controller, is also configured with control algorithms including perception layer command reception, motion control layer displacement signal output, and safety protection, so as to realize material sorting operations.

[0074] The specific embodiments described herein are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

Claims

1. A parallel robot system, characterized in that: include: Parallel robot body; A camera module is configured to identify the material in real time and determine the location information of the material to obtain the coordinate data of the center point of the material; A controller, wherein a sensing layer and a motion control layer are configured, wherein the sensing layer is used to store the center point coordinate data of the material, and the motion control layer is configured to determine the displacement of each joint of the parallel robot body according to the center point coordinate data; A driver configured to output a driving force of each joint according to the displacement generated by the controller, so as to drive the parallel robot body to move and grasp the material; The displacement is obtained by solving a pre-constructed kinematic model, and the displacement of each joint is adjusted in real time through multiple neural networks.

2. A parallel robot system according to claim 1, characterized in that: The kinematics of the parallel robot body is analyzed by the closed-loop vector law to obtain its inverse kinematic solution as its kinematic model.

3. A parallel robot system according to claim 2, characterized in that: The velocity Jacobian matrix of the parallel robot system is calculated according to the inverse kinematic solution to obtain a set of singular positions, and the set of singular positions forms an invalid working space of the parallel robot system.

4. A parallel robot system according to claim 1 or 2, characterized in that: The kinematic model is used to form a dynamic model of the parallel robot body, and the controller generates the displacement of each joint of the parallel robot body according to the dynamic model.

5. A parallel robot system according to claim 4, characterized in that: The plurality of neural networks are formed by pre-training the dynamics model.

6. A parallel robot system according to claim 5, characterized in that: There are three neural networks, which are used to approximate the inertial force, centripetal force, Coriolis force and gravity term of the parallel robot respectively. At the same time, an online learning update law of the neural network weights is configured to use the neural network to perform online learning on the centripetal force, Coriolis force and gravity term of the parallel robot.

7. The parallel robot system according to claim 1, characterized in that: A position encoder is configured on each of the joints, and based on the set of singular positions, it is determined that the joint position obtained by the position encoder is in a safe state; when the joint position is not in a safe state, the motion control layer is configured to switch the operating mode of the motor corresponding to the joint from a synchronous periodic torque mode to a synchronous periodic position mode.

8. A parallel robot system according to claim 7, characterized in that: The motion control layer is configured to send a position signal of a current joint position to the driver while changing the operation mode, so as to stop the driver from outputting a driving force.

9. A parallel robot material sorting method, characterized in that: Includes steps: Identify materials in real time to obtain the center point coordinate data of the materials; Determine the displacement of each joint of the parallel robot according to the center point coordinate data; the displacement is obtained by solving a pre-constructed kinematic model, and the displacement of each joint is adjusted in real time through multiple neural networks; Forming a dynamic model of the parallel robot through the kinematic model, and generating the displacement of each joint of the parallel robot according to the dynamic model; The driving force of each joint is output according to the generated displacement to drive the parallel robot to move and grasp the material.

10. A parallel robot material sorting method according to claim 9, characterized in that: There are three neural networks, which are respectively used to approximate the inertial force term, centripetal force, Coriolis force term and gravity term of the parallel robot, and an online learning update law of the neural network weights is configured to use the neural network to perform online learning on the centripetal force, Coriolis force term and gravity term of the parallel robot; The neural network is formed by pre-training the dynamics model.

Images