Kinematic error compensation technology for series industrial robot based on particle swarm algorithm

Abstract

The application discloses a kind of kinematic error compensation technology of series industrial robot based on particle swarm algorithm, including data batch acquisition, data storage, data processing and parameter identification and error compensation;Data batch acquisition is batched according to the acquisition data of single industrial robot, and the data storage library of batched acquisition data is established in the data storage library, including but not limited to pose point corresponding data, robot joint angle and end pose coordinate data robot data storage library;Data processing series robot kinematic error general model, Jacobian matrix condition data is used as objective function using particle swarm algorithm, and all industrial robots are executed to find optimal error matrix, and the optimal motion parameter error Jacobian matrix is obtained;Parameter identification and error compensation obtain high-precision MDH parameter.It can improve the accuracy of identification result, can meet the demand of large batch industrial robot error compensation, and improve the efficiency of large batch industrial robot error compensation.

Description

Technical Field

[0001] This invention relates to a serial industrial robot, and more particularly to a kinematic error compensation technology for serial industrial robots based on particle swarm optimization algorithm. Background Technology

[0002] As a typical representative of intelligent manufacturing equipment, serial industrial robots are affected by errors in machining, assembly, and linkage deformation, making it difficult to achieve a transformation and upgrade towards higher precision and intelligence. Currently, there are two methods to improve the end-effector positioning accuracy of industrial robots. The first is error prevention, which utilizes advanced machining methods to improve the precision of parts processing and assembly; the second is error compensation, which uses calibration technology to improve the robot's end-effector positioning accuracy—a software-based compensation method. However, error prevention requires high-precision machining technology and conditions, is costly, and has significant limitations. It cannot compensate for end-effector errors caused by factors such as mechanical wear and changes in component performance.

[0003] The core of error compensation methods is parameter identification. The process of parameter identification involves solving a set of error equations to obtain the robot's kinematic parameters and errors. Since there are many parameters to be determined in the error model, multiple sets of robot end-effector pose data are needed to construct the equation set for parameter identification. Currently, the most commonly used method for parameter identification is the least squares method, which is used to find the optimal solution that minimizes the error between theoretical and actual data.

[0004] Currently, there are two significant problems with kinematic error compensation for serial industrial robots:

[0005] (1) The data points selected in the error compensation process are randomly distributed in the workspace. This method of selecting points is random selection, which can easily cause the condition number of the error Jacobian matrix to be too large, resulting in inaccurate identification results.

[0006] (2) There is no efficient identification method for error compensation of large-scale industrial robots; Summary of the Invention

[0007] This invention addresses the current situation where improving the end-effector positioning accuracy of existing serial industrial robots can easily lead to an excessively large condition number in the Jacobian matrix, resulting in inaccurate identification results and difficulty in meeting the error compensation requirements of large-scale serial industrial robots. It provides a particle swarm optimization-based kinematic error compensation technology for serial industrial robots that can improve the accuracy of identification results, meet the error compensation requirements of large-scale industrial robots, and improve the error compensation efficiency of large-scale industrial robots.

[0008] To address the aforementioned technical problems, this invention employs a kinematic error compensation technology for serial industrial robots based on particle swarm optimization, comprising:

[0009] Batch data acquisition: This involves acquiring pose data from a single industrial robot's end effector. Based on the data acquired from this single robot, the data acquisition task is repeated for the remaining industrial robots to complete batch data acquisition. The data acquisition task for a single industrial robot includes the following methods for acquiring pose data from the end effector:

[0010] A1. Set up the experimental platform and correctly install the laser tracker system;

[0011] A2. Warm up and calibrate the laser tracker;

[0012] A3. Complete the establishment of the base coordinate system and the tool coordinate system;

[0013] A4. Select the robot end-effector position data acquisition point;

[0014] A5. The robot's movement is controlled by a robot teach pendant to output joint angles;

[0015] Data storage: After collecting the end-effector pose data of a single industrial robot, the data storage is performed. Robot data, including but not limited to pose point correspondence data, robot joint angles and end-effector pose coordinate data, is established in the data repository. The data storage is performed repeatedly for the remaining industrial robots to form a data repository.

[0016] Data processing: A general model of kinematic error for serial robots is used. The particle swarm optimization algorithm is used to find the Jacobian matrix as the objective function. The optimization error matrix is ​​iteratively performed on all industrial robots in the above part to obtain the optimal motion parameter error Jacobian matrix.

[0017] Parameter identification and error compensation: Perform parameter identification and large-scale error compensation to obtain high-precision MDH parameters.

[0018] It can quickly and efficiently achieve kinematic error compensation for large-scale serial industrial robots, enabling robot manufacturers to upgrade and improve existing welding robot models, further enhancing robot positioning accuracy and improving upgrade and development efficiency. It can improve the accuracy of identification results, meet the error compensation needs of large-scale industrial robots, and increase the efficiency of error compensation for large-scale industrial robots.

[0019] Preferably, the batch data acquisition method is as follows: First, data is acquired from the first robot. The end-effector pose of the serial robots is measured using a laser tracker. An experimental platform is first built, including the installation of the laser tracker and the test fixture. Then, the system is powered on and preheated to reach the set temperature. After preheating, the laser tracker is calibrated. After the preparation is completed, the robot pose is measured. First, the measurement coordinate system of the laser tracker and the base coordinate system of the robot are established separately, and the coordinate transformation between the two is performed. Then, the robot tool coordinate system is established so that the end-effector pose displayed in the robot teach pendant is the pose of the center of the target ball installed at the end. Finally, the robot end-effector pose points are acquired, and the joint angle corresponding to each end-effector pose point is transmitted. 1000±100 poses are selected as data acquisition points to ensure that these poses are distributed as evenly as possible throughout the workspace. Then, the above steps are repeated for the 2nd to nth robots.

[0020] This provides basic data support for subsequent data storage, data processing, parameter identification, and error compensation.

[0021] Preferably, the data storage method is to first store the data of the first robot, using Excel to store the collected robot data, storing the data of the first robot in the first sheet of Excel. Taking a six-axis serial robot as an example, and taking the data storage of a single robot as an example: the first column of Excel stores the data number, one pose point corresponds to a set of data, the second to seventh columns store the robot joint angles, and the last three columns store the robot end pose coordinate data.

[0022] It provides data support for subsequent data processing, parameter identification, and error compensation, and provides efficiency support for subsequent large-scale data processing, parameter identification, and error compensation.

[0023] Preferably, the data processing includes the following processing methods.

[0024] B1. First, the general model of kinematic errors for serial robots.

[0025] The error model was established based on the MDH (Modified DH Parameter Method) method of MDH parameter coordinates.

[0026] The general error model is as follows:

[0027] ΔP=J(q)ε (1)

[0028] J(q)=[J1(q)…J I (q)] (2)

[0029] ΔP=P r -P n(3)

[0030] Among them: J i (q) is the Jacobian error matrix of the parameter error, P r The position of the robot's end effector is represented by i = 1...I, where i = 1...I represents the 1-I joints of the serial industrial robot. n Let ΔP be the theoretical pose of the robot, and let ΔP be the error between the actual pose and the theoretical pose of the robot's end effector.

[0031] B2. Use the particle swarm optimization algorithm to find the 200 sets of data with the smallest condition number of the Jacobian matrix.

[0032] First, the formula for the constrained optimization problem is designed as follows:

[0033] min J1=cond(J)

[0034]

[0035] Where: J1 is the objective function, cond(J) represents the condition number of the Jacobian error matrix, and q(t) represents the angles of each joint of the robot. Indicates angular velocity. q represents angular acceleration. min , q represents the lower limits of angle, angular velocity, and angular acceleration, respectively. max , These represent the upper limits of angle, angular velocity, and angular acceleration, respectively; {s(q(t))}∈S represents the joint angle within the robot's entire motion space.

[0036] Due to the complexity of the objective function, traditional solution methods are very difficult. Particle swarm optimization (PSO) has been proven to be a very good intelligent optimization algorithm with advantages such as fast computation speed, few computational parameters, and ease of use. This invention uses PSO to solve the above-mentioned constrained optimization problem. The initial particle swarm population size is 300, the dimension is set to 200, and the maximum number of iterations is set to 1000. Using Matlab, the Excel spreadsheet configured in the data storage is loaded, and the optimization process is performed iteratively on n robots. The entire optimization process follows the standardized PSO algorithm procedure. Substituting the optimization results into formula (2) yields the most elegant ratio error matrix for the n robots, expressed by the formula:

[0037]

[0038] in, This represents the most elegant ratio error matrix for the first robot. Let represent the most elegant ratio error matrix of the nth robot.

[0039] It provides the optimal kinematic parameter Jacobian matrix for parameter identification and error compensation, and provides data and efficiency support for subsequent large-scale parameter identification and error compensation.

[0040] Preferably, the parameter identification and error compensation are performed using the least squares method to calculate the error, and the calculation formula is as follows:

[0041]

[0042] in: Let Δq be the most elegant ratio error matrix for the i-th robot. i Let ε be the end effector pose error of the i-th robot. i To calculate the error of the MDH parameters of the i-th robot, the error can be compensated for in the MDH parameters of the i-th robot to obtain high-precision MDH parameters.

[0043] It enables rapid and efficient kinematic identification and error compensation for large-scale serial industrial robots.

[0044] The beneficial effects of this invention are as follows: A method combining Excel and Matlab is proposed for large-scale error compensation; using the condition number of the Jacobian matrix of the error model of a serial industrial robot as the objective function, a particle swarm optimization algorithm is employed to find the optimal data points in the identification space; a method using particle swarm optimization to find the optimal data points in the identification space is proposed, solving the problem of random point selection in the current error compensation process, which leads to inaccurate identification; a large-scale error compensation method is proposed, solving the problem of low efficiency in current error compensation for large-scale industrial robots; it can quickly and efficiently achieve kinematic error compensation for large-scale serial industrial robots. This provides a way for general robot manufacturers to upgrade and improve existing welding robot models, further improving the robot's positioning accuracy and increasing the efficiency of upgrade and development. Attached image description:

[0045] Figure 1 This is a schematic diagram of the MDH link coordinate system structure of the kinematic error compensation technology for serial industrial robots based on particle swarm optimization algorithm of the present invention.

[0046] Figure 2 This is a schematic diagram of the framework structure of the kinematic error compensation technology for serial industrial robots based on the particle swarm optimization algorithm of the present invention.

[0047] Figure 3 This is a flowchart of the single robot end-effector pose data acquisition process for the kinematic error compensation technology of serial industrial robots based on particle swarm optimization algorithm, as described in this invention.

[0048] Figure 4This is a single robot data storage example of the kinematic error compensation technology for serial industrial robots based on particle swarm optimization in this invention. Detailed Implementation

[0049] Figure 1 , Figure 2 , Figure 3 , Figure 4 The illustrated embodiment presents a kinematic error compensation technique for serial industrial robots based on particle swarm optimization. The entire compensation technique consists of four steps: batch data acquisition, data storage, data processing, parameter identification, and error compensation (see...). Figure 2 The details are as follows:

[0050] Step 1. Batch Data Acquisition: Perform end-effector pose data acquisition for a single industrial robot. Based on the acquired data from this single robot, repeat the data acquisition task for the remaining industrial robots to complete batch data acquisition. The data acquisition task for a single industrial robot includes the following end-effector pose data acquisition methods: Figure 2 Taking a welding robot as an example, the robot collects data in batches of 1, 2, ... n, where n is a natural number. The collected data includes: laser tracker pose data 1 (thousands of data points), laser tracker pose data 2 (thousands of data points), ..., laser tracker pose data n (thousands of data points).

[0051] A1. Set up the experimental platform and correctly install the laser tracker system;

[0052] A2. Warm up and calibrate the laser tracker;

[0053] A3. Complete the establishment of the base coordinate system and the tool coordinate system;

[0054] A4. Select the robot end-effector position data acquisition point;

[0055] A5. The robot's movement is controlled by a robot teach pendant to output joint angles;

[0056] Step 2. Data Storage: After collecting the pose data of a single industrial robot's end effector, store the data in a data repository. This includes, but is not limited to, pose point correspondence data, robot joint angles, and end effector pose coordinate data. Repeat this process for the remaining industrial robots to collect and store data, forming a data repository. Figure 2 As shown, data package 1 is created in the data repository: MDH initial parameters, joint angles, and end-effector pose data for welding machine 1; MDH initial parameters, joint angles, and end-effector pose data for welding machine 2; ... MDH initial parameters, joint angles, and end-effector pose data for welding machine n.

[0057] Step 3. Data Processing: The general model of kinematic error of serial robots is used. The particle swarm optimization algorithm is used to find the Jacobian matrix condition data as the objective function. The 200 sets of data with the smallest Jacobian matrix condition number are used to iteratively optimize the error matrix for all industrial robots in the above part to obtain the optimal motion parameter error Jacobian matrix.

[0058] Step 4. Parameter Identification and Error Compensation: The least squares method is used to calculate the error, and error compensation is performed. Parameter identification and large-scale error compensation are then performed to obtain high-precision MDH parameters.

[0059] The batch data acquisition method involves first acquiring data from the first robot. A laser tracker is used to measure the end-effector pose of the serially connected robots. An experimental platform is first set up, including the installation of the laser tracker and test fixtures. The system is then powered on and preheated to the set temperature. After preheating, the laser tracker is calibrated. After preparation, robot pose measurement begins. First, the measurement coordinate system of the laser tracker and the robot's base coordinate system are established, and a coordinate transformation is performed between them. Then, the robot tool coordinate system is established so that the end-effector pose displayed in the robot teach pendant is the pose of the center of the target ball installed at the end. Finally, the robot end-effector pose points are acquired, and the joint angles corresponding to each end-effector pose point are transmitted. The overall process is as follows: Figure 3 As shown. Data acquisition points are selected at 1000 ± 100 poses, ensuring these poses are distributed as evenly as possible throughout the workspace. Then, the above steps are repeated for the 2nd to nth robots.

[0060] Data storage is performed by first storing the data for the first robot. The collected robot data is stored in an Excel spreadsheet, with the data from the first robot stored in the first sheet of the Excel spreadsheet. (Taking a six-axis serial robot as an example...) Figure 4 For example, data storage for a single robot is stored in the first column of an Excel file, where each pose point corresponds to a set of data. Columns 2-7 store the robot joint angles, and the last three columns store the robot end effector pose coordinates.

[0061] Data processing includes the following processing methods

[0062] B1. First, the general model of kinematic errors for serial robots.

[0063] The error model is established based on the MDH (Modified DH Parameter Method) parameter method. A coordinate diagram of the MDH parameter method is shown below. Figure 1 As shown. The general error model is as follows:

[0064] ΔP=J(q)ε (1)

[0065] J(q)=[J1(q)…JI (q)] (2)

[0066] ΔP=P r -P n (3)

[0067] Among them: J i (q) is the Jacobian error matrix of the parameter error, P r The position of the robot's end effector is represented by i = 1...I, where i = 1...I represents the 1-I joints of the serial industrial robot. n Let ΔP be the theoretical pose of the robot, and let ΔP be the error between the actual pose and the theoretical pose of the robot's end effector.

[0068] B2. Use the particle swarm optimization algorithm to find the 200 sets of data with the smallest condition number of the Jacobian matrix.

[0069] First, the formula for the constrained optimization problem is designed as follows:

[0070] min J1 = cond(J)

[0071]

[0072] Where: J1 is the objective function, cond(J) represents the condition number of the Jacobian error matrix, and q(t) represents the angles of each joint of the robot. Indicates angular velocity. q represents angular acceleration. min , q represents the lower limits of angle, angular velocity, and angular acceleration, respectively. max , These represent the upper limits of angle, angular velocity, and angular acceleration, respectively; {s(q(t))}∈S represents the joint angle within the robot's entire motion space.

[0073] Due to the complexity of the objective function, traditional solution methods are very difficult. Particle swarm optimization (PSO) has been proven to be a very good intelligent optimization algorithm with advantages such as fast computation speed, few computational parameters, and ease of use. This invention uses PSO to solve the above-mentioned constrained optimization problem. The initial particle swarm population size is 300, the dimension is set to 200, and the maximum number of iterations is set to 1000. Using Matlab, the Excel spreadsheet configured in the data storage is loaded, and the optimization process is performed iteratively on n robots. The entire optimization process follows the standardized PSO algorithm procedure. Substituting the optimization results into formula (2) yields the most elegant ratio error matrix for the n robots, expressed by the formula:

[0074]

[0075] in, This represents the most elegant ratio error matrix for the first robot. Let represent the most elegant ratio error matrix of the nth robot.

[0076] Parameter identification and error compensation use the least squares method to calculate the error. The calculation formula is as follows:

[0077]

[0078] in: Let Δq be the most elegant ratio error matrix for the i-th robot. i Let ε be the end effector pose error of the i-th robot. i To calculate the error of the MDH parameters of the i-th robot, the error can be compensated for in the MDH parameters of the i-th robot to obtain high-precision MDH parameters.

[0079] The above technical solutions and key points of the invention

[0080] 1) Establishment of error model for general serial industrial robot: First, establish the kinematic selection transformation matrix model of serial industrial robot based on MDH (improved DH parameter method); then establish the error model of a single link; finally, establish the kinematic parameter error model of serial kinematic chain.

[0081] (2) Method for finding the optimal data points in space: The Jacobian matrix condition number of the error model of the serial industrial robot is used as the objective function, and the particle swarm optimization algorithm is used to find the optimal data points.

[0082] (3) Large-scale error compensation method: A large-scale error compensation method is proposed by combining Excel and Matlab;

[0083] The above content and structure describe the basic principles, main features, and advantages of the product of this invention, which should be understood by those skilled in the art. The examples and descriptions above are merely illustrative of the principles of this invention. Various changes and modifications can be made to this invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A kinematic error compensation technology for serial industrial robots based on particle swarm optimization, characterized in that: include Batch data acquisition: This involves acquiring pose data from a single industrial robot's end effector. Based on the data acquired from this single robot, the data acquisition task is repeated for the remaining industrial robots to complete batch data acquisition. The data acquisition task for a single industrial robot includes the following methods for acquiring pose data from the end effector: A1. Set up the experimental platform and correctly install the laser tracker system; A2. Warm up and calibrate the laser tracker; A3. Complete the establishment of the base coordinate system and the tool coordinate system; A4. Select the robot end-effector position data acquisition point; A5. The robot's movement is controlled by a robot teach pendant to output joint angles; Data storage: After collecting the end-effector pose data of a single industrial robot, the data storage is performed. Robot data, including but not limited to pose point correspondence data, robot joint angles and end-effector pose coordinate data, is established in the data repository. The data storage is performed repeatedly for the remaining industrial robots to form a data repository. Data processing: A general kinematic error model for serial industrial robots is established. The particle swarm optimization algorithm is used to find the Jacobian matrix as the objective function. The optimization error matrix is ​​iteratively performed on all the above industrial robots to obtain the optimal motion parameter error Jacobian matrix. Parameter identification and error compensation: Perform parameter identification and large-scale error compensation to obtain high-precision MDH parameters; The data processing includes the following processing methods: B1. First, establish a general kinematic error model for serial industrial robots. The error model is established based on the MDH (Modified DH Parameter Method) parameter method, and the general error model is as follows: (1) (2) (3) in: The Jacobian error matrix represents the parameter error. This indicates the actual pose of the robot's end effector. Indicates serial industrial robots One joint, For the robot's theoretical pose, This represents the error between the actual pose and the theoretical pose of the robot's end effector. B2. Use the particle swarm optimization algorithm to find the 200 sets of data with the smallest condition number of the Jacobian matrix. First, the formula for the constrained optimization problem is designed as follows: (4) in: Let be the objective function. The condition number represents the Jacobi error matrix. Indicates the angles of each joint of the robot. Indicates angular velocity. Indicates angular acceleration; , , These represent the lower limits of angle, angular velocity, and angular acceleration, respectively. , , These represent the upper limits of angle, angular velocity, and angular acceleration, respectively. This indicates that the joint angles are within the robot's entire motion space; The particle swarm optimization algorithm is used to solve the above constrained optimization problem. The initial particle swarm population size is 300, the dimension is set to 200, and the maximum number of iterations is set to 1000. Using Matlab, the Excel spreadsheet configured in the data storage is loaded, and the optimization process is performed iteratively on n robots. The entire optimization process is carried out in accordance with the standardized particle swarm optimization algorithm process. Substituting the optimization results into formula (2) yields the most elegant ratio error matrix of n robots, which is expressed by the formula: (5) in, This represents the most elegant ratio error matrix for the first robot. This represents the most elegant ratio error matrix for the nth robot; The parameter identification and error compensation described above use the least squares method to calculate the error. The calculation formula is as follows: （6） in: For the first The most elegant ratio error matrix of the robot, For the first The end effector pose error of the robot. For the calculated first The MDH parameter error of the first robot is compensated to the first... High-precision MDH parameters can be obtained from the MDH parameters of the robot.

2. The kinematic error compensation technology for serial industrial robots based on particle swarm optimization algorithm as described in claim 1, characterized in that: The data batch acquisition method is as follows: First, data is acquired from the first robot. A laser tracker is used to measure the end-effector pose of the serially connected robots. An experimental platform is first set up, including the installation of the laser tracker and the test fixture. Then, the system is powered on and preheated to reach the set temperature. After preheating, the laser tracker is calibrated. After the preparation is completed, the robot pose is measured. First, the measurement coordinate system of the laser tracker and the base coordinate system of the robot are established separately, and the coordinate transformation between the two is performed. Then, the robot tool coordinate system is established so that the end-effector pose displayed in the robot teach pendant is the pose of the center of the target ball installed at the end. Finally, the robot end-effector pose points are acquired, and the joint angle corresponding to each end-effector pose point is transmitted. 1000±100 poses are selected for data acquisition points, so that these poses are distributed as evenly as possible throughout the workspace. Then, the above steps are repeated for the 2nd to nth robots.

3. The kinematic error compensation technology for serial industrial robots based on particle swarm optimization algorithm as described in claim 1, characterized in that... The data storage method is as follows: First, the data of the first robot is stored. The collected robot data is stored in Excel. The data of the first robot is stored in the first sheet of Excel. Taking a six-axis serial robot as an example, the data storage of a single robot is as follows: The first column of Excel stores the data number, one pose point corresponds to a group of data, the second to seventh columns store the robot joint angles, and the last three columns store the robot end pose coordinate data.

Images