Industrial robot end load parameter rapid online identification method and system

Abstract

This invention relates to a method and system for rapid online identification of end-effector load parameters of industrial robots, belonging to the field of robot parameter identification technology. The method includes: collecting joint state and joint torque data during natural static or low-speed short windows that occur during actual operations such as completion of grasping, before placement, path switching, or tool changing; reducing the impact of low-speed nonlinear friction on torque difference by pre-establishing an unloaded reference torque model that includes nonlinear friction compensation; and constructing a four-dimensional linear identification model of load parameters by embedding the end-effector load mass and the first-order moment of the center of mass into the robot's minimum inertia parameter expression. This allows for rapid online estimation of the end-effector load mass and center of mass position without requiring online recovery of the complete load inertia tensor or additional execution of a dedicated excitation trajectory. This invention not only avoids the complex computational problems associated with traditional high-dimensional parameter identification but also achieves rapid, real-time online identification of end-effector load parameters.

Description

Technical Field

[0001] This invention relates to the field of robot parameter identification technology, and in particular to a method and system for rapid online identification of end-effector load parameters of industrial robots. Background Technology

[0002] In industrial robot automation scenarios, the robot's end effector typically forms a rigid connection with external workpieces, tools, or other loads via flanges, grippers, suction cups, or various end effectors. During diverse workflows such as workpiece gripping, material handling, precision assembly, and tool changing, the mass, center of mass, and inertial parameters of the robot's end effector load dynamically change. Since these changes directly affect the driving torque required by each joint of the robot, if the control system still uses no-load parameters or fixed nominal load parameters, it can easily lead to a mismatch between the robot's dynamics model and the actual system. This results in increased joint torque prediction errors, decreased gravity compensation effectiveness, and reduced trajectory tracking accuracy, and in severe cases, may even affect the robot arm's operational stability and safety.

[0003] To address the dynamic control mismatch caused by variable end-effector loads, various robot end-effector load identification technologies have been developed. The mainstream solutions fall into three categories: offline load identification, online load identification, and external sensor-assisted load estimation. Each solution has inherent technical limitations, making it difficult to adapt to the high-frequency, variable-load, and continuous production demands of industrial environments. Offline load identification typically treats the end-effector load as an independent rigid body. By designing a dedicated excitation trajectory, the robot performs sufficiently stimulating motions under load, collecting joint position, velocity, acceleration, and joint torque data. Then, based on the robot's dynamics model, parameters such as the load's mass, center of mass position, and inertia tensor are calculated. This method offers a relatively complete modeling process and obtains comprehensive load dynamic parameters, making it suitable for laboratory calibration or fixed tool load scenarios. However, in actual industrial production, offline identification usually requires the robot to pause normal operations and perform additional excitation actions, resulting in long identification times. Furthermore, the excitation trajectory design and data acquisition process are complex, making it difficult to adapt to dynamic variable-load tasks such as continuous grasping, rapid handling, or frequent tool switching. Online load identification methods attempt to update load parameters in real time based on joint states and driving torque data during robot operation. Existing online methods typically include recursive least squares, extended Kalman filtering, online estimation based on parameter difference, and identification methods based on updating overall robot dynamic parameters. These methods offer better real-time performance compared to offline methods, but many still aim to recover the complete inertial parameters of the load, resulting in high-dimensional parameters and relatively complex calculations. Some methods also require the robot to perform specific movements or meet certain excitation conditions during the online phase to ensure parameter identifiability. In fast-paced industrial environments, robots often struggle to perform dedicated identification actions for extended periods, thus limiting the deployment of existing online methods in continuous operations. Furthermore, existing technologies can directly acquire load parameters by incorporating external measurement devices such as six-dimensional force sensors and end-effector weighing devices, avoiding complex dynamic model identification calculations and reducing algorithm design complexity. However, this technical solution relies on additional hardware, which not only significantly increases the hardware cost and structural complexity of the robot system, but is also limited by factors such as the installation space at the robot's end effector, sensor calibration errors, on-site vibration, dust, and electromagnetic interference, making it prone to problems such as high noise in the detection signal and unstable recognition accuracy. At the same time, for existing industrial robots that have already been deployed and put into production, retrofitting with external sensors is difficult and lacks versatility, making it impossible to achieve large-scale promotion and application of multiple robot models and multiple operating scenarios.

[0004] In summary, existing industrial robot end-effector load identification technologies generally suffer from poor adaptability, insufficient real-time performance, high hardware modification costs, and difficulty in adapting to continuous operation with dynamic load changes. They cannot simultaneously meet the industrial application requirements of identification accuracy, real-time performance, versatility, and low cost. There is an urgent need to develop a load identification solution to solve the problems of the existing technologies. Summary of the Invention

[0005] Therefore, the technical problem to be solved by the present invention is to overcome the problems of poor adaptability and insufficient real-time identification capability of the robot end-effector load identification scheme in the prior art, while avoiding the problems of high hardware modification cost and weak versatility of sensor identification scheme, as well as the difficulty of traditional identification methods in ensuring identification accuracy under dynamic load conditions.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a method for rapid online identification of end-effector load parameters of an industrial robot, comprising: S1. Pre-identify the robot's rigid body dynamics parameters and joint friction parameters, and establish an unloaded reference torque model; S2. Select the load mass and first moment of the centroid at the robot's end effector as the online identification objects, and construct the load parameter vector; S3. Based on the embedding relationship between the end-load mass and the first moment of the centroid in the robot's minimum inertial parameter set, extract and reorganize the column vectors corresponding to the load parameters from the robot's minimum parameter regression matrix to construct the load observation matrix. S4. Obtain the joint state data and load condition joint torque data of the robot; obtain the unloaded predicted joint driving torque based on the unloaded reference torque model and the joint state data; obtain the additional joint torque caused by the robot end-effector load based on the difference between the load condition joint torque data and the unloaded predicted joint driving torque; establish a load parameter linear identification model based on the additional joint torque, the load observation matrix, and the load parameter vector. S5. Stack multiple sampling point data within a preset time window, and solve the predicted load parameter vector of the load parameter vector according to the linear identification model of the load parameter. S6. Based on the predicted load parameter vector, obtain the predicted load mass and the first moment of the predicted centroid at the robot end-effector load location, and complete the rapid online identification and update of the robot end-effector load parameters.

[0007] In one embodiment of the present invention, the step of pre-identifying the robot's rigid body dynamics basis parameters and joint friction parameters, and establishing an unloaded reference torque model in step S1 is as follows: Collect joint state data of the robot when it moves along a preset excitation trajectory in an unloaded state; wherein, the joint state data includes joint angular displacement, joint angular velocity, joint angular acceleration and joint driving torque data; Based on the joint state data and the robot's unloaded dynamics model, a minimum parameter regression matrix is ​​constructed, and the rigid body dynamics basis parameters of the robot are obtained using the weighted least squares method. The robot's joints are controlled to move forward and backward within a preset speed range. The robot's joint angular velocity and driving torque data are collected, and a nonlinear least squares fitting method is used to identify nonlinear joint friction parameters. Based on the rigid body dynamics parameters and the joint friction parameters, an unloaded reference moment model is established; wherein, the expression for the unloaded reference moment model is: ; in, This indicates the predicted joint driving torque under no-load conditions. This represents the pre-identified rigid body dynamics basis parameters. This represents the pre-identified joint friction parameters. Represents the minimum parameter regression matrix. Represents the joint angle vector. Represents the joint angular velocity vector. This represents the joint angular acceleration vector. This represents the frictional torque of the joint.

[0008] In one embodiment of the present invention, the expression for the robot's unloaded dynamics model is: ; in, Indicates the joint driving torque. Represents the minimum parameter regression matrix. Represents the basic parameters of the robot's rigid body dynamics. This represents the parameters of the friction model.

[0009] In one embodiment of the present invention, the joint friction torque includes the nonlinear friction force of each joint of the robot; the expression for the nonlinear friction force of each joint of the robot is: ; in, Indicates the first Nonlinear frictional forces at each joint Indicates the first The static friction coefficient of each joint Represents the Coulomb coefficient of friction. Indicates the coefficient of viscous friction. Indicates the Stribeck characteristic velocity. Indicates the curve shape parameters, Indicates the first angular velocity of each joint Indicates the first Low-speed compensation coefficient for each joint This indicates the speed scale at which the low-speed compensation term applies. This indicates the frictional torque bias term. Represents a symbolic function.

[0010] In one embodiment of the present invention, the expression of the load parameter linear identification model is: ; in, This indicates the additional joint torque at the end of the load. This represents the measured joint torque under load conditions. This indicates the predicted joint driving torque under no-load conditions. Represents the load observation matrix. This represents the load parameter vector.

[0011] In one embodiment of the present invention, the expression for the measured joint torque under the load condition is: ; in, This represents the measured joint torque under load conditions. This represents the pre-identified rigid body dynamics basis parameters. This represents the pre-identified joint friction parameters. Represents the minimum parameter regression matrix. Represents the joint angle vector. Represents the joint angular velocity vector. This represents the joint angular acceleration vector. This represents the frictional torque of the joint.

[0012] In one embodiment of the present invention, the expression for the load parameter vector is: ; in, Represents the load parameter vector. Indicates the quality of the end load; This indicates the position of the load's centroid in the end coordinate system. The symbol for transpose of a matrix.

[0013] In one embodiment of the present invention, in step S5, the weighted least squares method is used to solve for the predicted load parameter vector of the load parameter vector; wherein, the expression of the predicted load parameter vector is: ; in, Represents the predicted load parameter vector. This represents the load identification weight matrix. Indicates in The load parameter vector at each sampling point Indicates in The end load plus joint torque obtained at each sampling point The symbol for transpose of a matrix.

[0014] In one embodiment of the present invention, the expressions for the predicted load mass and the predicted first moment of the centroid are: ; ; in, Indicates the predicted load quality; , and This indicates the predicted position of the load centroid in the end coordinate system; , , , These represent the first, second, third, and fourth components of the predicted load parameter vector, respectively.

[0015] Secondly, to solve the above-mentioned technical problems, the present invention provides a rapid online identification system for end-effector load parameters of an industrial robot, used to implement the above-mentioned rapid online identification method for end-effector load parameters of an industrial robot, comprising: The no-load reference torque establishment module is used to pre-identify the robot's rigid body dynamics basis parameters and joint friction parameters, and establish a no-load reference torque model; The load parameter vector construction module is used to select the load mass and first moment of the centroid at the robot's end effector as online identification objects and construct the load parameter vector. The load observation matrix construction module is used to extract and reorganize the column vectors corresponding to the load parameters from the robot's minimum parameter regression matrix based on the embedding relationship between the end load mass and the first moment of the centroid in the robot's minimum inertial parameter set, and construct the load observation matrix. The identification model building module is used to acquire the joint state data and load-condition joint torque data of the robot; obtain the unloaded predicted joint driving torque based on the unloaded reference torque model and the joint state data; obtain the additional joint torque caused by the robot's end-effector load based on the difference between the load-condition joint torque data and the unloaded predicted joint driving torque; and establish a load parameter linear identification model based on the additional joint torque, the load observation matrix, and the load parameter vector. The solution module is used to stack multiple sampling point data within a preset time window and solve for the predicted load parameter vector of the load parameter vector according to the linear identification model of the load parameter. The identification module is used to obtain the predicted load mass and the first moment of the predicted centroid at the robot end-effector load location based on the predicted load parameter vector, thereby completing the rapid online identification and updating of the robot end-effector load parameters.

[0016] Compared with the prior art, the above-described technical solution of the present invention has the following advantages: This invention discloses a method and system for rapid online identification of end-effector load parameters in industrial robots, effectively solving problems such as reliance on dedicated excitation trajectories, high dimensionality of identification parameters, large online computational load, and difficulty in integrating with the robot's normal operating rhythm in existing industrial robot end-effector load identification technologies. Addressing dynamic and variable load operation scenarios such as grasping, handling, assembly, and quick tool changing in industrial robots, this invention utilizes short pauses or low-speed operation periods during normal robot operation. Based on the difference between the measured joint torque under load conditions and the no-load reference torque, a low-dimensional linear identification model of the end-effector load mass and the first-order moment of the centroid is constructed, and a weighted least squares method is used to quickly estimate the load mass and centroid position. This invention does not require the execution of additional dedicated excitation trajectories online, nor does it require complete identification of the end-effector load inertia tensor; it can rapidly update the end-effector load mass and centroid parameters online to meet the needs of dynamic and variable load industrial operations. This invention is applicable to six-DOF serial joint industrial robots and can also be extended to other multi-DOF joint robots with dynamic models and joint torque acquisition capabilities, demonstrating strong engineering applicability and facilitating real-time deployment. Attached Figure Description

[0017] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0018] Figure 1 This is a flowchart of a method for rapid online identification of end-effector load parameters of an industrial robot according to a preferred embodiment of the present invention; Figure 2 This is a flowchart illustrating a preferred embodiment of a method for rapid online identification of end-effector load parameters of an industrial robot. Figure 3 This is a structural diagram of a rapid online identification system for end-effector load parameters of an industrial robot, as shown in a preferred embodiment of the present invention. Detailed Implementation

[0019] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0020] Example 1: Reference Figure 1 As shown, this embodiment of the invention provides a method for rapid online identification of end-effector load parameters of an industrial robot, including but not limited to the following steps: S1. Pre-identify the robot's rigid body dynamics parameters and joint friction parameters, and establish an unloaded reference torque model; S2. Select the load mass and first moment of the centroid at the robot's end effector as the online identification objects, and construct the load parameter vector; S3. Based on the embedding relationship between the end-load mass and the first moment of the centroid in the robot's minimum inertial parameter set, extract and reorganize the column vectors corresponding to the load parameters from the robot's minimum parameter regression matrix to construct the load observation matrix. S4. Obtain the robot's joint state data and load condition joint torque data; based on the no-load reference torque model and joint state data, obtain the no-load predicted joint driving torque; based on the difference between the load condition joint torque data and the no-load predicted joint driving torque, obtain the additional joint torque caused by the robot's end effector load; based on the additional joint torque, load observation matrix, and load parameter vector, establish a linear identification model for load parameters. S5. Stack multiple sampling point data within a preset time window, and solve the predicted load parameter vector based on the load parameter linear identification model. S6. Based on the predicted load parameter vector, obtain the predicted load mass and the first moment of the predicted centroid at the robot end-effector load location, and complete the rapid online identification and update of the robot end-effector load parameters.

[0021] This invention discloses a rapid online identification method for end-effector load parameters of industrial robots. It separates the additional joint torque generated by the end-effector load by subtracting the joint torque under load from the predicted torque under no-load conditions. This effectively eliminates torque components caused by the robot's link dynamics and joint friction, and isolates the interference of body model parameter errors on the load identification results. This method selects the load mass and the first-order moment of the center of mass as the identification objects. Utilizing the embedding relationship of the end-effector load parameters in the robot's minimum inertia parameter set, it extracts and reconstructs the column vectors corresponding to the load parameters from the minimum parameter regression matrix to construct the load observation matrix. It establishes only a low-dimensional linear identification model for the load mass and the first-order moment of the center of mass, without needing to identify the complete load inertia tensor, effectively reducing the dimensionality of the parameters to be identified, decreasing the computational load of matrix solving, and improving the numerical stability of the regression equation. Simultaneously, this method uses joint state and torque sampling data during normal robot operation, utilizing naturally occurring pauses and low-speed periods within the workflow to complete data acquisition. It does not require the execution of additional dedicated excitation trajectories online, nor does it interrupt the production cycle, making it adaptable to dynamic variable load industrial scenarios such as grasping, handling, assembly, and tool change. By stacking observation data from multiple sampling points within a preset time window to construct a model and solve for the load parameter vector, the influence of torque measurement noise can be effectively suppressed, and the estimated values ​​of the load mass and the first moment of the centroid can be quickly obtained, enabling real-time identification and updating of the end-effector load parameters. Furthermore, this method can be implemented based on the robot's own joint torque feedback, without the need for additional force sensing equipment. It is not only applicable to six-DOF tandem joint industrial robots but can also be extended to other multi-DOF jointed robots with dynamic models and joint torque acquisition capabilities. The algorithm is easy to deploy and integrate within the robot controller, exhibiting strong engineering applicability and providing reliable load parameter support for robot torque feedforward compensation, collision detection, and motion control optimization under variable load conditions.

[0022] Specifically, this invention focuses on articulated industrial robots, particularly six-degree-of-freedom articulated industrial robots. The robot's end effector is rigidly connected to an external load via a flange, gripper, or tool end. When the robot grips a workpiece, moves an object, or changes tools, the mass and center-of-gravity position of the end-load change, causing variations in the joint driving torque. To quickly obtain the end-load parameters without executing additional dedicated excitation trajectories, this invention utilizes the difference between the robot's no-load reference torque and the measured torque under load to construct a low-dimensional linear identification model of the end-load mass and the first-order moment of its center of gravity.

[0023] Specifically, for The dynamic model of a jointed industrial robot with multiple degrees of freedom is expressed as follows: ; Since the robot's rigid body dynamics model is linear with respect to inertial parameters, it can be reorganized as follows: ; in, Represents the minimum parameter regression matrix. Represents the basic parameters of the robot's rigid body dynamics. This represents the joint friction torque. Indicates the parameters of the friction model. Represents the joint angle vector. Represents the joint angular velocity vector. This represents the joint angular acceleration vector. The rigid body dynamics basis parameters are obtained from the robot's original dynamic parameters after eliminating linear correlation terms and recombining parameters. Their specific expression can be predetermined according to the robot configuration, link dimensions, joint offsets, and installation posture.

[0024] Furthermore, under static or low-speed operating conditions, the joint friction torque has a significant impact on the joint driving torque. To improve the prediction accuracy of the no-load reference torque, this embodiment of the invention introduces an improved nonlinear Stribeck friction model into the no-load dynamics model. Specifically, the joint friction torque includes the nonlinear friction force of each joint of the robot; the expression for the nonlinear friction force of each joint of the robot is as follows: ; in, Indicates the first Nonlinear frictional forces at each joint Indicates the first The static friction coefficient of each joint Represents the Coulomb coefficient of friction. Indicates the coefficient of viscous friction. Indicates the Stribeck characteristic velocity. Indicates the curve shape parameters, Indicates the first Angular velocity of each joint; Indicates the first The low-speed compensation coefficient of each joint is used to compensate for the local friction residuals that the standard Stribeck model still does not fully fit in the low-speed uniform range. This is the speed scale for the low-speed compensation term, used to limit the main application of the compensation term to the low-speed region. It is set according to the actual range of the low-speed region. This is the friction torque bias term, used to compensate for unmodeled errors in current zero bias, driver bias, and other approximate constant values; Represents a symbolic function.

[0025] In this embodiment of the invention, before performing online load identification, the robot's unloaded rigid body dynamics parameters and friction parameters are identified offline.

[0026] Specifically, in step S1, the robot runs a pre-designed excitation trajectory in an unloaded state and collects joint angular displacement, joint angular velocity, joint angular acceleration, and joint driving torque data. A minimum parameter regression matrix is ​​constructed, and the robot's rigid body dynamics basis parameters are obtained using the weighted least squares method. Simultaneously, the robot's joints are controlled to move forward and backward within a preset speed range, collecting joint angular velocity and driving torque data. Nonlinear least squares fitting is used to identify the parameters of the nonlinear friction model, thereby establishing an unloaded reference torque model that includes nonlinear friction compensation. This unloaded reference torque model is used to predict the joint driving torque of the robot in an unloaded state. The specific steps are as follows: S110. Collect joint state data of the robot when it moves along a preset excitation trajectory under no-load conditions; wherein, the joint state data includes joint angular displacement, joint angular velocity, joint angular acceleration and joint driving torque data.

[0027] S120. Based on the joint state data and the robot's unloaded dynamics model, construct the minimum parameter regression matrix and obtain the robot's rigid body dynamics basis parameters using the weighted least squares method. The expression for the robot's unloaded dynamics model is: ; in, Indicates the joint driving torque. Represents the minimum parameter regression matrix. Represents the basic parameters of the robot's rigid body dynamics. Indicates the parameters of the friction model. This represents the frictional torque of the joint.

[0028] S130. Control each joint of the robot to move forward and backward within a preset speed, collect the joint angular velocity and driving torque data of the robot, and use a nonlinear least squares fitting method to identify nonlinear joint friction parameters.

[0029] S140. Based on the rigid body dynamics basis parameters and joint friction parameters, establish the no-load reference moment model; wherein, the expression of the no-load reference moment model is: ; in, This indicates the predicted joint driving torque under no-load conditions. This represents the pre-identified rigid body dynamics basis parameters. This represents the pre-identified joint friction parameters. This represents the minimum parameter regression matrix.

[0030] Specifically, in industrial applications, the end effector load is typically rigidly connected to the robotic arm via a flange or tool end. Therefore, the end effector load can be considered as an integrated additional part rigidly coupled to the end effector link, and its mass and first-order moment of the center of mass are incorporated into the robot's overall minimum inertia parameter expression. Thus, in step S2, the end effector load is considered as an additional part rigidly coupled to the robot's end effector link, and only the load mass and the first-order moment of the center of mass are selected as online identification objects to construct the load parameter vector. Here, let the end effector load mass be... The position of the load centroid in the end coordinate system is The expression for defining the load parameter vector is: ; in, Represents the load parameter vector. Indicates the quality of the end load; This indicates the position of the load's center of mass in the end coordinate system, i.e., the first moment of the center of mass; The symbol for transpose of a matrix.

[0031] Specifically, in step S3, based on the embedding relationship between the end-load mass and the first moment of the centroid in the robot's minimum inertial parameter set, column vectors related to the load parameters are extracted and reorganized from the robot's minimum parameter regression matrix to construct the load observation matrix. .

[0032] For example, the load observation matrix can be obtained by recombining the column vectors related to the end-load mass and the first moment of the centroid in the robot's minimum parameter regression matrix. In a specific embodiment of a six-degree-of-freedom serial articulated industrial robot, let: ; in, The first parameter of the minimum parameter regression matrix represents the first parameter of the minimum parameter regression matrix. If the column is defined, then the load observation matrix can be expressed as: ; in, Depend on The column vectors related to the load mass are obtained by linear combination of the robot's structural parameters. , , respectively by It is obtained by combining column vectors related to the first moment of the load centroid.

[0033] For example, in a specific structure, the observation column vector corresponding to the load quality is represented as: ; in, , These represent the lengths of the robot's second and third links, respectively. , , These represent the offsets of the robot's 4th, 5th, and 6th joints, respectively.

[0034] The observation column vector corresponding to the first moment of the load centroid can be expressed as: .

[0035] It should be noted that the specific column combination relationship of the above-mentioned load observation matrix is ​​only one specific implementation method, and its specific expression is related to the robot configuration, the selection method of the minimum inertial parameter set, and the column order of the dynamic regression matrix. For other configurations of industrial robots, it can be re-determined based on the corresponding minimum parameter regression matrix and end-load embedding relationship. .

[0036] In actual industrial operations, robots typically experience brief periods of stillness or low-speed operation during tasks such as grasping, placement, tool switching, or path transitions. This invention utilizes these short time windows to collect a small amount of joint state and torque data without interrupting the normal workflow or requiring additional dedicated excitation trajectories. Therefore, this method can be embedded within the robot's normal operating cycle to update load parameters. Within this short time window, robot joint velocities and accelerations are relatively low, and the influence of the load's complete inertia tensor on the additional joint torques is relatively weak. The additional joint torques are mainly caused by the load mass and the first-order moment of the center of mass. Simultaneously, the robot's body dynamics and joint friction terms are compensated by the no-load reference torque model. Therefore, a low-dimensional load identification model can be constructed based on the difference between the measured joint torques under load conditions and the no-load reference torque.

[0037] Specifically, in step S4, online identification data is collected using a short window of actual operation, and a linear identification model for load parameters is established. (Refer to...) Figure 2 As shown, the system first determines whether the robot has entered a stationary or low-speed short-term working zone. If it has, the system utilizes the stationary or low-speed short-term window formed during the completion of grasping, before placement, after tool switching, or path transition phases to collect joint status data and measured joint torque data under load conditions. This data acquisition process does not require interruption of the normal workflow or additional planning of dedicated excitation trajectories. If the robot has not entered the aforementioned zone, it continues to perform grasping, handling, and tool switching tasks until a valid identification window is detected. The determination method involves real-time monitoring of the angular velocity of each joint. When the angular velocity of all joints remains below a preset angular velocity threshold and maintains the shortest sampling duration, it is determined that the robot has entered a valid identification window.

[0038] Furthermore, based on the difference between the measured joint torque under load conditions and the no-load reference torque, the additional joint torque caused by the end load is obtained, and a linear identification model of the load parameters is established. The expression for the linear identification model of the load parameters is: ; in, This indicates the additional joint torque at the end of the load. This represents the measured joint torque under load conditions. This indicates the predicted joint driving torque under no-load conditions. Represents the load observation matrix. This represents the load parameter vector.

[0039] Furthermore, in one embodiment of a six-degree-of-freedom serial articulated industrial robot, when the end effector load is rigidly connected to the sixth link, the relevant parameters of the sixth link can be determined from the parameters under no-load conditions. Updated to: ; in, These represent the mass of the robot's sixth joint, the centroid of the sixth joint along the X-axis, the centroid of the sixth joint along the Y-axis, and the centroid of the sixth joint along the Z-axis, respectively.

[0040] Therefore, the expression for the measured joint torque under load conditions is: ; in, This represents the measured joint torque (i.e., joint driving torque) under load conditions. This represents the pre-identified rigid body dynamics basis parameters. This represents the pre-identified joint friction parameters. This represents the minimum parameter regression matrix.

[0041] Specifically, in steps S5 and S6, multiple sampling point data are stacked within a short-term preset time window, and the load parameter vector is solved using the weighted least squares method. This further calculates the end-load mass and centroid position, enabling rapid online updates of the end-load parameters. The short-term preset time window corresponds to the following working conditions: low-speed uniform motion, low-speed reversing, and stationary waiting. In this embodiment, the low speed is determined based on the robot's joint angular velocity; as a preferred embodiment, a threshold value of ≤10° / s for each joint angular velocity can be set. It should be understood that this angular velocity threshold is merely an example, and those skilled in the art can select other suitable threshold values ​​based on the robot model and the actual working conditions.

[0042] It should be noted that this short-term preset time window can be manually set, and its duration can be determined based on actual working conditions or engineering experience. It only requires that the robot continuously operates at a low speed within this time window during its operation. Furthermore, the working conditions that form this time window are not limited to those naturally occurring during robot operation; they can also be actively configured. For example, after the robot picks up an object, to complete the load mass and center of gravity identification, a low-speed, uniform motion phase can be manually planned before the formal execution of the work path. This low-speed, uniform motion phase is part of the overall work path, thus constructing a short-term low-speed time window that meets the requirements.

[0043] The method described in this embodiment of the invention transforms the end-load identification problem into a linear solution problem of four-dimensional load parameters through the above steps. It can quickly update the load parameters within a short window of stillness or low speed during normal operation of the industrial robot, without the need for online identification of the complete load inertia tensor or the execution of additional dedicated excitation trajectories.

[0044] Furthermore, in actual online identification processes, data can be collected within a short time window when the robot is stationary or moving at low speed. Each sampling point has a specified number of sampling points. The load-added torque and load observation matrix corresponding to each sampling point are stacked to obtain... The expression for the linear identification model of load parameters at each sampling point is: ; in, express The end load plus joint torque obtained at each sampling point Indicates in The load parameter vector at each sampling point.

[0045] Further, in step S5, the weighted least squares method is used to solve for the predicted load parameter vector from the load parameter vector; wherein, the expression for the predicted load parameter vector is: ; in, Represents the predicted load parameter vector; This represents the load identification weight matrix, which is used to assign different weights to different sampling points based on the quality of the sampled data. Indicates in The end load and additional joint torque are obtained from each sampling point.

[0046] Furthermore, the expression for the identified load parameter vector is: ; in, , , , Let these represent the first, second, third, and fourth components of the predicted load parameter vector, respectively. The end load mass and centroid location can then be obtained. Therefore, the expressions for the predicted load mass and the first moment of the predicted centroid are: ; ; in, Indicates the predicted load quality; , and This indicates the predicted position of the load's centroid in the end-effector coordinate system. Based on the predicted load mass and the predicted first moment of the centroid, the robot's end-effector load parameters are rapidly identified and updated online.

[0047] The method described in this invention utilizes naturally occurring or artificially pre-set static or low-speed short windows during actual operations such as robot grasping completion, placement, path switching, or tool replacement to collect joint state and joint torque data. By pre-establishing an unloaded reference torque model that includes nonlinear friction compensation, the influence of low-speed friction nonlinearity on torque difference is reduced. Furthermore, the end-effector load mass and the first-order moment of the center of mass are embedded into the robot's minimum inertia parameter expression to construct a four-dimensional linear identification model of load parameters. This enables rapid online estimation of the end-effector load mass and center of mass position without the need for online recovery of the complete load inertia tensor or additional execution of a dedicated excitation trajectory.

[0048] Example 2: Based on the same inventive concept, this embodiment provides a rapid online identification system for end-effector load parameters of industrial robots. The principle of solving the problem is similar to that of the rapid online identification method for end-effector load parameters of industrial robots provided in Embodiment 1, and the repetitions will not be repeated.

[0049] Reference Figure 3 As shown, this embodiment provides a rapid online identification system for end-effector load parameters of an industrial robot, used to implement the rapid online identification method for end-effector load parameters of an industrial robot described in Embodiment 1, including: The no-load reference torque establishment module is used to pre-identify the robot's rigid body dynamics basis parameters and joint friction parameters, and establish a no-load reference torque model. The load parameter vector construction module is used to select the load mass and first moment of the centroid at the robot's end effector as online identification objects and construct the load parameter vector. The load observation matrix construction module is used to extract and reorganize the column vectors corresponding to the load parameters from the robot's minimum parameter regression matrix based on the embedding relationship between the end-load mass and the first moment of the centroid in the robot's minimum inertial parameter set, and construct the load observation matrix. The identification model building module is used to acquire the robot's joint state data and load-condition joint torque data; based on the no-load reference torque model and joint state data, the no-load predicted joint driving torque is obtained; based on the difference between the load-condition joint torque data and the no-load predicted joint driving torque, the additional joint torque caused by the robot's end effector load is obtained; based on the additional joint torque, load observation matrix, and load parameter vector, a linear identification model of load parameters is established. The solution module is used to stack multiple sampling point data within a preset time window and solve for the predicted load parameter vector based on the load parameter linear identification model. The identification module is used to obtain the predicted load mass and the first moment of the predicted centroid at the robot's end effector load based on the predicted load parameter vector, and to complete the rapid online identification and updating of the robot's end effector load parameters.

[0050] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0051] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0052] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0053] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0054] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for rapid online identification of end-effector load parameters of an industrial robot, characterized in that, include: S1. Pre-identify the robot's rigid body dynamics parameters and joint friction parameters, and establish an unloaded reference torque model; S2. Select the load mass and first moment of the centroid at the robot's end effector as the online identification objects, and construct the load parameter vector; S3. Based on the embedding relationship between the end-load mass and the first moment of the centroid in the robot's minimum inertial parameter set, extract and reorganize the column vectors corresponding to the load parameters from the robot's minimum parameter regression matrix to construct the load observation matrix. S4. Obtain the joint state data and load condition joint torque data of the robot; Based on the no-load reference torque model and the joint state data, the no-load predicted joint driving torque is obtained; based on the difference between the load condition joint torque data and the no-load predicted joint driving torque, the additional joint torque caused by the robot end-effector load is obtained; based on the additional joint torque, the load observation matrix, and the load parameter vector, a load parameter linear identification model is established. S5. Stack multiple sampling point data within a preset time window, and solve the predicted load parameter vector of the load parameter vector according to the linear identification model of the load parameter. S6. Based on the predicted load parameter vector, obtain the predicted load mass and the first moment of the predicted centroid at the robot end-effector load location, and complete the rapid online identification and update of the robot end-effector load parameters.

2. The method for rapid online identification of end-effector load parameters of an industrial robot according to claim 1, characterized in that, In step S1, the steps of pre-identifying the robot's rigid body dynamics parameters and joint friction parameters, and establishing the no-load reference torque model are as follows: Collect joint state data of the robot when it moves along a preset excitation trajectory in an unloaded state; wherein, the joint state data includes joint angular displacement, joint angular velocity, joint angular acceleration and joint driving torque data; Based on the joint state data and the robot's unloaded dynamics model, a minimum parameter regression matrix is ​​constructed, and the rigid body dynamics basis parameters of the robot are obtained using the weighted least squares method. The robot's joints are controlled to move forward and backward within a preset speed range. The robot's joint angular velocity and driving torque data are collected, and a nonlinear least squares fitting method is used to identify nonlinear joint friction parameters. Based on the rigid body dynamics parameters and the joint friction parameters, an unloaded reference moment model is established; wherein, the expression for the unloaded reference moment model is: ； in, This indicates the predicted joint driving torque under no-load conditions. This represents the pre-identified rigid body dynamics basis parameters. This represents the pre-identified joint friction parameters. Represents the minimum parameter regression matrix. Represents the joint angle vector. Represents the joint angular velocity vector. This represents the joint angular acceleration vector. This represents the frictional torque of the joint.

3. The method for rapid online identification of end-effector load parameters of an industrial robot according to claim 2, characterized in that, The expression for the robot's unloaded dynamics model is as follows: ； in, Indicates the joint driving torque. Represents the minimum parameter regression matrix. Represents the basic parameters of the robot's rigid body dynamics. This represents the parameters of the friction model.

4. The method for rapid online identification of end-effector load parameters of an industrial robot according to claim 2, characterized in that, The joint friction torque includes the nonlinear friction force of each joint of the robot; the expression for the nonlinear friction force of each joint of the robot is: ； in, Indicates the first Nonlinear frictional forces at each joint Indicates the first The static friction coefficient of each joint Represents the Coulomb coefficient of friction. Indicates the coefficient of viscous friction. Indicates the Stribeck characteristic velocity. Indicates the curve shape parameters, Indicates the first angular velocity of each joint Indicates the first Low-speed compensation coefficient for each joint This indicates the speed scale at which the low-speed compensation term applies. This indicates the frictional torque bias term. Represents a symbolic function.

5. The method for rapid online identification of end-effector load parameters of an industrial robot according to claim 1, characterized in that, The expression for the linear identification model of the load parameters is: ； in, This indicates the additional joint torque at the end of the load. This represents the measured joint torque under load conditions. This indicates the predicted joint driving torque under no-load conditions. Represents the load observation matrix. This represents the load parameter vector.

6. The method for rapid online identification of end-effector load parameters of an industrial robot according to claim 5, characterized in that, The expression for the measured joint torque under the load condition is: ； in, This represents the measured joint torque under load conditions. This represents the pre-identified rigid body dynamics basis parameters. This represents the pre-identified joint friction parameters. Represents the minimum parameter regression matrix. Represents the joint angle vector. Represents the joint angular velocity vector. This represents the joint angular acceleration vector. This represents the frictional torque of the joint.

7. The method for rapid online identification of end-effector load parameters of an industrial robot according to claim 1, characterized in that, The expression for the load parameter vector is: ； in, Represents the load parameter vector. Indicates the quality of the end load; This indicates the position of the load's centroid in the end coordinate system. The symbol for transpose of a matrix.

8. The method for rapid online identification of end-effector load parameters of an industrial robot according to claim 1, characterized in that, In step S5, the weighted least squares method is used to solve for the predicted load parameter vector from the load parameter vector; wherein, the expression for the predicted load parameter vector is: ； in, Represents the predicted load parameter vector. This represents the load identification weight matrix. Indicates in The load parameter vector at each sampling point Indicates in The end load plus joint torque obtained at each sampling point The symbol for transpose of a matrix.

9. The method for rapid online identification of end-effector load parameters of an industrial robot according to claim 1, characterized in that, The expressions for the predicted load mass and the predicted first moment of the centroid are: ； ； in, Indicates the predicted load quality; , and This indicates the predicted position of the load centroid in the end coordinate system; , , , These represent the first, second, third, and fourth components of the predicted load parameter vector, respectively.

10. A rapid online identification system for end-effector load parameters of an industrial robot, used to implement the rapid online identification method for end-effector load parameters of an industrial robot as described in any one of claims 1 to 9, characterized in that, include: The no-load reference torque establishment module is used to pre-identify the robot's rigid body dynamics basis parameters and joint friction parameters, and establish a no-load reference torque model; The load parameter vector construction module is used to select the load mass and first moment of the centroid at the robot's end effector as online identification objects and construct the load parameter vector. The load observation matrix construction module is used to extract and reorganize the column vectors corresponding to the load parameters from the robot's minimum parameter regression matrix based on the embedding relationship between the end load mass and the first moment of the centroid in the robot's minimum inertial parameter set, and construct the load observation matrix. The identification model building module is used to acquire the joint state data and joint torque data under load conditions of the robot; Based on the no-load reference torque model and the joint state data, the no-load predicted joint driving torque is obtained; Based on the difference between the joint torque data under load and the predicted joint driving torque under no-load conditions, the additional joint torque caused by the robot end-effector load is obtained; based on the additional joint torque, the load observation matrix and the load parameter vector, a linear identification model of the load parameters is established. The solution module is used to stack multiple sampling point data within a preset time window and solve for the predicted load parameter vector of the load parameter vector according to the linear identification model of the load parameter. The identification module is used to obtain the predicted load mass and the first moment of the predicted centroid at the robot end-effector load location based on the predicted load parameter vector, thereby completing the rapid online identification and updating of the robot end-effector load parameters.

Images