A method for identifying dynamics parameters of a robot arm

Abstract

The application relates to the technical field of mechanical arm control, and particularly provides a mechanical arm dynamics parameter identification method, which identifies friction parameters of a segmented nonlinear friction model according to initial values of joint friction torques, calculates the calculated values of the joint friction torques, calculates the calculated values of total inertia torques according to the calculated values of the joint friction torques, identifies inertia parameters of a semi-linearized dynamics model according to the calculated values of the total inertia torques, updates the semi-linearized dynamics model according to the identified inertia parameters and recalculates updated values of the total inertia torques, calculates updated values of the joint friction torques according to the updated values of the total inertia torques, replaces the initial values of the joint friction torques with the updated values of the joint friction torques to re-identify the friction parameters, and iterates until an error converges to a predetermined threshold. The identification method adopts the semi-linearized dynamics model and compensates the model, the identified parameters are closer to the actual situation, and the reliability and accuracy of the dynamics model are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mechanical arm control, and particularly provides a method for identifying dynamic parameters of a mechanical arm. BACKGROUND

[0002] The dynamic model of a mechanical arm can be used in the fields of advanced control algorithm design, collision detection, human-machine interaction, multi-arm cooperation, etc. A high-precision dynamic model can improve the control performance of the mechanical arm. However, the dynamic model is usually not provided by the manufacturer of the mechanical arm, and the dynamic parameters obtained from computer-aided design (CAD) software often show significant differences compared with the actual model, and experimental identification is a reliable method to obtain accurate dynamic parameters.

[0003] The experimental dynamic identification method can be generally divided into a model-free identification method and a model-based identification method. The model-free dynamic identification method does not require a predefined mathematical model of the mechanical arm, and this method relies on the measurement of the joint position and current of the mechanical arm. Typical model-free dynamic identification methods include a multilayer perceptron network compensator, a Gaussian process regression learning, a radial basis function network compensator, and an adaptive neural model optimization. The advantage of this method is more flexible, and compared with the model-based method, this method requires more computing resources, has lower precision, and the identified dynamic parameters can not cover all physical effects, and it is very difficult to accurately estimate the inertia parameters, and is not suitable for observer design.

[0004] The model-based dynamic identification usually includes dynamic modeling, excitation trajectory design, parameter identification, and model verification. The Newton-Euler method is a commonly used method for establishing a dynamic model of a mechanical arm, and using a tool box such as SYMORO or OpenSYMORO can quickly realize dynamic linearization. However, the identification accuracy based on the rigid body dynamic model is limited, and the identification error is large, especially when the joint speed is very low, the dynamic model identification error is large.

[0005] Accurate estimation of dynamic parameters is a key step in identifying the dynamics of a mechanical arm. Several commonly used dynamic parameter identification methods include a least squares algorithm, a weighted least squares algorithm, a maximum likelihood estimation, and a Kalman filter. Although these methods can quickly identify dynamic parameters, there are still the following problems:

[0006] 1. These algorithms are easily affected by noise and outliers. The collected data usually has a lot of noise, and these noises usually cause a lot of errors in the data, which affect the identification accuracy.

[0007] 2. The identified dynamic parameters do not satisfy the physical feasibility constraints, and the identified parameters can exceed the normal value, which will cause the identified model to have high numerical accuracy but not meet the actual physical conditions.

[0008] 3. Physical feasibility constraints can reduce the accuracy of dynamic identification, and may cause the identified dynamic parameters to be on the edge of the physical feasibility constraints.

[0009] 4. The limitation of the speed bandwidth causes the identified trajectory of the robot arm to be unable to achieve high-speed operation, resulting in the excitation trajectory being unable to be designed to be large, thereby causing the identification effect to be poor.

[0010] 5. These algorithms cannot effectively combine the nonlinear friction model for dynamic parameter identification;

[0011] The existing parameter identification method linearizes the entire model completely, and this method is not suitable for nonlinear friction models. SUMMARY

[0012] To solve the above problems, the application provides a robot arm dynamic parameter identification method, the identification method adopted by the application is a semi-linearized dynamic model, which integrates the nonlinearity of the friction model, so that the identified parameters are more accurate and more consistent with the actual physical situation.

[0013] The robot arm dynamic parameter identification method provided by the application comprises:

[0014] S1: According to the direction of the joint speed of the robot arm, a segmented joint nonlinear friction model of the robot arm is established:

[0015] ;

[0016] Wherein, represents any joint of the robot arm, represents the total number of joints of the robot arm, represents the viscous friction when the joint speed is positive, represents the viscous friction when the joint speed is negative, represents the Coulomb friction when the joint speed is positive, represents the Coulomb friction when the joint speed is negative, represents the bias term of the joint friction when the joint speed is positive, represents the bias term of the joint friction when the joint speed is negative, represents the joint speed of the first joint of the robot arm, represents the nonlinear friction parameter of the joint when the joint speed is positive, represents the nonlinear friction parameter of the joint when the joint speed is negative, represents the friction of the joint, represents the joint friction torque of the robot arm, ​​​​indicates a symbol function;

[0017] S2: A semi-linearized dynamics model of the manipulator is established by an iterative semi-linearized dynamics identification algorithm combined with the piecewise joint nonlinear friction model:

[0018] ;

[0019] wherein, indicates a total joint torque of the manipulator, indicates an inertia total torque of the manipulator, indicates a subset of the regression matrix of the frictionless term, indicates a dynamics minimum parameter set vector of the frictionless parameter, indicates a nonlinear, indicates a linear, indicates an inertia torque of the manipulator, indicates a Coriolis force torque of the manipulator, indicates a gravity torque of the manipulator;

[0020] S3: Each joint of the manipulator is independently driven according to a predetermined trajectory, and an initial value of the joint friction torque of each joint is calculated;

[0021] S4: According to the initial value of the joint friction torque, the following iteration is performed:

[0022] The initial value of the joint friction torque is substituted into a friction parameter identification equation, and the friction parameter of the piecewise joint nonlinear friction model is identified by using an interior point method, and the friction parameter identification equation is:

[0023] ;

[0024] wherein, indicates the initial value of the joint friction torque

[0025] According to the identified friction parameter, the piecewise nonlinear friction model is updated, and the calculated value of the joint friction torque is calculated according to the updated piecewise nonlinear friction model;

[0026] According to the following torque relationship and the calculated value of the joint friction torque, the calculated value of the inertia total torque of the manipulator is calculated:

[0027] ;

[0028] wherein, indicates a total joint torque of the manipulator;

[0029] According to the calculated value of the inertia total torque, the inertia parameter of the semi-linearized dynamics model is identified by using a weighted least square algorithm, and the inertia parameter identification equation is:

[0030] ;

[0031] wherein, represents a target function,

[0032] updating the semi-linearized dynamics model according to the inertia parameter;

[0033] determining whether the root mean square error of the mechanical arm torque and the update amount of the friction parameter are all converged into a predetermined threshold:

[0034] if yes, ending the identification process;

[0035] if no, recalculating the updated value of the total inertia torque of the mechanical arm according to the updated semi-linearized dynamics model, and calculating the updated value of the joint friction torque again according to the updated value of the total inertia torque through the torque relationship, replacing the initial value of the joint friction torque in step S4 with the updated value of the joint friction torque, and repeating step S4.

[0036] Preferably, the expression of the predetermined trajectory is as follows:

[0037] ;

[0038] wherein, represents a joint motion trajectory period, represents any time within 0 to time, represents the time required for the joint speed to reverse, represents the joint speed of the joint at time, represents the expected joint speed of the joint, represents the expected joint speed of the joint, represents a motion trajectory parameter.

[0039] Preferably, the calculation method of the initial value of the joint friction torque is as follows:

[0040] ;

[0041] ;

[0042] wherein, , represents the friction torque calculated according to the predetermined trajectory, represents the joint torque of the mechanical arm when the joint position is , the joint speed is , and the joint acceleration is 0, represents the joint torque of the mechanical arm when the joint position is , the joint speed is , and the joint acceleration is 0.

[0043] Preferably, S5 is further included: compensating the model error of the semi-linearized dynamic model by using a deep convolutional neural network.

[0044] Preferably, the loss function of the deep convolutional neural network is:

[0045] ;

[0046] wherein, represents the model error of the semi-linearized dynamic model derived according to the identified semi-linearized dynamic model, represents the output generated by the convolutional neural network, represents the total number of training samples.

[0047] Preferably, the compensated semi-linearized dynamic model is:

[0048] .

[0049] wherein, represents the model error of the semi-linearized dynamic model compensated by the neural network, represents the neural network.

[0050] Compared with the prior art, the present application can achieve the following beneficial effects:

[0051] The dynamic model established by the identification method proposed in the present application is a semi-linearized dynamic model, which combines a piecewise nonlinear friction model. The piecewise linear friction model considers the speed direction of the robot arm, effectively improves the identification accuracy of the dynamic parameters, and further improves the accuracy of the dynamic model. The dynamic parameters identified by the dynamic parameter identification method proposed in the present application can satisfy the physical feasibility constraint, so that the identified dynamic parameter value is more reasonable and more consistent with the actual situation. At the same time, the present application also compensates the uncertainty of the dynamic model by using a deep learning convolutional neural network, which can effectively handle the parameter identification error of the dynamic model, improve the accuracy of the dynamic model of the robot arm under low speed motion, and further improve the reliability and accuracy of the dynamic model. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 is a dynamic parameter identification flowchart provided according to an embodiment of the present application;

[0053] Figure 2 is an algorithm flowchart of the dynamic parameter identification method according to an embodiment of the present application;

[0054] Figure 3 is a whole structure diagram of the compensation convolutional neural network according to an embodiment of the present application. DETAILED DESCRIPTION

[0055] In the following description, embodiments of the invention will be described with reference to the accompanying drawings. In the description below, the same modules are denoted by the same reference numerals. Where the same reference numerals are used, their names and functions are also the same. Therefore, their detailed description will not be repeated.

[0056] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not constitute a limitation thereof.

[0057] like Figure 1 As shown, this embodiment of the invention provides a method for identifying the dynamic parameters of a robotic arm. The identified dynamic parameters include friction parameters and inertial parameters. The identification method specifically includes:

[0058] S1: To make the friction model more closely reflect the actual motion of the robotic arm, this embodiment of the invention retains the nonlinearity of the friction model. Based on the direction of the joint velocity of the robotic arm, a piecewise nonlinear friction model is established, as shown in the following expression:

[0059] ;

[0060] in, This represents any joint of the robotic arm. This indicates the total number of joints in the robotic arm. This represents the viscous friction force when the joint velocity is positive. This represents the viscous friction force when the joint velocity is negative. This represents the Coulomb friction force when the joint velocity is positive. This represents the Coulomb friction force when the joint velocity is negative. This term represents the bias term for joint friction when the joint velocity is positive. This term represents the bias term for joint friction when the joint velocity is negative. Indicates the first robotic arm Joint velocity of each joint When the joint velocity is positive, the joint... The nonlinear friction parameters, This indicates that when the joint velocity is negative, the joint... The nonlinear friction parameters, and The value is usually in the range of 0 to 1. Indicates joint friction, This represents the joint friction torque of the robotic arm. Represents a symbolic function, and have:

[0061] .

[0062] The segmented nonlinear friction model contains 8 unknown friction parameters, respectively 、 、 、 、 、 、 and .

[0063] S2: In order to realize the semi-linearization of the dynamic model, the nonlinear fusion of the segmented joint nonlinear friction model into the dynamic model is used, the iterative semi-linearization dynamic identification algorithm is used to linearize the dynamic model of the manipulator, and the segmented nonlinear joint friction model which retains the nonlinearity is combined, and the semi-linearized dynamic model is established as follows:

[0064] ;

[0065] Wherein, represents the total joint torque of the manipulator, represents the total inertia torque of the manipulator, represents the joint friction torque of the manipulator, represents the subset of the regression matrix without friction, represents the dynamic minimum parameter set vector without friction parameters, represents the nonlinearity, represents the linearity, represents the inertia torque of the manipulator, represents the Coriolis force torque of the manipulator, represents the gravity torque of the manipulator, the total inertia torque of the manipulator includes the Coriolis force torque of the manipulator , the inertia torque and the gravity torque .

[0066] S3: Before identifying the friction parameters of the segmented joint nonlinear friction model, the initial joint friction torque data needs to be calculated, which is called the initial value of the joint friction torque in the embodiment of the application.

[0067] According to the predetermined motion trajectory, the motion of each joint of the manipulator is driven, and the motion trajectory used in the embodiment of the application is as follows:

[0068] ;

[0069] Wherein, represents the joint motion trajectory period, represents 0 to any moment in time, denotes the time required for the joint velocity to reverse, denotes the joint velocity at the initial moment, denotes the joint velocity at the final moment, denotes the joint velocity at the moment, denotes the desired joint velocity of the joint, denotes the motion trajectory parameters, for ensuring the continuity of the trajectory, the trajectory parameters satisfy the following constraints:

[0070] ;

[0071] According to the above constraints, the values of the motion trajectory parameters can be calculated.

[0072] In order to ensure the continuity of the motion speed of the robot arm during the motion process, the joint velocity at the initial moment and the joint velocity at the end moment are both set to 0, and the initial value of the joint friction torque is calculated through the single-joint motion of the robot arm, and the calculation method is as follows:

[0073] ;

[0074] wherein, , denotes the friction torque calculated according to the above motion trajectory, denotes the joint torque of the robot arm when the joint position is , the joint velocity is , and the joint acceleration is 0, denotes the joint torque of the robot arm when the joint position is , the joint velocity is , and the joint acceleration is 0.

[0075] Through the above motion trajectory, the joint friction torque of each joint at different speeds is calculated, the set of different speeds of each joint is called a speed group, and the number of speeds in the speed group is set to , and is used to represent the calculated initial value of the joint friction torque, then the initial value of the joint friction torque is: According to the initial value of the joint friction torque, the identification of the friction parameter in the dynamic parameter is started.

[0076] S4: In order to ensure that the identified dynamic parameters can meet the precision requirements, the identification process of the dynamic parameters needs to be iterated, and the specific iteration process is as follows:

[0077] According to the initial value of the joint friction torque, the interior point method is used to identify the friction parameter, and when optimizing the segmented joint nonlinear friction model, the physical feasibility constraint is considered, and the friction parameter identification equation is as follows:

[0078] ;

[0079] ;

[0080] Based on the identified friction parameters, the friction parameters in the original piecewise joint nonlinear friction model are updated with the identified friction parameters to obtain the updated piecewise joint nonlinear friction model. The corresponding new joint friction torque is then calculated using the updated piecewise joint nonlinear friction model. In this embodiment of the invention, this joint friction torque is referred to as the calculated value of the joint friction torque to distinguish it from the initial value of the joint friction torque. Both the calculated value and the initial value of the joint friction torque are joint friction torques of the robotic arm, but they are different and have different sources. The initial value of the joint friction torque is the joint friction torque calculated based on the single joint motion in step S3, while the calculated value of the joint friction torque is obtained by calculating it using the updated piecewise nonlinear friction model after identifying the friction parameters.

[0081] Based on the calculated value of the joint friction torque, the calculated value of the total inertial torque of the robotic arm can be obtained. The joint friction torque and the total inertial torque of the robotic arm satisfy the following torque relationship:

[0082] ;

[0083] in, This represents the total torque of the robotic arm. The value of the total torque is output by the software tool and is a known, fixed value. Therefore, the inertial total torque is... The total inertial torque of the robotic arm calculated here is referred to as the calculated value of the total inertial torque in this embodiment of the invention.

[0084] After obtaining the calculated value of the total inertial moment, the inertial parameters of the semi-linear dynamic model are identified based on the calculated value of the total inertial moment using the weighted least squares algorithm. The process of establishing the inertial parameter identification equation is as follows:

[0085] First, establish the initial inertial parameter identification equations:

[0086] ;

[0087] ;

[0088] in, The inertial parameters of the robotic arm are those identified from the basic parameter set using the weighted least squares algorithm. Friction parameters are not included in these inertial parameters. Describe the objective function. This represents a physical feasibility constraint. Indicates the first The moment of inertia of the motor rotor at each joint Indicates the first a joint quality of the joint, a moment of inertia of the joint, a joint center position vector of the joint, a trace of a trace of a trace of , the trace being a mathematical operation of a matrix and being a prior art in mathematical calculation, a unit matrix.

[0089] In order to prevent the identified inertia parameter from reaching the boundary of the physical feasibility set, the embodiments of the present application modify the objective function and introduce a reference dynamic parameter, which is directly obtained according to existing software and includes a reference inertia parameter and a reference friction parameter, both of which are known fixed values. Here, only the reference inertia parameter in the reference dynamic parameter is introduced, which can serve as a guide to make the identified inertia parameter closer to the actual value. After modifying the objective function and introducing the reference inertia parameter, the initial inertia parameter identification equation is changed to:

[0090] ;

[0091] ;

[0092] wherein, represents the reference inertia parameter, represents the 2-norm Euclidean distance metric.

[0093] After changing the inertia parameter identification equation, the embodiments of the present application combine an iterative semi-linearization identification algorithm to re-identify the inertia parameter using the geometric Riemann metric method, further ensuring that the identified inertia parameter is far away from the edge of the physical feasibility constraint set, so that the identified inertia parameter is more consistent with the actual situation. The final inertia parameter identification equation is:

[0094] ;

[0095] wherein, represents the geometric metric, which is the reference inertia parameter of the joint, represents the iterative semi-linearization identification algorithm, which can effectively integrate the geometric metric to make the identification result more consistent with the actual physical system.

[0096] The inertia parameter identified according to the above inertia parameter identification equation is used to update the semi-linearization dynamic model. Whether the updated semi-linearization dynamic model converges to meet the accuracy requirement needs to be judged by a threshold value, and the setting of the threshold value needs to be calculated theoretically according to the accuracy requirement and the characteristics of the dynamic model to obtain a reasonable threshold range.

[0097] The root mean square error of the mechanical arm torque and the update amount of the friction parameter are calculated, the root mean square error of the mechanical arm torque is the error square sum between the actual torque of the mechanical arm and the torque predicted by the semi-linearized dynamics model, and then the average value is taken and the square root is taken, and the update amount of the friction parameter is the two norm of the deviation between the friction parameter identified in the last iteration and the friction parameter obtained in this iteration of identification. A threshold value is preset to determine whether the root mean square error of the mechanical arm torque and the update amount of the friction parameter converge to the threshold value range:

[0098] If yes, the identified dynamics parameters have met the accuracy requirement, and the identification process can be stopped.

[0099] If it exceeds the threshold range, the iterative identification of the dynamics parameters needs to be continued, and the total inertia torque of the mechanical arm is recalculated through the semi-linearized dynamics model updated in S4, which is referred to as the update value of the total inertia torque in the embodiment of the application. Similarly, the update value of the total inertia torque and the calculated value of the total inertia torque are both the total inertia torque of the mechanical arm, but their values and sources are different. The calculated value of the total inertia torque is calculated based on the calculated value of the joint friction torque, and the update value of the total inertia torque is obtained from the semi-linearized dynamics model with updated inertia parameters.

[0100] After obtaining the update value of the total inertia torque, the new joint friction torque is calculated again according to the torque relationship which is referred to as the update value of the joint friction torque. Similarly, the update value of the joint friction torque is still the joint friction torque of the mechanical arm, but its value and source are different from the initial value of the joint friction torque and the calculated value of the joint friction torque. The initial value of the total inertia torque in step S4 is replaced with the update value of the joint friction torque, and the friction parameter identification is performed again according to the friction parameter identification equation, that is, the update value of the joint friction torque is used as the new initial value of the total inertia torque to perform the friction parameter identification again, and the new friction parameter is obtained. Step S4 is repeatedly executed. Through this process, the identification accuracy is continuously iteratively identified, and the identification result converges to the threshold value range.

[0101] The identification error of the semi-linearized dynamics model parameters is also compensated in the embodiment of the application:

[0102] The identification error of the semi-linearized dynamics model parameters is related to the complex nonlinear relationship between various factors such as joint position, speed, acceleration, friction characteristics, temperature and mechanical arm torque. Extracting the correlation between these related factors and the identification error of the semi-linearized dynamics model parameters can effectively alleviate the uncertainty of the model.

[0103] After identifying the parameters of the semi-linear dynamic model, the identification error is recalculated using multiple pre-collected datasets based on the identified semi-linear dynamic model, thus constructing an identification error dataset. To improve identification accuracy, nonlinear machine learning algorithms can be used to fit and compensate for the identification error of the semi-linear dynamic model parameters. Therefore, this invention proposes a method for compensating for the identification error of semi-linear dynamic model parameters using deep learning algorithms.

[0104] Convolutional neural networks (CNNs) are a deep learning method used for nonparametric model compensation. In conventional techniques, CNNs are applied to image processing for feature extraction in tasks such as image recognition and classification. However, this invention applies them to processing the state information of a robotic arm reconstructed as a matrix. By utilizing the powerful feature extraction capabilities of CNNs, the complex nonlinear relationship between the state of the robotic arm and the identification error of the semi-linearized dynamic model parameters can be extracted.

[0105] This invention employs a deep convolutional neural network to compensate for parameter identification errors and model uncertainties in a semi-linear dynamics model. The convolutional neural network comprises an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer, with the overall structure as follows: Figure 2 As shown, the core of this method is to construct input and output datasets, provide them to a convolutional neural network to extract the dynamic features of the robotic arm, compensate for the error of the semi-linear dynamic model based on the extraction results, and improve the recognition accuracy using semi-parametric techniques.

[0106] The input image matrix of the convolutional neural network is shown below:

[0107] ;

[0108] ;

[0109] in, Indicates the first The joint angles of each joint. Indicates the first Joint velocity of each joint Indicates the first Joint acceleration of each joint Indicates the first Estimates of joint torques for each joint. Indicates the first Estimates of the joint friction torque of each joint. Indicates the first Estimates of the torque of the robotic arm at each joint.

[0110] The two matrices are input image matrices constructed for two nine-joint robot arms, and are input matrices of the convolutional neural network, and the dimension of the matrix is 9*9*2, and the output matrix is the model error of the semi-linearized dynamics model. The input matrix of the convolutional neural network of the robot arm with 6 joints can be constructed as 6*9*2, and the difference between the two matrices is that the joint arrangement order is different, which is to enable the convolutional neural network to better extract the dynamics characteristics of the robot arm.

[0111] As shown in Figure 3 The proposed dynamics model parameter identification compensation method based on convolutional neural network compensation comprises two convolutional layers, which aims to extract image features using convolution kernels. After each convolutional layer, an activation function layer is introduced to introduce nonlinearity into the convolutional neural network and enhance its ability to learn and adapt to complex data patterns. The activation function layer can alleviate the problem of gradient disappearance and promote network training and convergence. Figure 3 In the embodiment of the application, kernal, 16set represents 16 convolution kernels, kernal, 32set represents 32 convolution kernels, Flattrning represents the tiling expansion process after the completion of the convolutional layer. This process is to access data to the fully connected layer.

[0112] In view of the relatively low dimension of the input data in the embodiment of the application, the pooling layer is omitted in the design of the convolutional neural network. After the second activation function layer, two fully connected neural networks are used to convert the feature map derived from the convolutional layer into a vector, so that feature extraction and classification can be performed through a series of fully connected layers. These fully connected neural networks are composed of 256 neurons and 128 neurons, respectively. Finally, the output layer is responsible for generating the dynamics model error. In order to obtain the optimal parameters of the convolutional neural network, the random gradient descent algorithm with a learning rate of 0.0001 is adopted in the embodiment of the application, and the loss function is defined as follows:

[0113] ;

[0114] wherein, represents the model error of the semi-linearized dynamics model derived from the identified semi-linearized dynamics model. represents the output generated by the convolutional neural network, represents the total number of training samples. The loss function is used to train the convolutional neural network and update the parameters of the optimized convolutional neural network.

[0115] The mechanical arm semi-linearization dynamic model parameters identified by the embodiment of the present application include three parts, which are respectively friction parameters of the mechanical arm, inertia parameters of the mechanical arm and non-linear non-parameterized torque of the neural network compensation of the mechanical arm. The semi-linearization dynamic model after compensation is as follows:

[0116] ;

[0117] wherein, the model error of the semi-linearization dynamic model of the neural network compensation is represented by e (t), the neural network is represented by NN.

[0118] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. Those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

[0119] The specific embodiments of the present application do not constitute a limitation on the protection scope of the present application. Any various other corresponding changes and modifications made according to the technical concept of the present application shall be included in the protection scope of the claims of the present application.

Claims

1. A method for identifying dynamics parameters of a robot arm, characterized in that, Comprise: S1: according to the direction of the joint speed of the robot arm, the segmented joint nonlinear friction model of the robot arm is established: ； in, This represents any joint of the robotic arm. This indicates the total number of joints in the robotic arm. This represents the viscous friction force when the joint velocity is positive. This represents the viscous friction force when the joint velocity is negative. This represents the Coulomb friction force when the joint velocity is positive. This represents the Coulomb friction force when the joint velocity is negative. This term represents the bias term for joint friction when the joint velocity is positive. This term represents the bias term for joint friction when the joint velocity is negative. Indicates the first robotic arm Joint velocity of each joint When the joint velocity is positive, the joint... The nonlinear friction parameters, This indicates that when the joint velocity is negative, the joint... The nonlinear friction parameters, Indicates joint friction, This represents the joint friction torque of the robotic arm. Represents a symbolic function; S2: combined with the segmented joint nonlinear friction model, the semi-linearized dynamics model of the robot arm is established by the iterative semi-linearized dynamics identification algorithm: ； wherein, denotes the total joint torque of the robot arm, denotes the total inertia torque of the robot arm, denotes a subset of the regression matrix without the friction term, denotes the vector of the minimal parameter set of dynamics without the friction parameters, denotes the non-linear, denotes the linear, denotes the inertia torque of the robot arm, denotes the Coriolis torque of the robot arm, denotes the gravity torque of the robot arm; S3: according to the predetermined trajectory, each joint of the robot arm is independently driven, and the initial value of the joint friction torque of each joint is calculated; S4: according to the initial value of the joint friction torque, the following iteration is carried out: The initial value of the joint friction torque is substituted into the friction parameter identification equation, and the friction parameters of the segmented joint nonlinear friction model are identified by using the interior point method, and the friction parameter identification equation is: ； wherein represents an initial value of the joint friction torque According to the identified friction parameters, the segmented nonlinear friction model is updated, and the calculated value of the joint friction torque is calculated according to the updated segmented nonlinear friction model; According to the following torque relationship and the calculated value of the joint friction torque, the calculated value of the total inertia torque of the robot arm is calculated: ； wherein, Jtot represents the total joint torque of the robot arm; According to the calculated value of the total inertia torque, the inertia parameters of the semi-linearized dynamics model are identified by using the weighted least square algorithm, and the inertia parameter identification equation is: ； wherein represents the objective function, According to the inertia parameters, the semi-linearized dynamics model is updated; Determine whether the root mean square error of the robot arm torque and the update amount of the friction parameters are converged to the predetermined threshold: If yes, the identification process is ended; If not, according to the updated semi-linearized dynamics model, the updated value of the total inertia torque of the robot arm is recalculated, and the updated value of the joint friction torque is calculated again according to the total inertia torque of the updated value through the torque relationship, and the updated value of the joint friction torque is used to replace the initial value of the joint friction torque in step S4, and step S4 is repeated.

2. The method of claim 1, wherein, The expression of the predetermined trajectory is as follows: ； wherein, denotes the articulation trajectory period, denotes any time instant within the time interval denotes any time instant within the time interval denotes the time required for the articulation velocity to reverse, denotes the articulation velocity at the time instant denotes the articulation velocity of the articulation at the time instant denotes the desired articulation velocity of the articulation, denotes the desired articulation velocity of the articulation, denotes the motion trajectory parameter.

3. The method of claim 1, wherein, The calculation method of the initial value of the joint friction torque is: ； ； in, , This represents the frictional torque calculated based on the predetermined trajectory. Indicates the position of the joint. Joint velocity is The joint torque of the robotic arm when the joint acceleration is 0. Indicates the position of the joint. Joint velocity is The joint torque of the robotic arm when the joint acceleration is 0.

4. The method of claim 1, wherein, Further comprising S5: the model error of the semi-linearized dynamics model is compensated by using a deep convolutional neural network.

5. The method of claim 4, wherein, The loss function of the deep convolutional neural network is: ； wherein, represents a model error of the semi-linearized dynamics model derived from the identified semi-linearized dynamics model, represents an output generated by the convolutional neural network, represents the total number of training samples.

6. The method of identifying the mechanical arm dynamics parameters according to claim 5, wherein, The compensated semi-linearized dynamics model is: ； wherein, represents a model error of the semi-linearized dynamics model of the neural network compensation, represents a neural network.

Images