Identification method for mechanical arm kinetic parameters

Abstract

The invention relates to the technical field of mechanical arm control, in particular to a mechanical arm dynamic parameter identification method, which comprises the following steps of: identifying friction parameters of a piecewise nonlinear friction model according to an initial value of joint friction torque, and calculating a calculated value of the joint friction torque; calculating a calculated value of a total inertia moment according to the calculated value of the joint friction moment, identifying an inertia parameter of the semi-linear kinetic model according to the calculated value of the total inertia moment, updating the semi-linear kinetic model according to the identified inertia parameter, and recalculating an updated value of the total inertia moment; and calculating an updated value of the joint friction torque according to the updated value of the inertia total torque, replacing the initial value of the joint friction torque with the updated value of the joint friction torque to perform friction parameter identification again, and performing loop iteration until an error is converged to be within a preset threshold value. According to the identification method, the semi-linear kinetic model is adopted and is compensated, the identified parameters are more practical, and the reliability and the accuracy of the kinetic model are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of robot arm control, and specifically provides a method for identifying dynamic parameters of a robot arm. Background Art

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

[0003] Experimental dynamic identification methods can generally be divided into model-free identification methods and model-based identification methods. Model-free dynamic identification methods do not require a predefined mathematical model of the robot arm. This method relies on the measurement of the joint position and current of the robot arm. Typical model-free dynamic identification methods include multilayer perceptron network compensators, Gaussian process regression learning, radial basis function network compensators, and adaptive neural model optimization. Its advantage is that it is more flexible. Compared with model-based methods, this method requires more computing resources and has lower accuracy. The identified dynamic parameters may not cover all physical effects. Accurate estimation of inertial parameters is very difficult and is not suitable for observer design.

[0004] Model-based dynamic identification usually includes dynamic modeling, excitation trajectory design, parameter identification and model verification. The Newton-Euler method is a common method for establishing the dynamic model of a robotic arm. Dynamic linearization can be quickly achieved using toolboxes such as SYMORO and OpenSYMORO. However, the identification accuracy based entirely on the rigid body dynamics 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] Accurately estimating dynamic parameters is a key step in identifying the dynamics of the robot. Several commonly used dynamic parameter identification methods include least squares algorithm, weighted least squares algorithm, maximum likelihood estimation and Kalman filtering. Although these methods can quickly identify dynamic parameters, there are still the following problems: 1. These algorithms are easily affected by noise and outliers. The collected data usually has a lot of noise, which usually leads to large errors in the data, and these errors affect the recognition accuracy.

[0006] 2. The identified dynamic parameters do not meet the physical feasibility constraints, and the identified parameters will exceed the normal values, which will result in the identified model being numerically very accurate but not conforming to the actual physical situation.

[0007] 3. Physical feasibility constraints will reduce the accuracy of dynamic identification and may cause the identified dynamic parameters to be on the edge of the physical feasibility constraints.

[0008] 4. The limitation of speed bandwidth makes it impossible for the robot to realize high-speed operation in the recognition trajectory, which results in that the excitation trajectory cannot be designed to be large, thus resulting in poor recognition effect.

[0009] 5. These algorithms cannot effectively combine nonlinear friction models for dynamic parameter identification; The existing parameter identification method completely linearizes the entire model, which is not suitable for incorporating nonlinear friction models. Summary of the invention

[0010] In order to solve the above problems, the present invention provides a method for identifying the dynamic parameters of a robotic arm. The dynamic model of the identification method adopted by the present invention is a semi-linear dynamic model, which integrates the nonlinearity of the friction model, so that the identified parameters are more accurate and more in line with the actual physical conditions.

[0011] The method for identifying the dynamic parameters of a mechanical arm provided by the present invention comprises: S1: According to the direction of the robot arm joint velocity, establish the segmented joint nonlinear friction model of the robot arm: ; in, represents any joint of the robot arm, Represents the total number of robot arm joints, represents the viscous friction when the joint velocity is positive, represents the viscous friction when the joint velocity is negative, represents the Coulomb friction force when the joint velocity is positive, represents the Coulomb friction when the joint velocity is negative, It represents the bias term of the joint friction force when the joint velocity is positive, represents the bias term of the joint friction force when the joint velocity is negative, Indicates the robot arm The joint velocities of the joints, Indicates that the joint velocity is positive, the joint The nonlinear friction parameter, Indicates that when the joint velocity is negative, the joint The nonlinear friction parameter, Indicates joints The friction force, represents the joint friction torque of the robot arm, represents a symbolic function; S2: Combined with the nonlinear friction model of the segmented joints, the semi-linear dynamic model of the robot arm is established through an iterative semi-linear dynamic identification algorithm: ; in, represents the total joint torque of the robotic arm, represents the total moment of inertia of the robot arm, represents the subset of the regression matrix without friction terms, The vector of the dynamic minimum parameter set representing the frictionless parameters, represents nonlinearity, represents linearity, represents the inertia moment of the robot arm, represents the Coriolis torque of the robot arm, represents the gravitational torque of the robot arm; S3: driving each joint of the robot arm independently according to a predetermined trajectory, and calculating the initial value of the joint friction torque of each joint; S4: According to the initial value of the joint friction torque, perform the following iterations: Substitute the initial value of the joint friction torque into the friction parameter identification equation, and use the interior point method to identify the friction parameters of the segmented joint nonlinear friction model. The friction parameter identification equation is: ; in, Indicates the initial value of the joint friction torque According to the identified friction parameters, 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; According to the following torque relationship and the calculated value of the joint friction torque, calculate the total inertia moment of the robot arm: ; in, represents the total joint torque of the robot; According to the calculated value of the total moment of inertia, the weighted least squares algorithm is used to identify the inertia parameters of the semi-linear dynamic model. The inertia parameter identification equation is: ; in, represents the objective function, Update the semi-linear dynamic model according to the inertia parameters; Determine whether the root mean square error of the manipulator torque and the update amount of the friction parameter converge to the predetermined threshold: If yes, then the identification process ends; If not, then based on the updated semi-linearized dynamic model, recalculate the updated value of the total moment of inertia of the robotic arm, and through the torque relationship, based on the updated value of the total moment of inertia, recalculate the updated value of the joint friction torque, and replace the initial value of the joint friction torque in step S4 with the updated value of the joint friction torque, and repeat step S4.

[0012] Preferably, the expression of the predetermined trajectory is as follows: ; in, represents the joint motion trajectory cycle, Indicates 0 to At any time in time, Indicates the time required for the joint speed to change direction. Indicated in Time joints The joint speed, represents the desired joint velocity of the joint, Represents the motion trajectory parameters.

[0013] Preferably, the calculation method of the initial value of the joint friction torque is: ; ; in, , represents the friction torque calculated according to the predetermined trajectory, Indicates that the joint position is , the joint speed is , the joint torque of the robot arm when the joint acceleration is 0, Indicates that the joint position is , the joint speed is , the joint torque of the robotic arm when the joint acceleration is 0.

[0014] Preferably, the method further includes S5: compensating the model error of the semi-linear dynamics model by using a deep convolutional neural network.

[0015] Preferably, the loss function of the deep convolutional neural network is: ; in, represents the model error of the semilinearized dynamic model derived from the identified semilinearized dynamic model, represents the output generated by the convolutional neural network, Represents the total number of training samples.

[0016] Preferably, the compensated semi-linear dynamic model is: .

[0017] in, represents the model error of the semilinearized dynamics model compensated by the neural network, Represents a neural network.

[0018] Compared with the prior art, the present invention can achieve the following beneficial effects: The dynamic model established by the identification method proposed in the present invention is a semi-linear dynamic model, which is combined with a piecewise nonlinear friction model. The piecewise nonlinear friction model takes into account the speed direction of the robot arm, effectively improves the accuracy of dynamic parameter identification, and thus improves the accuracy of the dynamic model. The dynamic parameters identified by the dynamic parameter identification method proposed in the present invention can meet the physical feasibility constraints, so that the identified dynamic parameter values ​​are more reasonable and more in line with the actual situation. At the same time, the present invention also compensates for the uncertainty of the dynamic model through a deep learning convolutional neural network, which can effectively handle the parameter identification error of the dynamic model, and can improve the accuracy of the dynamic model of the robot arm under low-speed motion, thereby improving the reliability and accuracy of the dynamic model. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a kinetic parameter identification flow chart provided according to an embodiment of the present invention; Figure 2 is an algorithm flow chart of a kinetic parameter identification method provided according to an embodiment of the present invention; Figure 3 It is an overall structural diagram of a compensatory convolutional neural network provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0020] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the following description, the same modules are represented by the same reference numerals. In the case of the same reference numerals, their names and functions are also the same. Therefore, the detailed description thereof will not be repeated.

[0021] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not constitute a limitation of the present invention.

[0022] like Figure 1 As shown, an embodiment of the present invention provides a method for identifying dynamic parameters of a robot arm, wherein the identified dynamic parameters include friction parameters and inertia parameters, and the identification method specifically includes: S1: In order to make the friction model more suitable for the actual situation of the robot arm movement, the embodiment of the present invention retains the nonlinearity of the friction model and establishes a piecewise nonlinear friction model according to the direction of the robot arm joint velocity. The expression is as follows: ; in, represents any joint of the robot arm, Represents the total number of robot arm joints, represents the viscous friction when the joint velocity is positive, represents the viscous friction when the joint velocity is negative, represents the Coulomb friction force when the joint velocity is positive, represents the Coulomb friction when the joint velocity is negative, It represents the bias term of the joint friction force when the joint velocity is positive, represents the bias term of the joint friction force when the joint velocity is negative, Indicates the robot arm The joint velocity of each joint, Indicates that the joint velocity is positive, the joint The nonlinear friction parameter, Indicates that when the joint velocity is negative, the joint The nonlinear friction parameter, and The value of is usually in the range of 0 to 1. Indicates joints The friction force, represents the joint friction torque of the robot arm, represents a symbolic function, and have: .

[0023] The piecewise nonlinear friction model contains eight unknown friction parameters, namely , , , , , , and .

[0024] S2: To achieve the semi-linearization of the dynamics model, the nonlinearity of the segmented joint nonlinear friction model is integrated into the dynamics model. The iterative semi-linear dynamics identification algorithm is used to linearize the robot arm dynamics model. At the same time, the segmented nonlinear joint friction model that maintains nonlinearity is combined to establish a semi-linear dynamics model as follows: ; in, represents the total joint torque of the robotic arm, represents the total moment of inertia of the robot arm, represents the joint friction torque of the robot arm, represents the subset of the regression matrix without friction terms, The vector of the dynamic minimum parameter set representing the frictionless parameters, represents nonlinearity, represents linearity, represents the inertia moment of the robot arm, represents the Coriolis torque of the robot arm, Represents the gravitational moment of the robot arm and the total moment of inertia of the robot arm Including the Coriolis force moment of the robot arm , Moment of inertia and gravitational torque .

[0025] S3: Before identifying the friction parameters of the segmented joint nonlinear friction model, it is necessary to calculate the initial joint friction torque data. In the embodiment of the present invention, the joint friction torque data is referred to as the initial value of the joint friction torque.

[0026] Each joint of the robot arm is driven to move according to a predetermined motion trajectory. The motion trajectory used in the embodiment of the present invention is as follows: ; in, represents the joint motion trajectory cycle, Indicates 0 to At any time in time, Indicates the time required for the joint speed to change direction. Indicated in Time joints The joint speed, represents the desired joint velocity of the joint, Represents the motion trajectory parameters, which are used to ensure trajectory continuity. The trajectory parameters satisfy the following constraints: ; According to the above constraints, the motion trajectory parameters can be calculated The value of .

[0027] In order to ensure the continuous movement speed of the robot during the movement, the joint speed at the beginning and the end are set to 0. The initial value of the joint friction torque is calculated through the single joint movement of the robot. The calculation method is as follows: ; in, , represents the friction torque calculated according to the above motion trajectory, Indicates that the joint position is , the joint speed is , the joint torque of the robot arm when the joint acceleration is 0, Indicates that the joint position is , the joint speed is , the joint torque of the robotic arm when the joint acceleration is 0.

[0028] 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 ,use represents the initial value of the calculated joint friction torque, then the initial value of the joint friction torque is: , according to the initial value of the joint friction torque, start to identify the friction parameters in the dynamic parameters.

[0029] S4: To ensure that the identified kinetic parameters can meet the accuracy requirements, the kinetic parameter identification process needs to be iterated. The specific iterative process is as follows: According to the initial value of the joint friction torque, the interior point method is used to identify the friction parameters. When optimizing the nonlinear friction model of the segmented joint, the physical feasibility constraints are considered. The friction parameter identification equation is as follows: ; ; According to the identified friction parameters, the friction parameters in the original segmented joint nonlinear friction model are updated to the identified friction parameters to obtain an updated segmented joint nonlinear friction model, and the corresponding new joint friction torque is calculated through the updated segmented joint nonlinear friction model. The embodiment of the present invention refers to the joint friction torque as the calculated value of the joint friction torque to distinguish it from the initial value of the joint friction torque. The calculated value of the joint friction torque and the initial value of the joint friction torque are both the joint friction torque of the robotic arm, but their values ​​are different and their sources are different. The initial value of the joint friction torque is the joint friction torque obtained in step S3 based on the single joint motion calculation, and the calculated value of the joint friction torque is obtained by calculating the updated segmented nonlinear friction model after the friction parameters are identified.

[0030] Based on the calculated value of the joint friction torque, the calculated value of the total moment of inertia of the robot arm can be calculated. The joint friction torque and the total moment of inertia of the robot arm satisfy the following torque relationship: ; in, Represents the total torque of the robot arm. The value of the total torque is output by the software tool and is a known fixed value. Then there is the total moment of inertia The total moment of inertia of the robotic arm calculated here is referred to as the calculated value of the total moment of inertia in this embodiment of the present invention.

[0031] After obtaining the calculated value of the total moment of inertia, the inertia parameters of the semi-linearized dynamic model are identified by the weighted least squares algorithm based on the calculated value of the total moment of inertia. The establishment process of the inertia parameter identification equation is as follows: First, establish the initial inertia parameter identification equation: ; ; in, Represents the inertia parameters of the robot arm, which are the inertia parameters in the basic parameter set identified using the weighted least squares algorithm. The inertia parameters do not include friction parameters. represents the objective function, represents the physical feasibility constraint, Indicates The motor rotor moment of inertia of each joint, Indicates The joint quality of each joint, Indicates The moment of inertia of the joints, Indicates The joint center of mass position vector of each joint, express The trace of a matrix is ​​a mathematical operation and is a state-of-the-art technique in mathematical calculations. Represents the identity matrix.

[0032] In order to prevent the identified inertia parameters from reaching the boundary of the physical feasibility set, the embodiment of the present invention modifies the objective function and introduces reference dynamic parameters. The reference dynamic parameters are directly obtained based on existing software, which include reference inertia parameters and reference friction parameters, and both are known fixed values. Here, only the reference inertia parameters in the reference dynamic parameters are introduced. The reference inertia parameters can be used as a guide to make the identified inertia parameters closer to the actual values. After modifying the objective function and introducing the reference inertia parameters, the initial inertia parameter identification equation is changed to: ; ; in, represents the reference inertia parameter, represents the 2-norm Euclidean distance metric.

[0033] After changing the inertia parameter identification equation, the embodiment of the present invention combines an iterative semi-linear identification algorithm and uses a geometric Riemann metric method to re-identify the inertia parameters, further ensuring that the identified inertia parameters are far away from the edge of the physical feasibility constraint set, so that the identified inertia parameters are more in line with the actual situation. The inertia parameter identification equation finally obtained is: ; in, Represents geometric measurement, which is The reference inertial parameters of the joints are expressed as follows: The iterative semi-linear identification algorithm can effectively integrate geometric metrics, making the identification results more consistent with the actual physical system.

[0034] According to the inertia parameters identified by the above inertia parameter identification equation, the semi-linearized dynamic model is updated. Whether the updated semi-linearized dynamic model converges to meet the accuracy requirements requires a threshold judgment. The setting of the threshold requires theoretical calculation based on the accuracy requirements and the characteristics of the dynamic model to obtain a reasonable threshold range.

[0035] Calculate the root mean square error of the manipulator torque and the update of the friction parameters. The root mean square error of the manipulator torque is to sum the square of the error between the actual torque of the manipulator and the torque predicted by the semi-linear dynamic model, take the average value, and then take the square root. The update of the friction parameters is the second norm of the deviation between the friction parameters identified in the previous iteration and the friction parameters identified in this iteration. Preset the threshold to determine whether the root mean square error of the manipulator torque and the update of the friction parameters converge to the threshold range: If yes, the identified dynamic parameters have met the accuracy requirements, and the identification process can be stopped; If it exceeds the threshold range, it is necessary to continue the iterative identification of the dynamic parameters, and recalculate the total moment of inertia of the robotic arm through the semi-linearized dynamic model updated in S4. The embodiment of the present invention calls the total moment of inertia the updated value of the total moment of inertia. Similarly, the updated value of the total moment of inertia and the calculated value of the total moment of inertia are both the total moment of inertia of the robotic arm, but their values ​​and sources are different. The calculated value of the total moment of inertia is obtained based on the calculated value of the joint friction torque, and the updated value of the total moment of inertia is obtained by the semi-linearized dynamic model after updating the inertia parameters.

[0036] After obtaining the updated value of the total moment of inertia, according to the updated value of the total moment of inertia, the moment relationship is used Calculate the new joint friction torque again, which is called the updated value of the joint friction torque. Similarly, the updated value of the joint friction torque is still the joint friction torque of the robot arm, but the value is different from the initial value of the joint friction torque and the calculated value of the joint friction torque. Replace the initial value of the total moment of inertia in step S4 with the updated value of the joint friction torque, and re-identify the friction parameters according to the friction parameter identification equation, that is, use the updated value of the joint friction torque as the new initial value of the total moment of inertia to identify the friction parameters again, obtain the new friction parameters, and repeatedly execute step S4. Utilize this process to continuously iterate the identification to achieve convergence of the identification accuracy until the identification result converges to within the threshold range.

[0037] The embodiment of the present invention also compensates for the identification error of the semi-linear dynamic model parameters: The identification error of the semi-linear dynamic model parameters is related to the complex nonlinear relationship between various factors such as joint position, velocity, acceleration, friction characteristics, temperature and robot arm torque. Extracting the correlation between these related factors and the identification error of the semi-linear dynamic model parameters can effectively alleviate the uncertainty of the model.

[0038] After identifying the parameters of the semi-linearized dynamic model, the identification error is recalculated based on the identified semi-linearized dynamic model using multiple sets of data sets collected in advance to construct an identification error data set. In order to improve the identification accuracy, a nonlinear machine learning algorithm can be used to fit and compensate for the semi-linearized dynamic model parameter identification error. To this end, an embodiment of the present invention proposes a method for compensating the semi-linearized dynamic model parameter identification error using a deep learning algorithm.

[0039] A convolutional neural network is a deep learning method used for non-parametric model compensation. In conventional technology, a convolutional neural network is applied to image processing for feature extraction in tasks such as image recognition and classification. However, an embodiment of the present invention applies it to process state information of a robotic arm reconstructed as a matrix. By utilizing the powerful feature extraction capability of a convolutional neural network, the complex nonlinear relationship between the state of the robotic arm and the parameter identification error of a semi-linearized dynamic model can be extracted.

[0040] The embodiment of the present invention uses a deep convolutional neural network to compensate for the semi-linear dynamic model parameter identification error and model uncertainty. The convolutional neural network includes an input layer, a convolution layer, a pooling layer, a fully connected layer and an output layer. The overall structure is as follows: Figure 2 As shown in the figure, the core of this method is to construct input data sets and output data sets, provide them to the convolutional neural network to extract the dynamic characteristics of the robot arm, compensate for the semi-linear dynamic model error based on the extraction results, and use semi-parametric technology to improve the identification accuracy.

[0041] The input image matrix of the convolutional neural network is as follows: ; ; in, Indicates The joint angles of the joints, Indicates The joint velocities of the joints, Indicates The joint acceleration of each joint, Indicates The estimated values ​​of the joint torques of the joints, Indicates The estimated value of the joint friction torque of each joint, Indicates Estimated values ​​of the robot arm torque at each joint.

[0042] The above two matrices are the input image matrices constructed for two nine-joint robotic arms, and are also the input matrices of the convolutional neural network. The dimension of the matrix is ​​9*9*2, and the output matrix is ​​the model error of the semi-linearized dynamic model. The convolutional neural network input matrix of the six-joint robotic arm can be constructed as 6*9*2. The difference between the above two matrices is that the joint arrangement order is different, which is to enable the convolutional neural network to better extract the dynamic characteristics of the robotic arm.

[0043] like Figure 3 As shown, the dynamic model parameter identification and compensation method based on convolutional neural network compensation proposed in the present invention includes two convolution layers, which are intended to use convolution kernels to extract image features. After each convolution layer, an activation function layer is introduced to introduce nonlinearity into the convolutional neural network to enhance its ability to learn and adapt to complex data patterns. The activation function of the activation function layer can alleviate the problem of gradient disappearance and promote network training and convergence. Figure 3 middle, kernal, 16set means 16 The convolution kernel, kernal, 32set means 32 The convolution kernel of , Flattrning refers to the flattening process performed after the convolution layer is completed, which is to access the data to the fully connected layer.

[0044] Given the relatively low dimensionality of the input data in the embodiment of the present invention, 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 maps derived from the convolutional layer into vectors, 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 dynamic model error. In order to obtain the optimal parameters of the convolutional neural network, the embodiment of the present invention uses a stochastic gradient descent algorithm with a learning rate of 0.0001, and the loss function is defined as follows: ; in, represents the model error of the semilinearized dynamic model derived from the identified semilinearized dynamic 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 and optimize the parameters of the convolutional neural network.

[0045] The semi-linear dynamic model parameters of the robot arm identified by the embodiment of the present invention include three parts in total, namely the friction parameters of the robot arm, the inertia parameters of the robot arm and the nonlinear non-parametric torque of the neural network compensation of the robot arm. The compensated semi-linear dynamic model is as follows: ; in, represents the model error of the semilinearized dynamics model compensated by the neural network, Represents a neural network.

[0046] Although the embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and cannot be understood as limiting the present invention. Those skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

[0047] The above specific implementations of the present invention do not constitute a limitation on the protection scope of the present invention. Any other corresponding changes and modifications made based on the technical concept of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A method for identifying dynamic parameters of a robot arm, characterized in that: include: S1: According to the direction of the robot arm joint velocity, establish the segmented joint nonlinear friction model of the robot arm: ； in, represents any joint of the robot arm, Represents the total number of robot arm joints, represents the viscous friction when the joint velocity is positive, represents the viscous friction when the joint velocity is negative, represents the Coulomb friction force when the joint velocity is positive, represents the Coulomb friction when the joint velocity is negative, It represents the bias term of the joint friction force when the joint velocity is positive, represents the bias term of the joint friction force when the joint velocity is negative, Indicates the robot arm The joint velocity of each joint, Indicates that the joint velocity is positive, the joint The nonlinear friction parameter, Indicates that when the joint velocity is negative, the joint The nonlinear friction parameter, Indicates joints The friction force, represents the joint friction torque of the robot arm, represents a symbolic function; S2: Combined with the nonlinear friction model of the segmented joints, the semi-linear dynamic model of the robot arm is established through an iterative semi-linear dynamic identification algorithm: ； in, represents the total joint torque of the robotic arm, represents the total moment of inertia of the robot arm, represents the subset of the regression matrix without friction terms, The vector of the dynamic minimum parameter set representing the frictionless parameters, represents nonlinearity, represents linearity, represents the inertia moment of the robot arm, represents the Coriolis torque of the robot arm, represents the gravitational torque of the robot arm; S3: driving each joint of the robot arm independently according to a predetermined trajectory, and calculating the initial value of the joint friction torque of each joint; S4: According to the initial value of the joint friction torque, perform the following iterations: Substitute the initial value of the joint friction torque into the friction parameter identification equation, and use the interior point method to identify the friction parameters of the segmented joint nonlinear friction model. The friction parameter identification equation is: ； in, Indicates the initial value of the joint friction torque According to the identified friction parameters, 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; According to the following torque relationship and the calculated value of the joint friction torque, calculate the total inertia moment of the robot arm: ； in, represents the total joint torque of the robot; According to the calculated value of the total moment of inertia, the inertia parameters of the semi-linear dynamic model are identified using a weighted least squares algorithm. The inertia parameter identification equation is: ； in, represents the objective function, updating the semi-linearized dynamics model according to the inertia parameters; Determine whether the root mean square error of the manipulator torque and the update amount of the friction parameter converge to the predetermined threshold: If yes, then the identification process ends; If not, then based on the updated semi-linearized dynamic model, recalculate the updated value of the total moment of inertia of the robotic arm, and through the torque relationship, based on the updated value of the total moment of inertia, recalculate the updated value of the joint friction torque, and use the updated value of the joint friction torque to replace the initial value of the joint friction torque in step S4, and repeat step S4.

2. The method for identifying the dynamic parameters of a robot arm according to claim 1, characterized in that: The expression of the predetermined trajectory is as follows: ； in, represents the joint motion trajectory cycle, Indicates 0 to At any time in time, Indicates the time required for the joint speed to change direction. Indicated in Time joints The joint speed, represents the desired joint velocity of the joint, Represents the motion trajectory parameters.

3. The method for identifying the dynamic parameters of a robot arm according to claim 1, characterized in that: The calculation method of the initial value of the joint friction torque is: ； ； in, , represents the friction torque calculated according to the predetermined trajectory, Indicates that the joint position is , the joint speed is , the joint torque of the robot arm when the joint acceleration is 0, Indicates that the joint position is , the joint speed is , the joint torque of the robotic arm when the joint acceleration is 0.

4. The method for identifying the dynamic parameters of a robot arm according to claim 1, characterized in that: It also includes S5: using a deep convolutional neural network to compensate for the model error of the semi-linear dynamic model.

5. The method for identifying the dynamic parameters of a robot arm according to claim 4, characterized in that: The loss function of the deep convolutional neural network is: ； in, represents the model error of the semilinearized dynamic model derived from the identified semilinearized dynamic model, represents the output generated by the convolutional neural network, Represents the total number of training samples.

6. The method for identifying the dynamic parameters of a robot arm according to claim 5, characterized in that: The compensated semi-linear dynamic model is: 。 in, represents the model error of the semilinearized dynamics model compensated by the neural network, Represents a neural network.

Images