Kinetic parameter identification method for building robot

Abstract

The invention provides a kinetic parameter identification method for a building robot. The kinetic parameter identification method comprises the following steps: step 1, establishing a kinetic model based on a Newton-Euler method and carrying out linearization processing; 2, designing and optimizing an excitation track based on an improved crayfish optimization algorithm; thirdly, mechanical arm joint state collection and data processing are carried out; and 4, mechanical arm parameter estimation and model verification are carried out based on an improved iterative weighted least square method, so that a high-precision building robot dynamic model is efficiently established. According to the method, the problems of poor identification precision and low identification efficiency of the dynamic parameters of the building robot in a severe environment in a traditional identification method are solved, and the identification precision and efficiency of the dynamic parameters of the mechanical arm are improved under the condition that no additional sensor is needed.

Description

Technical Field

[0001] The invention relates to the technical field of robots, and in particular to a dynamic parameter identification method for a construction robot. Background Art

[0002] Construction robots refer to robots that are used in the field of civil engineering. They can greatly improve the efficiency and safety of construction projects. During the construction process, construction robots are mostly used to perform heavy-load assembly and handling work. The objects to be assembled and handled are usually large in size and heavy in load. Large-space, high-speed and heavy-load are typical characteristics of construction robot operations. In the actual production of robots, there are inevitable processing and manufacturing errors, assembly errors, structural deformation, etc. This causes deviations between the actual values ​​of parameters such as the length of the actual robot rod, the center of gravity, and the mass, and the theoretical values, and the errors gradually increase with the increase in joints. In addition, due to the large load of heavy-duty robots, the dynamic model is susceptible to environmental noise and temperature interference, and has strong coupling and strong time-varying characteristics, which lead to uncertainty in robot assembly operations.

[0003] Since physical experiment estimation requires disassembling the robot and then using special measuring devices to weigh and balance each component, the process is complicated and cumbersome. Therefore, it is not suitable to use physical experiment estimation methods to identify robot dynamic parameters; since computer-aided design technology estimation directly measures some parameters of the robot model in the CAD database, but the CAD model does not completely match the actual parts, it is not suitable to use computer-aided design technology estimation methods to identify robot dynamic parameters. Most of the previous robot dynamic parameter identification algorithms have problems with poor identification accuracy and low identification efficiency, or over-emphasize identification accuracy and ignore identification efficiency. Therefore, high-precision and high-efficiency identification of dynamic parameters is the core issue of this study. Summary of the invention

[0004] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0005] In view of the above-mentioned deficiencies, the present invention provides a method for identifying dynamic parameters of a construction robot.

[0006] The technical solution adopted by the present invention to solve its technical problems is: a method for dynamic parameter identification of a construction robot, comprising the following steps: step one, establishing a dynamic model based on the Newton-Euler method and linearizing it; step two, designing and optimizing an excitation trajectory based on an improved crayfish optimization algorithm; step three, collecting and processing data on the joint states of a robotic arm; step four, estimating the parameters of the robotic arm and verifying the model based on an improved iterative weighted least squares method, thereby efficiently establishing a high-precision dynamic model of a construction robot.

[0007] Furthermore, step 1 includes: the research object of the present invention is a six-degree-of-freedom serial robot in the field of construction, which has 6 independent rotation joints. First, the kinematic model of the robot arm is constructed according to the MD-H method; secondly, the dynamic model of the robot arm is established according to the ideas of extrapolation iteration and internalization iteration in the Newton-Euler method; then, the dynamic model is converted into a linear expression form according to the parallel axis theorem; finally, the dynamic model constructed only by the basic inertial parameter set is obtained according to the QR decomposition.

[0008] Furthermore, step 2 includes: the purpose of designing the excitation trajectory is to use the trajectory to excite the role of relevant parameters in the dynamic equation. First, the excitation trajectory expression containing optimization parameters is designed based on the cubic Fourier function; secondly, the optimization model of the excitation trajectory is established with the minimum observation matrix condition number as the objective function, and the robot joint limit constraint conditions are added to the optimization model in the form of a penalty function; finally, the improved crayfish optimization algorithm is used to find the optimal solution for the optimization model.

[0009] Furthermore, step three includes: robot arm joint state acquisition and data processing. First, the excitation trajectory is input into the robot controller for operation; second, the robot joint current, angle and other joint state information are acquired; finally, the joint angle is differentiated, the joint current is multiplied by the proportional coefficient, and then the Butterworth filter is used for filtering.

[0010] Furthermore, step 4 includes: estimation of robot arm parameters based on improved iterative weighted least squares method. First, weighted least squares is used to obtain initial residuals based on processed data; second, a clustering algorithm is used to obtain residual thresholds based on the initial residuals; finally, weighted iterative calculations are performed to finally obtain high-precision dynamic parameters.

[0011] The beneficial effect of the present invention is that a method for identifying the dynamic parameters of a construction robot is proposed, which solves the problem of poor identification accuracy and low identification efficiency of the dynamic parameters of a construction robot under harsh environments under traditional identification methods, and improves the identification accuracy and efficiency of the dynamic parameters of the robotic arm without the need for additional sensors. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0013] Figure 1 It is a flow chart of the present invention;

[0014] Figure 2 This is a schematic diagram of the robot structure of the present invention;

[0015] Figure 3 MD-H parameter diagram of the robot of the present invention;

[0016] Figure 4 The schematic diagram of the Newton-Euler method of the present invention;

[0017] Figure 5 This is a result diagram after the excitation trajectory optimization in the joint space of the present invention;

[0018] Figure 6 This is a diagram showing the processing result of the robot arm joint state data of the present invention;

[0019] Figure 7 This is a flow chart of step 4 of the present invention;

[0020] Figure 8 This is a data diagram comparing the root mean square of the identification error between the kinetic parameter identification algorithm of the present invention and other algorithms;

[0021] Fig. 9 This is a data diagram comparing the identification efficiency of the kinetic parameter identification algorithm of the present invention and other algorithms. DETAILED DESCRIPTION

[0022] Reference Figure 1 A method for identifying dynamic parameters of a construction robot comprises the following steps: step one, establishing a dynamic model based on the Newton-Euler method and linearizing it; step two, designing and optimizing the excitation trajectory based on the improved crayfish optimization algorithm; step three, collecting the joint states of the robot and processing the data; step four, estimating the parameters of the robot and verifying the model based on the improved iterative weighted least squares method, so as to efficiently establish a high-precision dynamic model of the construction robot.

[0023] Reference Figure 1 , Figure 2 , Figure 3 , Figure 4, step 1 includes: the research object of the present invention is a six-degree-of-freedom serial robot in the field of construction, which has six independent rotating joints. The dynamic model of the robot arm is established according to the Newton-Euler method. Considering the friction of the joints, the dynamics of the robot can be expressed as:

[0024]

[0025] Where M(q) represents the inertia matrix, represents the Coriolis force and centrifugal force matrix, G(q) represents the gravitational moment, Respectively represent the angular displacement, angular velocity, angular acceleration of each joint, τ and τ f They represent the driving torque and friction torque of each joint respectively.

[0026] Using the parallel axis theorem, the parameter P in the dynamic equation is extracted to achieve the linearization of the equation:

[0027]

[0028] In the formula, represents the linear regression matrix, which can be obtained by extracting dynamic parameters.

[0029] By using the singular value decomposition method, some inertial parameters with common coefficients in the full inertial parameter set are combined together to construct a dynamic model constructed by the basic inertial parameter set, namely:

[0030]

[0031] In the formula, represents the full-rank observation matrix after rearrangement, P b is defined as the minimum inertia parameter set.

[0032] Reference Figure 1 , Figure 5 , step 2 includes: designing the excitation trajectory according to the cubic Fourier function, the trajectory expression is:

[0033]

[0034] In the formula, They represent the joint angular displacement, angular velocity, and angular acceleration of joint n at time t, respectively. i,0 is a constant term, a i，k , b i，k Represents the coefficients of each order of Fourier function, ω b It is the fundamental frequency of the Fourier function, and the period of the function is T = 2π / ω b K is the number of harmonic terms in the Fourier function, and each Fourier function contains 2k+1 parameters.

[0035] The constant coefficients of the above trajectory are optimized, the joint angle, angular velocity, and angular acceleration are set to be in accordance with the actual constraints, and the condition number of the observation matrix is ​​set as the optimization criterion for the trajectory coefficients. The specific optimization criteria and optimization model are as follows:

[0036]

[0037] The optimization of the excitation trajectory is a typical multivariate nonlinear optimization problem. To solve this problem, the present invention adopts an improved crayfish optimization algorithm to solve it.

[0038] Reference Figure 1 , Figure 6 , step three includes: inputting the excitation trajectory into the robot controller to run, and repeating the operation 5 times to average the data set to reduce the error. The robot obtains joint current, angle, and angular velocity at a sampling frequency of 1 kHz. Since the robotic arm cannot directly obtain the joint angular acceleration and torque, the joint angular velocity is first filtered and then differentiated to obtain the angular acceleration, and the joint current is filtered and then multiplied by the proportional coefficient to obtain the joint torque. In order to achieve filtering, a 5th-order Butterworth filter with a cutoff frequency of 3 is used for a zero-phase digital filter, and the sampling frequency is set to 1 / 0.008.

[0039] Reference Figure 1 , Figure 7 , step 4 includes: Assuming that a single experiment collects m data points, the observation matrix and observation response can be calculated as:

[0040]

[0041] In the formula, the matrix The size of is (mn×p), the matrix The size of is (mn×1).

[0042] The identification problem is expressed as a multivariate linear regression problem, and the identification equation is modified as follows:

[0043]

[0044] Where P b represents the inertial parameter set to be solved, W is the weight function associated with the residual, usually called the weight matrix. Q is P b The cofactor matrix of

[0045] Use the WLS algorithm to estimate the parameters:

[0046]

[0047] The initial residuals determined by WLS are:

[0048]

[0049] Manhattan distance is used as the classification basis, and the residual threshold δ of weighted iteration is obtained according to the clustering algorithm. In the i-th iteration process, the weights Δw are assigned to the collection points in turn. ij (The weight calculation rule adopts the IGG III method), and its increment is recorded as the diagonal matrix ΔW i , whose elements are K ij Δw ij (K ij is the selection factor, when ψ ij When ≥δ, K ij =1; when ψ ij When <δ, K ij =0). Then the new cofactor matrix Q of the estimated parameters is i+1 for:

[0050]

[0051] According to the matrix inversion lemma, it can be expanded as:

[0052] Q i+1 =Q i +ΔQ i

[0053]

[0054] In the formula, Q i represents the cofactor matrix before the weight matrix is ​​updated, ΔQ i Represents the incremental matrix of the cofactor matrix after the weight matrix is ​​updated.

[0055] Similarly, Ω i+I It can be obtained by the following formula:

[0056]

[0057] In the formula, To extract the matrix In and ΔW i The non-zero elements correspond to the matrix composed of row vectors,

[0058] In summary, we can get:

[0059]

[0060] In the formula, represents the parameter estimate at the i+1th iteration, represents the parameter estimate at the i-th iteration, Represents the incremental estimate of the parameters at this iteration.

[0061] The identification algorithm is iterated until the error vector is smaller than the set threshold.

[0062] pass Figure 8 It can be seen that the root mean square error of the algorithm of the present invention is small and conforms to the rule that the first joint identification accuracy is low. Compared with the classic LS algorithm, the root mean square average error of IIRLS is reduced by 19.3%, and the correlation coefficient is increased by 1.96‰. Compared with the IRLS algorithm, the root mean square average error of IIRLS is reduced by 5.7%, and the correlation coefficient is increased by 0.5‰. This shows that to a certain extent, the IIRLS algorithm significantly improves the accuracy of the robot arm dynamics model.

[0063] pass Figure 8 It can be seen that the computational efficiency of the algorithm of the present invention is higher than that of the classical IRLS algorithm, especially when there are more sampling point data in the identification model, the computational efficiency is significantly improved. When the sampling data m=1001, the computational time required for the improved algorithm is only 1 / 30 of that of the classical algorithm. In addition, when the sampling data points increase, the computational time of the IIRLS algorithm increases slightly. For example, when the sampling data points increase from 501 to 1001, the identification time of the classical algorithm increases by about 3.5 times, while the identification time of the IIRLS algorithm only increases by 2 times, which shows that the IIRLS algorithm is very efficient for processing a large number of sampling points.

[0064] The above is only a specific embodiment of the present invention, and cannot be used to limit the scope of the invention. Therefore, the replacement of equivalent components, or equivalent changes and modifications made according to the scope of protection of the patent of the present invention, should still fall within the scope of this patent. In addition, the technical features of the present invention can be freely combined with each other, with each other, and with each other.

Claims

1. A method for identifying dynamic parameters of a construction robot, characterized in that: The method comprises the following steps: step 1, establishing a dynamic model based on the Newton-Euler method and performing linearization processing; step 2, designing and optimizing the excitation trajectory based on the improved crayfish optimization algorithm; step 3, collecting the joint states of the robotic arm and performing data processing; step 4, estimating the parameters of the robotic arm and performing model verification based on the improved iterative weighted least squares method, thereby efficiently establishing a high-precision dynamic model of the construction robot.

2. A method for identifying dynamic parameters of a construction robot according to claim 1, characterized in that: Step one includes: the research object of the present invention is a six-degree-of-freedom serial robot in the construction field, which has 6 independent rotation joints. First, a robot arm kinematic model is constructed according to the MD-H method; secondly, a robot arm dynamic model is established according to the ideas of extrapolation iteration and internalization iteration in the Newton-Euler method; then, the dynamic model is converted into a linear expression according to the parallel axis theorem; finally, a dynamic model constructed only by a basic inertial parameter set is obtained according to QR decomposition.

3. A method for identifying dynamic parameters of a construction robot according to claim 2, characterized in that: The joint friction considered is Coulomb-viscous friction, which belongs to the linear friction of robot joints. There are 48 parameters to be identified in the minimum inertia parameter set.

4. The method for identifying dynamic parameters of a construction robot according to claim 1, characterized in that: In step 2, the excitation trajectory is designed according to the cubic Fourier function, which has a total of 42 parameters to be optimized.

5. A method for identifying dynamic parameters of a construction robot according to claim 4, characterized in that: When constructing the optimization model, the optimization objective function is to minimize the condition number of the observation matrix. The equality constraints and inequality constraints are added to the objective function in the form of penalty functions. According to the optimization model, the improved crayfish optimization algorithm (COA) is used to seek the optimal solution.

6. A method for identifying dynamic parameters of a construction robot according to claim 1, characterized in that: In step three, after obtaining the joint status information, the joint angular velocity is first filtered and differentiated again to obtain the angular acceleration, the joint current is filtered and then multiplied by the proportional coefficient to obtain the joint torque, and then a 5th-order Butterworth filter with a cutoff frequency of 3 is used for data filtering and smoothing, and the sampling frequency is set to 1 / 0.

008.

7. The method for identifying dynamic parameters of a construction robot according to claim 1, characterized in that: In step 4, the dynamic model identification problem is transformed into a multivariate linear regression problem, and the identification equation is corrected to P b =Q·Ω, decompose Q and Ω into two parts, one of which is the parameter P before iteration b The other part is the incremental matrix of the cofactor matrix after determining the weight matrix W, thus avoiding the inverse operation of the cofactor matrix in the calculation of the classic IRLS algorithm.

8. The method for identifying dynamic parameters of a construction robot according to claim 7, characterized in that: The residual vector of weighted least squares is calculated according to the least squares algorithm, and the robot joints corresponding to the data points are weighted according to the residual values. At the same time, the heteroscedasticity of the error is corrected to obtain the initial residual Ψ0 of the improved iterative weighted least squares algorithm with a smaller residual value.

9. A method for identifying dynamic parameters of a construction robot according to claim 8, characterized in that: For the initial residual Ψ0, the Kmeans clustering algorithm is used to divide it into two categories. The category with a larger residual value is considered to have a larger data error and is used as the selection object. At the same time, the minimum residual value of this category is used as the residual threshold δ.

10. A method for identifying dynamic parameters of a construction robot according to claim 9, characterized in that: According to the threshold δ, the data points are assigned weights and weight judgment factors in turn, and the high-dimensional matrix ΔQ is converted according to whether the weight judgment factor is zero. i , ΔΩ i Convert to a low-dimensional matrix to improve the computational efficiency of the triple inversion of the complex matrix and iteratively calculate the incremental estimate of the parameter until the set threshold is met.

Images