CN109773794B - 6-axis robot dynamics parameter identification method based on neural network - Google Patents
6-axis robot dynamics parameter identification method based on neural network Download PDFInfo
- Publication number
- CN109773794B CN109773794B CN201910143907.6A CN201910143907A CN109773794B CN 109773794 B CN109773794 B CN 109773794B CN 201910143907 A CN201910143907 A CN 201910143907A CN 109773794 B CN109773794 B CN 109773794B
- Authority
- CN
- China
- Prior art keywords
- joint
- robot
- neural network
- fourier series
- groups
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 38
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 33
- 238000005457 optimization Methods 0.000 claims abstract description 32
- 238000002474 experimental method Methods 0.000 claims abstract description 24
- 239000011159 matrix material Substances 0.000 claims abstract description 23
- 230000005284 excitation Effects 0.000 claims abstract description 22
- 238000001914 filtration Methods 0.000 claims abstract description 8
- 238000012545 processing Methods 0.000 claims abstract description 8
- 238000005070 sampling Methods 0.000 claims abstract description 6
- 238000012795 verification Methods 0.000 claims abstract description 3
- 230000001133 acceleration Effects 0.000 claims description 17
- 238000013377 clone selection method Methods 0.000 claims description 7
- 238000010367 cloning Methods 0.000 claims description 4
- 238000004458 analytical method Methods 0.000 claims description 3
- 239000000427 antigen Substances 0.000 claims description 3
- 102000036639 antigens Human genes 0.000 claims description 3
- 108091007433 antigens Proteins 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 230000005484 gravity Effects 0.000 claims description 3
- 230000005764 inhibitory process Effects 0.000 claims description 3
- 230000035772 mutation Effects 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 2
- 238000012163 sequencing technique Methods 0.000 claims 1
- 238000010586 diagram Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000002068 genetic effect Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 230000036039 immunity Effects 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Images
Landscapes
- Feedback Control In General (AREA)
Abstract
本发明公开了一种基于神经网络的6轴机器人动力学参数辨识方法。步骤如下：第一步：机器人动力学建模以及线性化；第二步：激励轨迹优化，采用人工免疫算法来优化激励轨迹；第三步：实验采样，让机器人跟随激励轨迹运动，获取多组观测矩阵与关节力矩作为实验数据。第四步：数据处理，采用三标准差准则以及中位值平均滤波法对实验采集的数据进行预处理，降低数据噪声带来的影响。第五步：动力学参数估计，采用神经网络对动力学参数进行估计。第六步：参数验证，再让机器人跟随一条与激励轨迹不同的可执行轨迹，再次采样实验数据，根据辨识所得的动力学参数来预测理论关节力矩，用力矩残差根来评价辨识的动力学参数的可靠性。
The invention discloses a 6-axis robot dynamic parameter identification method based on neural network. The steps are as follows: first step: robot dynamics modeling and linearization; second step: excitation trajectory optimization, using artificial immune algorithm to optimize the excitation trajectory; third step: experimental sampling, let the robot follow the excitation trajectory, and obtain multiple sets of The observation matrix and joint torque are used as experimental data. The fourth step: data processing, the three standard deviation criterion and the median average filtering method are used to preprocess the data collected in the experiment to reduce the impact of data noise. The fifth step: dynamic parameter estimation, using neural network to estimate the dynamic parameters. Step 6: Parameter verification, let the robot follow an executable trajectory different from the excitation trajectory, sample the experimental data again, predict the theoretical joint torque according to the identified dynamic parameters, and use the torque residual root to evaluate the identified dynamic parameters reliability.
Description
Technical Field
The invention relates to a method for identifying kinetic parameters of a robot, in particular to a method for identifying kinetic parameters of a 6-axis robot based on a neural network, which relates to a series industrial robot widely used in the fields of industrial production, logistics transportation and the like, and accurate kinetic parameters are the basis of high-precision control of the robot.
Background
Robot dynamics is the basis for discussing robot control, so dynamics parameter identification is a very key technology in robot control, industrial robots achieve high-precision control, and smooth motion tracks all need high-precision dynamics parameters. At present, a genetic algorithm is mostly adopted for the robot dynamics parameter identification to optimize an excitation track, the least square method is used for iterative estimation of dynamics parameters, the accuracy of the dynamics parameter estimation is low, and the industrial robot is difficult to accurately control.
Disclosure of Invention
Aiming at the defects of the prior art, the invention discloses a method for identifying the kinetic parameters of a 6-axis robot based on a neural network, which improves the estimation precision of the kinetic parameters and has higher fault tolerance.
A method for identifying dynamic parameters of a 6-axis robot based on a neural network comprises the following steps:
the first step of dynamic modeling is carried out, a D-H connecting rod coordinate system of the 6-axis robot is established, and a dynamic equation of the robot is established. And deducing an explicit linear expression of the robot according to a Lagrange kinetic equation, simplifying the expression into a linear form according to a minimum inertia principle, and providing kinetic parameters and an observation matrix which need to be identified.
And (2) second step of excitation trajectory optimization, namely performing Fourier series expansion of joint angles, wherein in order to obtain a Fourier series with continuous boundaries, joint angular velocities and angular accelerations equal to 0 at the beginning of a period, the Fourier series needs to be optimized (improved), a quintic polynomial is used for replacing a constant term to obtain an improved Fourier series, and the related formula is as follows:
wherein i is the ith joint, wfFundamental frequency of Fourier series, aik，bikIs a Fourier series, and is characterized in that,
is a constant term coefficient, and M is the order of a Fourier series.
Optimizing to obtain Fourier series coefficients by an immune clone algorithm to obtain an optimized excitation track;
thirdly, experimental sampling, namely enabling the robot to move along with the excitation track, collecting joint angles of a plurality of groups of robots by using a photoelectric encoder in a plurality of periods at different moments to obtain a plurality of groups of joint moments as experimental samples, obtaining motor feedback current at the same time, and solving motor output moment according to a torque coefficient to obtain the joint moment;
the fourth step of data processing, namely performing data preprocessing on the data acquired by the experiment through a three-standard-deviation and median average filtering method to reduce the influence of data noise on the experiment;
and fifthly, estimating dynamic parameters, namely identifying the dynamic parameters through a neural network according to a plurality of groups of observation matrixes and joint moments obtained through experiments.
And sixthly, verifying parameters, namely giving an executable track different from the excitation track, enabling the robot to move along the track, collecting joint angles of a plurality of groups of robots in a plurality of periods at different moments, feeding back current by a motor, obtaining theoretical joint torque according to the solved kinetic parameters, comparing the theoretical joint torque with joint torque obtained by experiments, and evaluating the reliability of the kinetic parameters by using a torque residual root.
Further, the optimized fourier series of the immune clone algorithm is specifically that each joint of the improved excitation track has 10 degrees of freedom, and the total of 3 joints has 30 degree of freedom coefficients, and meanwhile, the position, speed and acceleration of the joints of the robot are constrained, the optimized result is obtained by adopting the immune clone selection algorithm, the optimized index is the condition number of the observation matrix is minimum, and the optimized steps of the immune clone selection algorithm are as follows:
(1) firstly, antigen is identified, namely, the problem needing optimization is analyzed, and optimization characteristics are selected. And carrying out feasibility analysis on the characteristics to be optimized, and then constructing an affinity function of the characteristics to be optimized to serve as an evaluation index of the problem and construct a constraint condition of the problem. The optimization problem in this context is the 30 degree of freedom coefficients, the minimum condition number as its affinity function, the range of joint motion of the robot, the maximum velocity and the maximum acceleration as its constraints.
(2) The antibody is then initialized. Because the binary coding method has the disadvantages of low efficiency and "hamming cliff", the initial solution adopts a real number coding method. A set of random initial solutions satisfying constraints is generated by encoding feasible solutions to the problem into antibodies in an immune clonal selection algorithm.
(3) The affinity was calculated for each individual antibody in the population. The performance of the antibodies was evaluated by calculating the affinity of the individual.
(4) The antibodies in the population are ranked according to the magnitude of affinity, and the top 50% of good individuals are selected as good individuals.
(5) And carrying out cloning, mutation and clone inhibition operations on the selected excellent individuals, and deleting and selecting to obtain the same number of filial individuals.
(6) And randomly generating new populations with the number equal to that of the original populations, calculating the affinity of the new populations, and deleting the top 50% of good offspring.
(7) Combining the immune population generated in (5) and (6) with the new population to replace the original population.
(8) And judging whether the obtained new population meets the termination condition of the algorithm, wherein the termination condition is that the iteration times exceed 200 times, if so, terminating the optimization searching process, and otherwise, continuing to perform the iteration optimization searching.
Further, the fourth step is specifically to firstly perform a three-standard-deviation criterion on a plurality of groups of observation matrix data obtained by sampling, and remove gross errors; and then, processing the experimental data by using a median average filtering method to reduce the influence caused by data noise.
Further, the fifth step is to use an observation matrix obtained through an experiment as the input of the neural network, set the initial weight to be a random value of 0 to 1, use the joint moment obtained through the experiment as a label of the output of the neural network, and use the theoretical joint moment as the output of the neural network. And estimating the kinetic parameters by adopting a batch gradient descent optimization algorithm, wherein the weights of the trained neural network are the kinetic parameters to be estimated.
Further, in the first step, for the n-joint robot system, the dynamic equation can be expressed by a second-order nonlinear differential equation by considering the influence of joint friction:
wherein M (q) represents an inertial matrix and q is a joint angle vector;
is the joint angular velocity vector;
is a joint angular acceleration vector;
is a matrix of centrifugal and coriolis forces; g (q) is a robot gravity term; τ is the control force vector.
Its kinetic model can be linearized and simplified:
wherein W is an observation matrix formed by variables of each joint of the robot,
is the kinetic parameter to be identified.
Furthermore, in the third step, the joint angle obtained by the experiment is fitted into a Fourier series, the differential is obtained to obtain the joint angular velocity, the differential is obtained to obtain the joint angular acceleration, and a plurality of groups of observation matrixes are formed to be used as experiment samples.
The invention has the advantages and beneficial effects that:
the traditional method for identifying the kinetic parameters comprises a disintegration measurement method, a CAD method and an integral identification method, and the method provided by the invention belongs to the integral identification method.
The invention provides a method for processing data by using a three-standard-difference criterion and a median average filtering method, which removes gross errors caused by misoperation or incorrect operation, integrates the advantages of two filtering algorithms, can inhibit random interference and filter obvious pulse interference. The excitation tracks are mostly optimized by adopting a genetic algorithm, local optimization is easy to fall into, the calculation is complex, the situation of local optimization can be effectively avoided by adopting an artificial immune algorithm, and the convergence speed of the algorithm is high.
The invention provides a method for estimating kinetic parameters by using a neural network, wherein the neural network has higher fault tolerance and the capability of searching an optimal solution at high speed, and the kinetic parameters of a robot can be quickly and accurately estimated by the neural network.
Drawings
FIG. 1 is a block diagram of a process for estimation of kinetic parameters of a robot in accordance with the present invention;
FIG. 2 is a flow chart of an immune clone optimization algorithm of the present invention;
FIG. 3 is a schematic diagram of the optimization effect of the immune cloning algorithm of the present invention;
FIG. 4 is a schematic diagram of the structure of a neural network in the present invention;
FIG. 5 is a schematic diagram of the gradient descent optimization algorithm of the present invention;
FIG. 6 is a flow chart of the present invention for verifying kinetic parameter reliability.
Detailed Description
The technical scheme of the invention is further explained by combining the attached drawings of the specification.
And selecting a proper 6-axis robot to obtain information such as D-H parameters of the robot.
D-H parameter
alpha1＝pi/2；a1＝160；d1＝0；theta1＝q1；init_theta1＝0
alpha2＝0；a2＝575；d2＝0；theta2＝q2；init_theta2＝pi/2
alpha3＝pi/2；a3＝130；d3＝644；theta3＝q3；init_theta3＝0
alpha4＝-pi/2；a4＝0；d4＝0；theta4＝q4；init_theta4＝0
alpha5＝pi/2；a5＝0；d5＝0；theta5＝q5；init_theta5＝-pi/2
alpha6＝0；a6＝0；d6＝109.5；theta6＝q6；init_theta6＝pi/2
The invention adopts the following steps:
a method for identifying dynamic parameters of a 6-axis robot based on a neural network mainly comprises the following steps:
the first step of dynamic modeling is carried out, a D-H connecting rod coordinate system of the 6-axis robot is established, a dynamic equation of the robot is established, the rear three axes mainly influence the tail end posture of the robot and have small influence on joint moment, so that the robot is not considered, and the front three joint dynamic parameters are mainly identified. And deducing an explicit linear expression of the robot according to a Lagrange kinetic equation, simplifying the expression into a linear form according to a minimum inertia parameter principle, and providing kinetic parameters to be identified and an observation matrix.
For an n-joint robot system, the dynamic equation can be expressed by a second-order nonlinear differential equation by considering the influence of joint friction:
wherein M (q) represents an inertial matrix and q is a joint angle vector;
is the joint angular velocity vector;
is a joint angular acceleration vector;
is a matrix of centrifugal and coriolis forces; g (q) is a robot gravity term; τ is the control force vector.
Its kinetic model can be linearized and simplified:
wherein W is an observation matrix formed by variables of each joint of the robot,
is the kinetic parameter to be identified.
And (2) second step of excitation trajectory optimization, namely performing Fourier series expansion on joint angles, and in order to obtain a Fourier series with continuous boundaries, joint angular velocities and angular accelerations equal to 0 at the beginning of a period, optimizing (improving) the Fourier series, and replacing a constant term by a fifth-order polynomial to obtain an improved Fourier series, wherein the related formula is as follows:
wherein i is the ith joint, wfFundamental frequency of Fourier series, aik，bikIs a Fourier series, and is characterized in that,
is a constant term coefficient, and M is the order of a Fourier series.
And optimizing to obtain Fourier series coefficients by using an immune clone algorithm, wherein the Fourier series optimized by the immune clone algorithm is specifically that each joint of the improved excitation track has 10 degrees of freedom, and the total of 3 joints is 30 degree of freedom coefficients. Meanwhile, the joint position, the speed and the acceleration of the robot are constrained, so that the excitation track optimization problem is a multi-constraint optimization problem. The optimization index is the minimum condition number of the observation matrix, and the traditional method is complex in solving abnormity, so that a better optimization result is obtained by adopting an immune clone selection algorithm.
The optimization steps of the immune clone selection algorithm are as follows:
(1) firstly, antigen is identified, namely, the problem needing optimization is analyzed, and optimization characteristics are selected. And performing feasibility analysis on the characteristics to be optimized, and then constructing an affinity function of the characteristics, namely, the affinity function is the minimum condition number of the observation matrix and is used as an evaluation index of the problem, and meanwhile, a constraint condition of the problem is constructed. The optimization problem in this context is the 30 degree of freedom coefficients, the minimum condition number as its affinity function, the range of joint motion of the robot, the maximum velocity and the maximum acceleration as its constraints.
(2) The antibody is then initialized. Because the binary coding method has the disadvantages of low efficiency and "hamming cliff", the initial solution adopts a real number coding method. A set of random initial solutions satisfying constraints is generated by encoding feasible solutions to the problem into antibodies in an immune clonal selection algorithm.
(3) The affinity was calculated for each individual antibody in the population. The performance of the antibodies was evaluated by calculating the affinity of the individual.
(4) The antibodies in the population are ranked according to the magnitude of affinity, and the top 50% of good individuals are selected as good individuals.
(5) And carrying out cloning, mutation and clone inhibition operations on the selected excellent individuals, and deleting and selecting to obtain the same number of filial individuals.
(6) And randomly generating new populations with the number equal to that of the original populations, calculating the affinity of the new populations, and deleting the top 50% of good offspring.
(7) Combining the immune population generated in (5) and (6) with the new population to replace the original population.
And the second step is to judge whether the obtained new population meets the termination condition of the algorithm, wherein the termination condition is that the iteration times exceed 200 times. If the termination condition is met, the optimization process is terminated, otherwise, the iterative optimization is continued. And (4) carrying out iterative calculation on the immune clone algorithm to obtain an optimized Fourier series coefficient and generate an optimized excitation track. The algorithm flow chart is shown in figure 2, and the effect chart of the artificial immunity algorithm optimization is shown in figure 3.
Third step experiment sampling, wherein the robot moves along with the excitation track, joint angles of a plurality of groups of robots are acquired by using a photoelectric encoder at different times in a plurality of periods, the joint angles obtained by the experiment are fitted into Fourier series, the differential is solved to obtain the joint angular velocity, the differential is solved to obtain the joint angular acceleration, a plurality of groups of observation matrixes are formed to be used as experiment samples, and the specific steps of deriving the observation matrixes according to the inertia matrix and the Cogowski force matrix are as follows:
w1_1＝ddq1；w1_2＝ddq1*sin(q2)^2+dq1*dq2*sin(2*q2)；w1_3＝-ddq1*sin(2*q2)-2*dq1*dq2*cos(2*q2)；w1_4＝ddq2*sin(q2)+dq2^2*cos(q2)；w1_5＝ddq2*cos(q2)-dq2^2*sin(q2)；w1_6＝0；w1_7＝2*a2*(ddq1*cos(q2)-dq1*dq2*sin(q2))；w1_8＝-2*a2*(ddq1*sin(q2)-dq1*dq2*cos(q2))；w1_9＝-0.5*ddq1*cos(2*q2+2*q3)+dq1*(dq2+dq3)*sin(2*q2+2*q3)；w1_10＝-ddq1*sin(2*q2+2*q3)-2*dq1*(dq2+dq3)*cos(2*q2+2*q3)；
w1_11＝(ddq2+ddq3)*sin(q2+q3)+(dq2+dq3)^2*cos(q2+q3)；
w1_12＝(ddq2+ddq3)*cos(q2+q3)+(dq2+dq3)^2*sin(q2+q3)；
w1_13＝0；
w1_14＝ddq1*(a3*cos(q3)+a3*cos(2*q2+q3)+2*a2*cos(q2+q3))-2*dq1*dq2*(a3*sin(2*q2+q3)+a2*sin(q2+q3))-dq1*dq3*(2*a2*sin(q2+q3)+a3*sin(q3)+a3*sin(q3+2*q2))；
w1_15＝-ddq1*(a3*sin(q3)+a3*sin(2*q2+q3)+2*a2*sin(q2+q3))-2*dq1*dq2*(a3*cos(2*q2+q3)+a2*cos(q2+q3))-dq1*dq3*(2*a2*cos(q2+q3)+a3*cos(q3)+a3*cos(q3+2*q2))；
w2_1＝0；w2_2＝-dq1^2*sin(q2)*cos(q2)；w2_3＝dq1^2*cos(2*q2)；
w2_4＝ddq1*sin(q2)；w2_5＝ddq1*cos(q2)；w2_6＝ddq2；
w2_7＝dq1^2*a2*sin(q2)+g*cos(q2)；w2_8＝dq1^2*a2*cos(q2)-g*sin(q2)；
w2_9＝-dq1^2*sin(q2+q3)*cos(q2+q3)；w2_10＝dq1^2*cos(2*q2+2*q3)；
w2_11＝ddq1*sin(q2+q3)；w2_12＝ddq1*cos(q2+q3)；w2_13＝ddq2+ddq3；
w2_14＝(2*ddq2*ddq3)*a3*cos(q3)-a3*sin(q3)*(2*dq2+dq3)*dq3+(a2*sin(q2+q3)+a3*sin(q3+2*q2))*ddq1^2+g*cos(q2+q3)；
w2_15＝-(2*ddq2*ddq3)*a3*sin(q3)-a3*cos(q3)*(2*dq2+dq3)*dq3+(a2*cos(q2+q3)+a3*cos(q3+2*q2))*ddq1^2-g*sin(q2+q3)；
w3_1＝0；w3_2＝0；w3_3＝0；w3_4＝0；w3_5＝0；w3_6＝0；w3_7＝0；
w3_8＝0；w3_9＝-ddq1*sin(q2+q3)*cos(q2+q3)；w3_10＝dq1^2*cos(2*q2+2*q3)；
w3_11＝ddq1*sin(q2+q3)；w3_12＝ddq1*cos(q2+q3)；w3_13＝ddq2+ddq3；
w3_14＝ddq2*a3*cos(q3)+q1^2*(a2*sin(q2+q3)+a3*sin(q2+q3)*cos(q2)+dq2^2*a3*sin(q3)+g*cos(q2+q3)；
w3_15＝-ddq2*a3*sin(q3)+dq1^2*(a2*cos(q2+q3)+0.5*a3*(cos(q3)+cos(2*q2+q3)))+dq2^2*a3*cos(q3)-g*sin(q2+q3)；
wherein q1, q2, q3 represent the angle of each joint, dq1, dq2, dq3 represent the angular velocity of each joint, ddq1, ddq2, ddq3 represent the angular acceleration of each joint, and g represents the gravitational acceleration.
And simultaneously, the feedback current of the motor is obtained, and the output torque of the motor, which is obtained according to the torque coefficient, is the joint torque.
T＝K*Ia
Wherein T is joint torque, K is a current torque coefficient, and Ia is motor feedback current;
and fourthly, data processing, namely removing gross errors in the experimental data through a three-standard-deviation criterion, and processing the data acquired by the experiment through a median average filtering method to reduce the influence of data noise on the experiment.
And fifthly, estimating dynamic parameters, namely identifying the dynamic parameters through a neural network according to a plurality of groups of observation matrixes and joint moments obtained through experiments. Specifically, a plurality of groups of observation matrixes obtained through experiments are used as input of the neural network, initial weight is set to be a random value of 0-1, a plurality of groups of joint moments obtained through experiments are used as labels of output of the neural network, and theoretical joint moments are used as output of the neural network. And (3) training the model by adopting a batch gradient descent method, wherein the weight of the trained neural network is the kinetic parameter to be identified, the structure of the neural network is shown in figure 4, and the schematic diagram of the gradient descent optimization algorithm is shown in figure 5.
And sixthly, verifying parameters, namely giving an executable track different from the excitation track, enabling the robot to move along the track, collecting joint angles of a plurality of groups of robots in a plurality of periods at different moments, feeding back current by a motor, obtaining theoretical joint torque according to the solved kinetic parameters, comparing the theoretical joint torque with joint torque obtained by experiments, and evaluating the reliability of the kinetic parameters by using a torque residual root. The parameter verification flow chart is as shown in figure 6.
Claims (5)
1. A6-axis robot dynamics parameter identification method based on a neural network is characterized by comprising the following steps:
the method comprises the steps of firstly, performing dynamic modeling, establishing a D-H connecting rod coordinate system of the 6-axis robot, and constructing a dynamic equation of the robot, wherein rear three axes mainly influence the tail end posture of the robot and have small influence on joint moment, so that the rear three axes are not considered, and the front three joint dynamic parameters are mainly identified; deducing an explicit linear expression of the robot according to a Lagrange kinetic equation, simplifying the expression into a linear form according to a minimum inertia principle, and providing kinetic parameters to be identified and an observation matrix;
and secondly, excitation track optimization is carried out, Fourier series expansion of joint angles is carried out, in order to obtain Fourier series with continuous boundaries, joint angular velocity and angular acceleration equal to 0 at the beginning of a period, the Fourier series needs to be optimized, a quintic polynomial replaces a constant term to obtain improved Fourier series, and a correlation formula is as follows:
wherein i is the ith joint, wfFundamental frequency of Fourier series, aik，bikIs a Fourier series, and is characterized in that,
is a constant term coefficient, and M is the order of a Fourier series;
optimizing to obtain Fourier series coefficients by an immune clone algorithm to obtain an optimized excitation track;
thirdly, experimental sampling, namely enabling the robot to move along with the excitation track, collecting joint angles of a plurality of groups of robots by using a photoelectric encoder in a plurality of periods at different moments to obtain a plurality of groups of joint moments as experimental samples, obtaining motor feedback current at the same time, and solving motor output moment according to a torque coefficient to obtain the joint moment;
the fourth step of data processing, namely performing data preprocessing on the data acquired by the experiment through a three-standard-deviation and median average filtering method to reduce the influence of data noise on the experiment;
estimating dynamic parameters, namely identifying the dynamic parameters through a neural network according to a plurality of groups of observation matrixes and joint moments obtained by experiments;
sixthly, parameter verification is carried out, an executable track different from the excitation track is given, the robot moves along the track, a plurality of cycles are carried out, joint angles of a plurality of groups of robots are collected at different moments, current is fed back by a motor, theoretical joint torque obtained according to the solved kinetic parameters is compared with joint torque obtained through experiments, and reliability of the kinetic parameters is evaluated by using a torque residual root;
the optimized Fourier series of the immune clone algorithm is characterized in that each joint of an improved excitation track has 10 degrees of freedom, the total of 3 joints has 30 degree of freedom coefficients, meanwhile, the position, the speed and the acceleration of the joints of the robot are constrained, the optimized result is obtained by adopting the immune clone selection algorithm, the optimized index is the condition number of an observation matrix is minimum, and the optimized steps of the immune clone selection algorithm are as follows:
(1) firstly, identifying an antigen, namely analyzing a problem needing to be optimized, selecting optimization characteristics, carrying out feasibility analysis on the characteristics needing to be optimized, then constructing an affinity function of the affinity function as an evaluation index of the problem, and constructing constraint conditions of the problem at the same time, wherein the optimization problem is 30 degree of freedom coefficients, the minimum condition number is used as the affinity function, the joint motion range of the robot, and the maximum speed and the maximum acceleration are used as the constraint conditions;
(2) then initializing the antibody, wherein the initial solution adopts a real number coding mode, and converts the feasible solution of the problem into the antibody in the immune clone selection algorithm through coding to generate a group of random initial solutions meeting constraint conditions;
(3) calculating the affinity of each individual antibody in the population, and evaluating the performance of the antibody by calculating the affinity of the individual;
(4) sequencing the antibodies in the population according to the size of the affinity, and selecting the top 50% of excellent individuals as excellent individuals;
(5) carrying out cloning, mutation and clone inhibition operations on the selected excellent individuals, and deleting and selecting to obtain the same number of filial generation individuals;
(6) randomly generating new species groups with the same number as the original species groups, calculating the affinity of the new species groups, and deleting the top 50% of excellent offspring groups;
(7) combining the immune population generated in the steps (5) and (6) with a newborn population to replace the original population;
(8) and judging whether the obtained new population meets the termination condition of the algorithm, wherein the termination condition is that the iteration times exceed 200 times, if so, terminating the optimization searching process, and otherwise, continuing to perform the iteration optimization searching.
2. The method for identifying the kinetic parameters of the 6-axis robot based on the neural network as claimed in claim 1, wherein: the fourth step is specifically that three standard deviation criteria are firstly carried out on a plurality of groups of observation matrix data obtained by sampling, and coarse errors are removed; and then, processing the experimental data by using a median average filtering method to reduce the influence caused by data noise.
3. The method for identifying the kinetic parameters of the 6-axis robot based on the neural network as claimed in claim 1, wherein: and the fifth step is specifically that the observation matrix obtained through the experiment is used as the input of the neural network, the initial weight is set to be a random value of 0-1, the joint moment obtained through the experiment is used as a label of the output of the neural network, the theoretical joint moment is used as the output of the neural network, the dynamic parameters are estimated by adopting a batch gradient descent optimization algorithm, and the weight of the trained neural network is the dynamic parameters to be estimated.
4. The method for identifying the kinetic parameters of the 6-axis robot based on the neural network as claimed in claim 1, wherein in the first step, for the robot system with n joints, the kinetic equation can be expressed by a second-order nonlinear differential equation by considering the influence of joint friction:
wherein M (q) represents an inertial matrix and q is a joint angle vector;
is the joint angular velocity vector;
is a joint angular acceleration vector;
is a matrix of centrifugal and coriolis forces; g (q) is a robot gravity term; tau is a control force vector;
its kinetic model can be linearized and simplified:
wherein W is an observation matrix formed by variables of each joint of the robot,
is the kinetic parameter to be identified.
5. The method for identifying the kinetic parameters of the 6-axis robot based on the neural network as claimed in claim 1, wherein in the third step, the experimental joint angles are fitted to Fourier series, the differential is performed to obtain the angular velocity of the joint, the differential is performed to obtain the angular acceleration of the joint, and a plurality of groups of observation matrices are formed to serve as experimental samples.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910143907.6A CN109773794B (en) | 2019-02-26 | 2019-02-26 | 6-axis robot dynamics parameter identification method based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910143907.6A CN109773794B (en) | 2019-02-26 | 2019-02-26 | 6-axis robot dynamics parameter identification method based on neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109773794A CN109773794A (en) | 2019-05-21 |
| CN109773794B true CN109773794B (en) | 2022-01-25 |
Family
ID=66485825
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910143907.6A Active CN109773794B (en) | 2019-02-26 | 2019-02-26 | 6-axis robot dynamics parameter identification method based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109773794B (en) |
Families Citing this family (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110253576B (en) * | 2019-06-17 | 2022-06-24 | 东北大学 | Inspection robot joint motion control method and inspection robot |
| CN110488608B (en) * | 2019-08-14 | 2022-05-27 | 深圳市烨嘉为技术有限公司 | Intelligent kinetic parameter identification method and module for driving and controlling integrated control system |
| CN110716557B (en) * | 2019-11-18 | 2021-05-11 | 河北工业大学 | Robot parameter identification and contact force monitoring method based on prior dynamics knowledge |
| CN111122899B (en) * | 2019-12-11 | 2020-11-17 | 南京航空航天大学 | Incidence angle sideslip angle estimation method for flying in atmospheric disturbance |
| CN111046335B (en) * | 2019-12-26 | 2023-04-28 | 哈尔滨工程大学 | A Method for Determining the Lift Function Coefficient of Integral Low Noise Cam Profile Based on Weighted Least Squares Method |
| CN111546336B (en) * | 2020-04-29 | 2021-01-29 | 北京航空航天大学 | Ash box model parameter identification method and system for robot system |
| CN112179551B (en) * | 2020-08-20 | 2021-06-22 | 清华大学 | Method and device for synchronizing test method and device of joint motor torque coefficient and friction force of robot |
| CN112109084A (en) * | 2020-08-21 | 2020-12-22 | 华南理工大学 | End position compensation method based on robot joint angle compensation and its application |
| CN112327630A (en) * | 2020-11-19 | 2021-02-05 | 上海交通大学 | Semi-parameter industrial robot dynamics modeling method of convolutional neural network |
| CN112518753B (en) * | 2020-12-04 | 2022-02-01 | 浙江理工大学 | Industrial robot trajectory tracking system and method based on neural network iterative compensation |
| CN112948999B (en) * | 2021-02-26 | 2023-08-15 | 北京空间飞行器总体设计部 | Joint reliability verification method for space robot |
| CN114169230B (en) * | 2021-11-24 | 2024-12-20 | 华南理工大学 | A method for identifying robot dynamic parameters |
| CN114516050B (en) * | 2022-03-03 | 2024-02-13 | 上海交通大学 | Robot external force estimation method and system based on configuration Jacobian condition number optimization |
| CN114800536B (en) * | 2022-06-30 | 2022-09-16 | 上海捷勃特机器人有限公司 | Method and device for identifying kinetic parameters of multi-link robot |
| CN115107035B (en) * | 2022-07-26 | 2023-01-06 | 珞石（北京）科技有限公司 | Hybrid control method for monitoring robot single-joint artificial immune system |
| CN117331311B (en) * | 2023-09-21 | 2024-05-14 | 中山大学 | Robot dynamics parameter estimation method based on acceleration-free recursive filtering regression |
| CN118386256B (en) * | 2024-06-28 | 2024-09-06 | 广东海洋大学 | A motion control method and medium based on neurodynamics |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5357424A (en) * | 1992-03-31 | 1994-10-18 | Nippon Telegraph And Telephone Corporation | Dynamic model parameter identification system |
| JPH0869313A (en) * | 1994-08-30 | 1996-03-12 | Nishijima Yoshitaro | Identification method for dynamics in orbit scheme of industrial manipulator |
| CN104614991A (en) * | 2014-12-31 | 2015-05-13 | 南京埃斯顿机器人工程有限公司 | Method for improving robot parameter identification accuracy |
| CN106346513A (en) * | 2016-10-17 | 2017-01-25 | 华南理工大学 | Device and method for identifying kinetic parameters of terminal loads of six-degree-of-freedom robot |
| CN107498562A (en) * | 2017-04-21 | 2017-12-22 | 浙江工业大学 | Six-degree-of-freedom mechanical arm dynamics model identification method |
| CN107671861A (en) * | 2017-11-13 | 2018-02-09 | 无锡信捷电气股份有限公司 | A kind of improved SCARA Identification of Dynamic Parameters of Amanipulator method |
-
2019
- 2019-02-26 CN CN201910143907.6A patent/CN109773794B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5357424A (en) * | 1992-03-31 | 1994-10-18 | Nippon Telegraph And Telephone Corporation | Dynamic model parameter identification system |
| JPH0869313A (en) * | 1994-08-30 | 1996-03-12 | Nishijima Yoshitaro | Identification method for dynamics in orbit scheme of industrial manipulator |
| CN104614991A (en) * | 2014-12-31 | 2015-05-13 | 南京埃斯顿机器人工程有限公司 | Method for improving robot parameter identification accuracy |
| CN106346513A (en) * | 2016-10-17 | 2017-01-25 | 华南理工大学 | Device and method for identifying kinetic parameters of terminal loads of six-degree-of-freedom robot |
| CN107498562A (en) * | 2017-04-21 | 2017-12-22 | 浙江工业大学 | Six-degree-of-freedom mechanical arm dynamics model identification method |
| CN107671861A (en) * | 2017-11-13 | 2018-02-09 | 无锡信捷电气股份有限公司 | A kind of improved SCARA Identification of Dynamic Parameters of Amanipulator method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109773794A (en) | 2019-05-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109773794B (en) | 6-axis robot dynamics parameter identification method based on neural network | |
| CN106557844B (en) | Path planning method for welding robot | |
| CN108621159B (en) | Robot dynamics modeling method based on deep learning | |
| CN111251295B (en) | Visual mechanical arm grabbing method and device applied to parameterized parts | |
| CN111798491B (en) | A Maneuvering Target Tracking Method Based on Elman Neural Network | |
| CN107671861A (en) | A kind of improved SCARA Identification of Dynamic Parameters of Amanipulator method | |
| CN110561438A (en) | Industrial robot manpower/position compliance control method based on kinetic parameter identification | |
| Tang et al. | Learning collaborative pushing and grasping policies in dense clutter | |
| CN107084714A (en) | A multi-robot collaborative target positioning method based on RoboCup3D | |
| Li et al. | Manipulation skill acquisition for robotic assembly based on multi-modal information description | |
| CN113609999B (en) | Human body model building method based on gesture recognition | |
| Shen et al. | Online identification of time-varying dynamical systems for industrial robots based on sparse Bayesian learning | |
| CN116352724A (en) | Mechanical arm dynamics identification method based on neural network moment prediction | |
| Omar et al. | A framework for data driven dynamic modeling of serial manipulators | |
| CN109284580B (en) | Method for identifying kinetic parameters of series robot | |
| CN110110469A (en) | Parallel robot dynamic parameters identification method based on singular value decomposition | |
| Xiao et al. | Joint torque prediction of industrial robots based on PSO-LSTM deep learning | |
| CN114417711A (en) | Neural network mechanical arm dynamics modeling method based on genetic algorithm optimization | |
| CN113987961A (en) | Robot jump landing detection method | |
| CN113290554A (en) | Intelligent optimization control method for Baxter mechanical arm based on value iteration | |
| CN112405568A (en) | Humanoid robot falling prediction method | |
| CN111369038A (en) | Multi-target dynamic scheduling method based on genetic algorithm | |
| CN111829521A (en) | A Data-driven Consistent Target Tracking Method | |
| CN116652967A (en) | Industrial robot dynamic parameter identification method | |
| CN118952190A (en) | A method for identifying dynamic parameters of a measuring robot based on IABC-ANN neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |