CN109732605A - A kind of compensation method and system of joint of robot moment of friction - Google Patents
A kind of compensation method and system of joint of robot moment of friction Download PDFInfo
- Publication number
- CN109732605A CN109732605A CN201910052603.9A CN201910052603A CN109732605A CN 109732605 A CN109732605 A CN 109732605A CN 201910052603 A CN201910052603 A CN 201910052603A CN 109732605 A CN109732605 A CN 109732605A
- Authority
- CN
- China
- Prior art keywords
- neural network
- friction
- joint
- data
- angular velocity
- 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.)
- Granted
Links
Landscapes
- Feedback Control In General (AREA)
Abstract
The present invention discloses the compensation method and system of a kind of joint of robot moment of friction.The problem of using compensation method and system of the invention, in industrial robot operational process, after the parameter changes such as joint temperature, load, lubrication and abrasion, friction model also can accordingly change, and can be avoided the feedforward compensation method failure based on fixed friction model.On this basis, the moment of friction training neural network generated using the friction model prediction of identification can faster approach ideal effect so that the training learning process of neural network is more efficient, improve the tracking speed and tracking accuracy of joint of robot.Further, the present invention is sampled by Gauss and carries out moment of friction feedforward compensation, compensate for the disadvantage that neural network output does not have exploration, neural network is set preferably to restrain, it avoids falling into local optimum, it further ensures that tracking accuracy, even if the parameters such as joint temperature, load, lubrication and abrasion change, can also fast implement the purpose of high precision tracking.
Description
Technical field
The present invention relates to robot fields, a kind of compensation method more particularly to joint of robot moment of friction and are
System.
Background technique
Industrial robot joint friction is one of the principal element for causing joint of robot tracking accuracy to reduce, in order to reduce
Friction influences robot control system bring, it is necessary to compensate to moment of friction.In general, friction torque compensation uses
Feedforward compensation method based on fixed friction model, but due in industrial robot operational process, joint temperature, load, lubrication
Constantly changing with parameters such as abrasions, making the friction model of feedforward compensation that significant changes occur, so as to cause fixed friction is based on
The feedforward compensation method of model fails, and causes the reduction of joint of robot tracking accuracy.
Summary of the invention
The object of the present invention is to provide the compensation method and system of a kind of joint of robot moment of friction, can be realized quickly
The purpose of high precision tracking joint of robot.
To achieve the above object, the present invention provides following schemes:
A kind of compensation method of joint of robot moment of friction, the compensation method include:
It obtains the first joint angular velocity data of robot start-up course and corresponds to first joint angular velocity data
First moment of friction data, the start-up course, which is the robot, to be run with the speed increment starting set to rated speed
Process；
Friction model identification is carried out using first joint angular velocity data and the first moment of friction data, is obtained
Friction model, the input of the friction model are joint angular speed, and the output of the friction model is joint-friction torque；
Neural network is established, the input of the neural network is joint angular speed, and the output of the neural network is joint
Moment of friction；
Obtain the second joint angular velocity data of robot start-up course；
The second joint angular velocity data is inputted into the friction model, obtains the second joint angular velocity data pair
The the first torque prediction data answered；
Using the second joint angular velocity data and the first torque prediction data as the training of the neural network
Data update weight and the biasing of the neural network using gradient descent method, obtain neural network prediction model；
The random angular velocity data of robot is obtained, the random angular velocity data is that the robot is transported with random velocity
Angular velocity data when row, the random velocity are less than or equal to two times of rated speeds；
The random angular velocity data is inputted into the neural network prediction model, obtains the second torque prediction data；
It is that variance carries out Gauss sampling using the second torque prediction data as mean value, 1, obtains friction torque compensation value；
Friction torque compensation is carried out to the joint of robot according to the friction torque compensation value.
Optionally, it is described according to the friction torque compensation value to the joint of robot carry out friction torque compensation it
Afterwards, further includes:
Obtain joint tracing deviation；
Judge whether the joint tracing deviation is greater than deviation threshold；
If so, updating weight and the biasing of the neural network according to the joint tracing deviation, updated mind is obtained
Through Network Prediction Model.
Optionally, weight and the biasing that the neural network is updated according to the joint tracing deviation, specifically includes:
According to formula:Update weight and the biasing of the neural network, wherein θnew
For updated neural network parameter collection, neural network parameter collection includes weight and the biasing of neural network, θoldBefore updating
Neural network parameter collection, α are the learning rate of neural network, NetθFor neural network,For neural network parameter gradient,
E is joint tracing deviation.
Optionally, described to carry out friction mould using first joint angular velocity data and the first moment of friction data
Type identification, obtains friction model, specifically includes:
Friction model identification is carried out using first joint angular velocity data and the first moment of friction data, is obtained
This Trebek model；
Each parameter to be identified in this described Trebek model is determined using nonlinear least square method, is obtained each
Friction parameter；
The friction model of joint of robot is determined according to each friction parameter and this described Trebek model.
Optionally, the neural network includes 1 hidden layer and 10 neurons, the connection type of the neural network are
Full connection, the activation primitive of the neural network are line rectification function.
Optionally, the expression formula of the neural network are as follows:
Wherein, x indicates the input of neural network, wijIndicate the power of neural network
Value, bjIndicate the biasing of neural network, SjIndicate the input of hidden layer, yjIndicate the output of hidden layer, Net (x) indicates nerve net
The output of network.
Optionally, weight and the biasing that the neural network is updated using gradient descent method, is specifically included:
According to formula:Update weight and the biasing of the neural network, wherein θnewFor
Updated neural network parameter collection, neural network parameter collection include weight and the biasing of neural network, θoldFor the mind before update
Through network parameter collection, α is the learning rate of neural network, NetθFor neural network,For neural network parameter gradient.
A kind of compensation system of joint of robot moment of friction, the compensation system include:
First data acquisition module, for obtaining described in the first joint angular velocity data and the correspondence of robot start-up course
First moment of friction data of the first joint angular velocity data, the start-up course are speed increment of the robot to set
Starting is run to the process of rated speed；
Model Distinguish module, for being carried out using first joint angular velocity data and the first moment of friction data
Friction model identification obtains friction model, and the input of the friction model is joint angular speed, and the output of the friction model is
Joint-friction torque；
Neural network module, for establishing neural network, the input of the neural network is joint angular speed, described
The output of neural network is joint-friction torque；
Second data acquisition module, for obtaining the second joint angular velocity data of robot start-up course；
First torque prediction module obtains institute for the second joint angular velocity data to be inputted the friction model
State the corresponding first torque prediction data of second joint angular velocity data；
Neural network determining module, for making the second joint angular velocity data and the first torque prediction data
For the training data of the neural network, weight and the biasing of the neural network are updated using gradient descent method, obtain nerve
Network Prediction Model；
Speed acquiring module, for obtaining the random angular velocity data of robot, the random angular velocity data is described
Angular velocity data when robot is run with random velocity, the random velocity are less than or equal to two times of rated speeds；
Second torque prediction module is obtained for the random angular velocity data to be inputted the neural network prediction model
Obtain the second torque prediction data；
Gauss sampling module is obtained for being that variance carries out Gauss sampling using the second torque prediction data as mean value, 1
Obtain friction torque compensation value；
Torque compensation module, for carrying out moment of friction benefit to the joint of robot according to the friction torque compensation value
It repays.
The specific embodiment provided according to the present invention, the invention discloses following technical effects:
The compensation method and system of a kind of joint of robot moment of friction provided by the invention, utilize the first joint angular speed
Data and the first moment of friction data carry out friction model identification, obtain friction model.Second joint angular velocity data is inputted
Friction model obtains the corresponding first torque prediction data of second joint angular velocity data；Utilize second joint angular velocity data
With the first torque prediction data training neural network, weight and the biasing of neural network are updated in conjunction with gradient descent method, obtains mind
Through Network Prediction Model.Random angular velocity data is inputted into neural network prediction model, obtains the second torque prediction data.With
Two torque prediction data be mean value, 1 be variance carry out Gauss sampling, obtain friction torque compensation value.According to friction torque compensation
Value carries out friction torque compensation to joint of robot.As it can be seen that the present invention carries out friction model identification first, friction model is obtained.
In industrial robot operational process, after the parameter changes such as joint temperature, load, lubrication and abrasion, friction model also can be corresponding
The problem of changing, can be avoided the feedforward compensation method failure based on fixed friction model.On this basis, rubbing using identification
The moment of friction training neural network that model prediction generates is wiped, so that the training learning process of neural network is more efficient, it can
Ideal effect is faster approached, the tracking speed and tracking accuracy of joint of robot are improved.Further, the present invention passes through Gauss
Sampling carries out moment of friction feedforward compensation, compensates for the disadvantage that neural network output does not have exploration, enables neural network more
Good convergence, avoids falling into local optimum, further ensures that tracking accuracy, even if joint temperature, load, lubrication and abrasion etc. are joined
Number changes, and can also fast implement the purpose of high precision tracking.
Detailed description of the invention
It in order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, below will be to institute in embodiment
Attached drawing to be used is needed to be briefly described, it should be apparent that, the accompanying drawings in the following description is only some implementations of the invention
Example, for those of ordinary skill in the art, without creative efforts, can also obtain according to these attached drawings
Obtain other attached drawings.
Fig. 1 is a kind of flow chart of the compensation method of joint of robot moment of friction provided in an embodiment of the present invention；
Fig. 2 is a kind of structural block diagram of the compensation system of joint of robot moment of friction provided in an embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
The object of the present invention is to provide the compensation method and system of a kind of joint of robot moment of friction, can be realized quickly
The purpose of high precision tracking joint of robot.
In order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, with reference to the accompanying drawing and specific real
Applying mode, the present invention is described in further detail.
Fig. 1 is a kind of flow chart of the compensation method of joint of robot moment of friction provided in an embodiment of the present invention.Such as Fig. 1
Shown, a kind of compensation method of joint of robot moment of friction, the compensation method includes:
Step 101: the first joint angular velocity data and corresponding first joint angle for obtaining robot start-up course are fast
First moment of friction data of degree evidence, the start-up course, which is the robot, to be run with the speed increment starting set to volume
Determine the process of revolving speed；
Step 102: carrying out friction model using first joint angular velocity data and the first moment of friction data
Identification obtains friction model, and the input of the friction model is joint angular speed, and the output of the friction model is joint-friction
Torque.
Wherein, step 102: being rubbed using first joint angular velocity data and the first moment of friction data
Model Distinguish obtains friction model, specifically includes:
Friction model identification is carried out using first joint angular velocity data and the first moment of friction data, is obtained
This Trebek model；
Each parameter to be identified in this described Trebek model is determined using nonlinear least square method, is obtained each
Friction parameter；
The friction model of joint of robot is determined according to each friction parameter and this described Trebek model.
Step 103: establish neural network, the input of the neural network is joint angular speed, the neural network it is defeated
It is out joint-friction torque.The neural network includes 1 hidden layer and 10 neurons, the connection type of the neural network
To connect entirely, the activation primitive of the neural network is line rectification function.The expression formula of the neural network are as follows:
Wherein, x indicates the input of neural network, wijIndicate the power of neural network
Value, bjIndicate the biasing of neural network, SjIndicate the input of hidden layer, yjIndicate the output of hidden layer, Net (x) indicates nerve net
The output of network.
Step 104: obtaining the second joint angular velocity data of robot start-up course；
Step 105: the second joint angular velocity data being inputted into the friction model, obtains the second joint angle speed
Degree is according to corresponding first torque prediction data；
Step 106: using the second joint angular velocity data and the first torque prediction data as the nerve net
The training data of network updates weight and the biasing of the neural network using gradient descent method, obtains neural network prediction model.
Weight and the biasing that the neural network is updated using gradient descent method, is specifically included:
According to formula:Update weight and the biasing of the neural network, wherein θnewFor
Updated neural network parameter collection, neural network parameter collection include weight and the biasing of neural network, θoldFor the mind before update
Through network parameter collection, α is the learning rate of neural network, NetθFor neural network,For neural network parameter gradient.
Step 107: obtain the random angular velocity data of robot, the random angular velocity data be the robot with
Angular velocity data when machine speed is run, the random velocity are less than or equal to two times of rated speeds；
Step 108: the random angular velocity data being inputted into the neural network prediction model, obtains the prediction of the second torque
Data；
Step 109: being that variance carries out Gauss sampling using the second torque prediction data as mean value, 1, obtain moment of friction
Offset；
Step 110: friction torque compensation is carried out to the joint of robot according to the friction torque compensation value.
Preferably, it executes step 110: moment of friction is carried out to the joint of robot according to the friction torque compensation value
After compensation, further includes:
Step 111: obtaining joint tracing deviation；
Step 112: judging whether the joint tracing deviation is greater than deviation threshold；
If so, executing step 113；
Step 113: updating weight and the biasing of the neural network according to the joint tracing deviation, obtain updated
Neural network prediction model.
In the present embodiment, step 113: weight and the biasing of the neural network are updated according to the joint tracing deviation,
It specifically includes:
According to formula:Update weight and the biasing of the neural network, wherein θnew
For updated neural network parameter collection, neural network parameter collection includes weight and the biasing of neural network, θoldBefore updating
Neural network parameter collection, α are the learning rate of neural network, NetθFor neural network,For neural network parameter gradient,
E is joint tracing deviation.
Fig. 2 is a kind of structural block diagram of the compensation system of joint of robot moment of friction provided in an embodiment of the present invention.Such as
Shown in Fig. 2, a kind of compensation system of joint of robot moment of friction, the compensation system includes:
First data acquisition module 201, for obtaining the first joint angular velocity data and correspondence of robot start-up course
First moment of friction data of first joint angular velocity data, the start-up course are speed of the robot to set
Increment starting is run to the process of rated speed；
Model Distinguish module 202, for utilizing first joint angular velocity data and the first moment of friction data
Carry out friction model identification, obtain friction model, the input of the friction model is joint angular speed, the friction model it is defeated
It is out joint-friction torque；
Neural network module 203, for establishing neural network, the input of the neural network is joint angular speed,
The output of the neural network is joint-friction torque；
Second data acquisition module 204, for obtaining the second joint angular velocity data of robot start-up course；
First torque prediction module 205 is obtained for the second joint angular velocity data to be inputted the friction model
The corresponding first torque prediction data of the second joint angular velocity data；
Neural network determining module 206, for the second joint angular velocity data and first torque to be predicted number
According to the training data as the neural network, weight and the biasing of the neural network are updated using gradient descent method, are obtained
Neural network prediction model；
Speed acquiring module 207, for obtaining the random angular velocity data of robot, the random angular velocity data is institute
Angular velocity data when robot is run with random velocity is stated, the random velocity is less than or equal to two times of rated speeds；
Second torque prediction module 208, for the random angular velocity data to be inputted the neural network prediction model,
Obtain the second torque prediction data；
Gauss sampling module 209, for being that variance carries out Gauss sampling using the second torque prediction data as mean value, 1,
Obtain friction torque compensation value；
Torque compensation module 210, for carrying out frictional force to the joint of robot according to the friction torque compensation value
Square compensation.
A kind of implementation process of the compensation system of joint of robot moment of friction provided by the invention is as follows:
(1) design joint running track makes joint bring into operation directly since zero-speed with 30 revs/min of speed increment
To arrival normal speed.
(2) in the operational process of joint, a joint angular speed and joint moment data is acquired every 2ms, obtains first
First moment of friction data of joint angular velocity data and corresponding first joint angular velocity data.
(3) the first joint angular velocity data and the first moment of friction data are utilized, Model Distinguish, this obtained trie are carried out
Bake curve (Stribeck) friction model formula is as follows:
In formula, f (v) is moment of friction, and v is joint angular speed, fcFor Coulomb friction torque, fsFor maximum static friction torque,
vsFor Stribeck friction velocity, δ is empirical coefficient, fvFor viscosity friction coefficient.
Wherein, parameter to be identified is (fc, fs, vs, δ, fv), the present embodiment is established by nonlinear least square method and is recognized
Model, formula are as follows:
In formula, fi(v) moment of friction being calculated for Stribeck friction model, yiFor the joint reality collected
Moment of friction, i.e. the first moment of friction data, n are data sample number, the i.e. number of the first moment of friction data.
Then it carries out solving parameter to be identified using gradient descent method.One group of initial solution (f any given firstc, fs,
vs, δ, fv)=(0,0,0,0,0), the gradient of above formula is then sought, solution is updated with a fixed step size, works as minimum mean-square error
When narrowing down in minimum mean-square error threshold range, Stribeck friction model parameter (f can be obtainedc、fs、vs、δ、fv) take
Value, to obtain Sribeck friction model.In the present embodiment, minimum mean-square error threshold value is less than or equal to 0.0001, updates
The step-length range of solution are as follows: 0.001~0.1.
(4) neural network is established, network inputs are joint angular speed, and network output is joint-friction torque, nerve
Network has a hidden layer, includes 10 neurons, and connection type is full connection, and activation primitive is line rectification function
(Rectified LinearUnit, ReLU).Neural network formula is as follows:
Wherein, x indicates the input of neural network, wijIndicate the power of neural network
Value, bjIndicate the biasing of neural network, SjIndicate the input of hidden layer, yjIndicate the output of hidden layer, Net (x) indicates nerve net
The output of network.
(5) in zero-speed to 3000 data points of uniform sampling between rated speed, and the Stribeck obtained by identification
Model calculates moment of friction corresponding to each revolving speed.Then using revolving speed and moment of friction as the training data of neural network,
Neural network is trained using gradient descent method, undated parameter.If neural network parameter integrates as θ, then parameter more new formula
Are as follows:
Wherein, θnewFor updated neural network parameter collection, neural network parameter Ji Bao
Include weight and the biasing of neural network, θoldFor the neural network parameter collection before update, α is the learning rate of neural network, Netθ
For neural network,For neural network parameter gradient.
(6) joint running track is designed, the control period is 2 milliseconds, with random in zero-speed to twice of nominal speed range
Speed moves back and forth.
(7) in the process of running, acquisition angular speed is inputted as neural network in real time, the nerve established by step 5
Network query function obtains network output, i.e. joint-friction torque.Then using joint-friction torque as mean value, 1 is variance, carries out Gauss
Sampling, it is as follows that Gauss samples formula:
U1, U2=r and (0,1)
In formula, U1, U2 be (0,1] two consistent random numbers in section, Z is positive state value, and m is mean value, δ2For variance,
X is Gauss sampled result, friction torque compensation value.
After joint-friction torque feedforward compensation value is calculated by above formula, offset is output to feedthrough before joint
In road, and joint tracing deviation is acquired in next control period.
(8) in the operational process described in step 7, before joint angular speed of 2 milliseconds of acquisitions, joint-friction torque
Offset and joint tracking error are presented, and as one group of training data, and is stored in database.
(9) it samples in the database, carries out neural metwork training, updated according to joint tracing deviation using following formula
The weight of neural network and biasing:
Wherein, θnewFor updated neural network parameter collection, neural network parameter collection
Weight and biasing including neural network, θoldFor the neural network parameter collection before update, α is the learning rate of neural network,
NetθFor neural network,For neural network parameter gradient, e is joint tracing deviation.
(10) repeat step 7- step 9, joint of robot friciton compensation is made to reach an ideal state, i.e. joint
Tracing deviation is less than or equal to deviation threshold set by user.
As it can be seen that the present invention has recognized Stribeck friction model first, and produced using the Stribeck friction model of identification
The training data of raw neural network, is trained neural network, is equivalent to and joined expertise, so that the instruction of neural network
Practice learning process effectively, faster approaches ideal effect.
The output of neural network is a definite value, and the present invention is sampled by Gauss and carries out moment of friction feedforward compensation, made up
Neural network output does not have the disadvantage of exploration, so that neural network is preferably restrained, avoids falling into local optimum.
Meanwhile neural network parameter more new direction is modified with the absolute value of joint tracing deviation, so that training is total
It can be carried out towards advantageous direction.
Therefore, the friction torque compensation method provided by the invention sampled based on neural network and Gauss can be in industrial machine
Network parameter is corrected and adjusted in real time in people's operational process, even if the parameters such as joint temperature, load, lubrication and abrasion become
Change, can also be adjusted to new ideal compensation state rapidly, it is ensured that joint of robot tracking accuracy with higher.
Each embodiment in this specification is described in a progressive manner, the highlights of each of the examples are with other
The difference of embodiment, the same or similar parts in each embodiment may refer to each other.For system disclosed in embodiment
For, since it is corresponded to the methods disclosed in the examples, so being described relatively simple, related place is said referring to method part
It is bright.
Used herein a specific example illustrates the principle and implementation of the invention, and above embodiments are said
It is bright to be merely used to help understand method and its core concept of the invention；At the same time, for those skilled in the art, foundation
Thought of the invention, there will be changes in the specific implementation manner and application range.In conclusion the content of the present specification is not
It is interpreted as limitation of the present invention.
Claims (8)
1. a kind of compensation method of joint of robot moment of friction, which is characterized in that the compensation method includes:
Obtain the first joint angular velocity data and the first of corresponding first joint angular velocity data of robot start-up course
Moment of friction data, the start-up course, which is the robot, to be run with the speed increment starting set to the mistake of rated speed
Journey；
Friction model identification is carried out using first joint angular velocity data and the first moment of friction data, is rubbed
Model, the input of the friction model are joint angular speed, and the output of the friction model is joint-friction torque；
Neural network is established, the input of the neural network is joint angular speed, and the output of the neural network is joint-friction
Torque；
Obtain the second joint angular velocity data of robot start-up course；
The second joint angular velocity data is inputted into the friction model, it is corresponding to obtain the second joint angular velocity data
First torque prediction data；
Using the second joint angular velocity data and the first torque prediction data as the training data of the neural network,
Weight and the biasing of the neural network are updated using gradient descent method, obtain neural network prediction model；
The random angular velocity data of robot is obtained, when the random angular velocity data is that the robot is run with random velocity
Angular velocity data, the random velocity be less than or equal to two times of rated speeds；
The random angular velocity data is inputted into the neural network prediction model, obtains the second torque prediction data；
It is that variance carries out Gauss sampling using the second torque prediction data as mean value, 1, obtains friction torque compensation value；
Friction torque compensation is carried out to the joint of robot according to the friction torque compensation value.
2. compensation method according to claim 1, which is characterized in that it is described according to the friction torque compensation value to described
Joint of robot carries out after friction torque compensation, further includes:
Obtain joint tracing deviation；
Judge whether the joint tracing deviation is greater than deviation threshold；
If so, updating weight and the biasing of the neural network according to the joint tracing deviation, updated nerve net is obtained
Network prediction model.
3. compensation method according to claim 1, which is characterized in that described according to joint tracing deviation update
The weight of neural network and biasing, specifically include:
According to formula:Update weight and the biasing of the neural network, wherein θnewTo update
Neural network parameter collection afterwards, neural network parameter collection include weight and the biasing of neural network, θoldFor the nerve net before update
Network parameter set, α are the learning rate of neural network, NetθFor neural network,For neural network parameter gradient, e is to close
Save tracing deviation.
4. compensation method according to claim 1, which is characterized in that it is described using first joint angular velocity data and
The first moment of friction data carry out friction model identification, obtain friction model, specifically include:
Friction model identification is carried out using first joint angular velocity data and the first moment of friction data, obtains this spy
In Bake model；
Each parameter to be identified in this described Trebek model is determined using nonlinear least square method, obtains each friction
Parameter；
The friction model of joint of robot is determined according to each friction parameter and this described Trebek model.
5. compensation method according to claim 1, which is characterized in that the neural network includes 1 hidden layer and 10
Neuron, the connection type of the neural network are full connection, and the activation primitive of the neural network is line rectification function.
6. compensation method according to claim 5, which is characterized in that the expression formula of the neural network are as follows:
Wherein, x indicates the input of neural network, wijIndicate the weight of neural network, bjTable
Show the biasing of neural network, SjIndicate the input of hidden layer, yjIndicate the output of hidden layer, Net (x) indicates the defeated of neural network
Out.
7. compensation method according to claim 1, which is characterized in that described to update the nerve net using gradient descent method
The weight of network and biasing, specifically include:
According to formula:Update weight and the biasing of the neural network, wherein θnewAfter updating
Neural network parameter collection, neural network parameter collection includes weight and the biasing of neural network, θoldFor the neural network before update
Parameter set, α are the learning rate of neural network, NetθFor neural network,For neural network parameter gradient.
8. a kind of compensation system of joint of robot moment of friction, which is characterized in that the compensation system includes:
First data acquisition module, for obtaining the first joint angular velocity data and corresponding described first of robot start-up course
First moment of friction data of joint angular velocity data, the start-up course are the robot with the speed increment starting set
It runs to the process of rated speed；
Model Distinguish module, for being rubbed using first joint angular velocity data and the first moment of friction data
Model Distinguish obtains friction model, and the input of the friction model is joint angular speed, and the output of the friction model is joint
Moment of friction；
Neural network module, for establishing neural network, the input of the neural network is joint angular speed, the nerve
The output of network is joint-friction torque；
Second data acquisition module, for obtaining the second joint angular velocity data of robot start-up course；
First torque prediction module obtains described for the second joint angular velocity data to be inputted the friction model
The corresponding first torque prediction data of two joint angular velocity datas；
Neural network determining module, for using the second joint angular velocity data and the first torque prediction data as institute
The training data for stating neural network updates weight and the biasing of the neural network using gradient descent method, obtains neural network
Prediction model；
Speed acquiring module, for obtaining the random angular velocity data of robot, the random angular velocity data is the machine
Angular velocity data when people is run with random velocity, the random velocity are less than or equal to two times of rated speeds；
Second torque prediction module, for the random angular velocity data to be inputted the neural network prediction model, acquisition the
Two torque prediction data；
Gauss sampling module is rubbed for being that variance carries out Gauss sampling using the second torque prediction data as mean value, 1
Wipe torque compensation value；
Torque compensation module, for carrying out friction torque compensation to the joint of robot according to the friction torque compensation value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910052603.9A CN109732605B (en) | 2019-01-21 | 2019-01-21 | Robot joint friction torque compensation method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910052603.9A CN109732605B (en) | 2019-01-21 | 2019-01-21 | Robot joint friction torque compensation method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109732605A true CN109732605A (en) | 2019-05-10 |
| CN109732605B CN109732605B (en) | 2021-02-26 |
Family
ID=66365403
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910052603.9A Active CN109732605B (en) | 2019-01-21 | 2019-01-21 | Robot joint friction torque compensation method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109732605B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110919633A (en) * | 2019-12-06 | 2020-03-27 | 泉州市微柏工业机器人研究院有限公司 | Robot position deviation method based on torque control |
| CN111428317A (en) * | 2020-04-06 | 2020-07-17 | 宁波智诚祥科技发展有限公司 | Joint friction torque compensation method based on 5G and recurrent neural network |
| CN111639749A (en) * | 2020-05-25 | 2020-09-08 | 上海智殷自动化科技有限公司 | Industrial robot friction force identification method based on deep learning |
| CN112152539A (en) * | 2020-09-29 | 2020-12-29 | 中国船舶重工集团公司第七二四研究所 | Neural network compensation motor load moment observer implementation method |
| CN112677156A (en) * | 2020-12-30 | 2021-04-20 | 法奥(淄博)智能装备有限公司 | Robot joint friction force compensation method |
| CN113295311A (en) * | 2021-04-27 | 2021-08-24 | 北京交通大学 | Method for determining friction torque between rolling bearing roller and raceway and testing device |
| CN113742865A (en) * | 2021-09-10 | 2021-12-03 | 遨博（北京）智能科技有限公司 | Data processing method |
| CN113821006A (en) * | 2020-05-07 | 2021-12-21 | 牧今科技 | Method and computing system for determining values of error parameters indicative of robot calibration quality |
| CN113855473A (en) * | 2021-08-25 | 2021-12-31 | 上海傅利叶智能科技有限公司 | Method and device for controlling exoskeleton robot and exoskeleton robot |
| CN114028164A (en) * | 2021-11-18 | 2022-02-11 | 深圳华鹊景医疗科技有限公司 | Rehabilitation robot control method and device and rehabilitation robot |
| CN114800498A (en) * | 2022-04-20 | 2022-07-29 | 上海捷勃特机器人有限公司 | SCARA robot moment feedforward compensation method |
| CN115042191A (en) * | 2022-08-12 | 2022-09-13 | 季华实验室 | Pre-training model fine-tuning training method and device, electronic equipment and storage medium |
| US11717966B2 (en) | 2020-05-07 | 2023-08-08 | Mujin, Inc. | Method and computing system for determining a value of an error parameter indicative of quality of robot calibration |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPS61231611A (en) * | 1985-04-05 | 1986-10-15 | Nissan Motor Co Ltd | Controller of robot |
| JPH04343690A (en) * | 1991-05-22 | 1992-11-30 | Nippon Telegr & Teleph Corp <Ntt> | Joint friction compensating method for multi-joint manipulator |
| WO1997010081A1 (en) * | 1995-09-11 | 1997-03-20 | Kabushiki Kaisha Yaskawa Denki | Robot controller |
| CN101332604A (en) * | 2008-06-20 | 2008-12-31 | 哈尔滨工业大学 | Control method of man machine interaction mechanical arm |
| CN105196294A (en) * | 2015-10-29 | 2015-12-30 | 长春工业大学 | Reconfigurable mechanical arm decentralized control system and control method adopting position measuring |
| CN106313044A (en) * | 2016-09-20 | 2017-01-11 | 华南理工大学 | Feedforward torque compensating method used for industrial robot |
| CN108714896A (en) * | 2018-05-28 | 2018-10-30 | 华南理工大学 | A kind of robot flexibility torque feed forward control method based on kinetics of deformable bodies model |
-
2019
- 2019-01-21 CN CN201910052603.9A patent/CN109732605B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPS61231611A (en) * | 1985-04-05 | 1986-10-15 | Nissan Motor Co Ltd | Controller of robot |
| JPH04343690A (en) * | 1991-05-22 | 1992-11-30 | Nippon Telegr & Teleph Corp <Ntt> | Joint friction compensating method for multi-joint manipulator |
| WO1997010081A1 (en) * | 1995-09-11 | 1997-03-20 | Kabushiki Kaisha Yaskawa Denki | Robot controller |
| CN101332604A (en) * | 2008-06-20 | 2008-12-31 | 哈尔滨工业大学 | Control method of man machine interaction mechanical arm |
| CN105196294A (en) * | 2015-10-29 | 2015-12-30 | 长春工业大学 | Reconfigurable mechanical arm decentralized control system and control method adopting position measuring |
| CN106313044A (en) * | 2016-09-20 | 2017-01-11 | 华南理工大学 | Feedforward torque compensating method used for industrial robot |
| CN108714896A (en) * | 2018-05-28 | 2018-10-30 | 华南理工大学 | A kind of robot flexibility torque feed forward control method based on kinetics of deformable bodies model |
Non-Patent Citations (1)
| Title |
|---|
| 吴晓敏等: ""机器人关节摩擦建模与补偿研究"", 《仪器仪表学报》 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110919633A (en) * | 2019-12-06 | 2020-03-27 | 泉州市微柏工业机器人研究院有限公司 | Robot position deviation method based on torque control |
| CN111428317A (en) * | 2020-04-06 | 2020-07-17 | 宁波智诚祥科技发展有限公司 | Joint friction torque compensation method based on 5G and recurrent neural network |
| CN111428317B (en) * | 2020-04-06 | 2023-06-09 | 宁波智诚祥科技发展有限公司 | Joint friction torque compensation method based on 5G and cyclic neural network |
| CN113821006B (en) * | 2020-05-07 | 2022-10-14 | 牧今科技 | Method and computing system for determining values of error parameters indicative of robot calibration quality |
| US11717966B2 (en) | 2020-05-07 | 2023-08-08 | Mujin, Inc. | Method and computing system for determining a value of an error parameter indicative of quality of robot calibration |
| CN113821006A (en) * | 2020-05-07 | 2021-12-21 | 牧今科技 | Method and computing system for determining values of error parameters indicative of robot calibration quality |
| CN111639749A (en) * | 2020-05-25 | 2020-09-08 | 上海智殷自动化科技有限公司 | Industrial robot friction force identification method based on deep learning |
| CN112152539A (en) * | 2020-09-29 | 2020-12-29 | 中国船舶重工集团公司第七二四研究所 | Neural network compensation motor load moment observer implementation method |
| CN112677156A (en) * | 2020-12-30 | 2021-04-20 | 法奥(淄博)智能装备有限公司 | Robot joint friction force compensation method |
| CN113295311A (en) * | 2021-04-27 | 2021-08-24 | 北京交通大学 | Method for determining friction torque between rolling bearing roller and raceway and testing device |
| CN113295311B (en) * | 2021-04-27 | 2022-04-29 | 北京交通大学 | Method for determining friction torque between rolling bearing roller and raceway and testing device |
| CN113855473A (en) * | 2021-08-25 | 2021-12-31 | 上海傅利叶智能科技有限公司 | Method and device for controlling exoskeleton robot and exoskeleton robot |
| CN113855473B (en) * | 2021-08-25 | 2023-08-15 | 上海傅利叶智能科技有限公司 | Method and device for controlling exoskeleton robot and exoskeleton robot |
| CN113742865A (en) * | 2021-09-10 | 2021-12-03 | 遨博（北京）智能科技有限公司 | Data processing method |
| CN113742865B (en) * | 2021-09-10 | 2024-10-29 | 遨博(北京)智能科技股份有限公司 | Data processing method |
| CN114028164A (en) * | 2021-11-18 | 2022-02-11 | 深圳华鹊景医疗科技有限公司 | Rehabilitation robot control method and device and rehabilitation robot |
| CN114800498A (en) * | 2022-04-20 | 2022-07-29 | 上海捷勃特机器人有限公司 | SCARA robot moment feedforward compensation method |
| CN115042191A (en) * | 2022-08-12 | 2022-09-13 | 季华实验室 | Pre-training model fine-tuning training method and device, electronic equipment and storage medium |
| CN115042191B (en) * | 2022-08-12 | 2022-11-08 | 季华实验室 | Pre-training model fine-tuning training method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109732605B (en) | 2021-02-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109732605A (en) | A kind of compensation method and system of joint of robot moment of friction | |
| CN110806759A (en) | Aircraft route tracking method based on deep reinforcement learning | |
| CN105321345B (en) | A kind of road traffic flow prediction method filtered based on ARIMA models and kalman | |
| CN110376882A (en) | Pre-determined characteristics control method based on finite time extended state observer | |
| CN107498562A (en) | Six-degree-of-freedom mechanical arm dynamics model identification method | |
| CN106487297A (en) | A kind of PMSM parameter identification method based on covariance matching Unscented kalman filtering algorithm | |
| CN107607101A (en) | MEMS gyro sliding-mode control based on interference observer | |
| CN109605377A (en) | A kind of joint of robot motion control method and system based on intensified learning | |
| CN103616818A (en) | Self-adaptive fuzzy neural global rapid terminal sliding-mode control method for micro gyroscope | |
| CN108942935A (en) | Single arm robot adaptive controller and control method based on output position | |
| CN102681443B (en) | Micro electromechanical system gyroscope fuzzy self-adaptive control system based on fuzzy compensation | |
| CN102411305A (en) | Design method of comprehensive disturbance rejection control system for single-rotor wing helicopter/turboshaft engine | |
| CN111308979B (en) | U control system based on multi-rate delay state space model | |
| CN109062046A (en) | Gyroscope system super-twisting sliding mode control method based on RBF neural | |
| CN102411302A (en) | Control method of MEMS (micro-electromechanical system) micro-gyroscope based on direct self-adaptive fuzzy control | |
| CN113091768A (en) | MIMU integral dynamic intelligent calibration compensation method | |
| CN107607102A (en) | MEMS gyro sliding formwork based on interference observer buffets suppressing method | |
| CN116604532A (en) | Intelligent control method for upper limb rehabilitation robot | |
| CN114936409A (en) | Digital twinning modeling method for civil engineering structure | |
| Boegli et al. | A Smoothed GMS friction model for Moving Horizon friction state and parameter estimation | |
| CN206224153U (en) | A kind of control device suitable for uneven torque servo system | |
| CN112904713B (en) | High-precision strong-robustness control method of spatial motion servo mechanism | |
| CN113733117B (en) | Reconfigurable robot human intention identification control method and device | |
| CN107608217A (en) | MEMS gyroscope modified fuzzy sliding mode controlling method based on Hybrid Learning | |
| Dong et al. | Barrier function-based asymptotic tracking control of uncertain nonlinear systems with multiple states constraints |
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 |