CN109623819A - The acquisition methods and device of a kind of robot harmonic drive joint actual torque value - Google Patents
The acquisition methods and device of a kind of robot harmonic drive joint actual torque value Download PDFInfo
- Publication number
- CN109623819A CN109623819A CN201811587023.1A CN201811587023A CN109623819A CN 109623819 A CN109623819 A CN 109623819A CN 201811587023 A CN201811587023 A CN 201811587023A CN 109623819 A CN109623819 A CN 109623819A
- Authority
- CN
- China
- Prior art keywords
- harmonic drive
- drive joint
- machine learning
- value
- gaussian process
- 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
- 238000000034 method Methods 0.000 title claims abstract description 226
- 238000010801 machine learning Methods 0.000 claims abstract description 207
- 238000012549 training Methods 0.000 claims abstract description 48
- 230000008569 process Effects 0.000 claims description 203
- 239000011159 matrix material Substances 0.000 claims description 58
- 230000005540 biological transmission Effects 0.000 claims description 42
- 230000006870 function Effects 0.000 claims description 30
- 238000004364 calculation method Methods 0.000 claims description 10
- 239000003638 chemical reducing agent Substances 0.000 claims description 10
- 238000001914 filtration Methods 0.000 claims description 10
- 230000009467 reduction Effects 0.000 claims description 6
- 238000005259 measurement Methods 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J9/00—Programme-controlled manipulators
- B25J9/16—Programme controls
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J19/00—Accessories fitted to manipulators, e.g. for monitoring, for viewing; Safety devices combined with or specially adapted for use in connection with manipulators
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J9/00—Programme-controlled manipulators
- B25J9/16—Programme controls
- B25J9/1628—Programme controls characterised by the control loop
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J9/00—Programme-controlled manipulators
- B25J9/16—Programme controls
- B25J9/1628—Programme controls characterised by the control loop
- B25J9/1633—Programme controls characterised by the control loop compliant, force, torque control, e.g. combined with position control
Landscapes
- Engineering & Computer Science (AREA)
- Robotics (AREA)
- Mechanical Engineering (AREA)
- Feedback Control In General (AREA)
Abstract
The invention discloses a kind of acquisition methods of robot harmonic drive joint actual torque value, 1) method includes:, utilizes the set of harmonic drive flexibility of joint error model acquisition harmonic drive joint output torque model value；2) sample data sets, are obtained, use sample data sets training machine learning model, then according to the position signal of the motor side in harmonic drive joint, the position signal of the connecting-rod head in harmonic drive joint and current of electric data, the predicted value of harmonic drive joint output torque is obtained；3), harmonic drive joint output torque model value and predicted value mean value are filtered, obtain the corresponding actual torque value of harmonic drive joint output torque model value.The invention discloses a kind of acquisition device of robot harmonic drive joint actual torque value.Using the embodiment of the present invention, the accuracy of the robot harmonic drive joint actual torque value of acquisition can be improved.
Description
Technical Field
The invention relates to a method and a device for acquiring a torque value, in particular to a method and a device for acquiring an actual torque value of a harmonic drive joint of a robot.
Background
Based on the flexible error model, the actual torque of the harmonic drive joint can be measured by using the motor current, the position information of the motor end and the position information of the connecting rod end.
However, when the actual moment of the harmonic drive joint is fitted by using the position errors of the motor end and the connecting rod end, accurate parameters are required for modeling the flexibility of the harmonic drive joint. The selection of parameters in the model directly affects the accuracy of the flexible model. In the harmonic drive flexible modeling process, fitting formulas and empirical formulas are mostly used, so that inherent errors are generated in the model.
Therefore, the problem that the accuracy of the obtained actual torque value of the harmonic transmission joint of the robot is not high exists in the prior art.
Disclosure of Invention
The invention aims to provide a method and a device for acquiring an actual torque value of a harmonic drive joint of a robot, so as to improve the accuracy of the acquired actual torque value of the harmonic drive joint of the robot.
The invention solves the technical problems through the following technical scheme:
the embodiment of the invention provides a method for acquiring an actual torque value of a harmonic drive joint of a robot, which comprises the following steps:
1) acquiring a set of harmonic drive joint output torque model values according to a position signal of a motor end of a harmonic drive joint, a position signal of a connecting rod end of the harmonic drive joint and motor current data by using a harmonic drive joint flexible error model;
2) acquiring a sample data set, training a machine learning model by using the sample data set, and acquiring a predicted value of the output torque of the harmonic drive joint according to a position signal of a motor end of the harmonic drive joint, a position signal of a connecting rod end of the harmonic drive joint and motor current data;
3) and filtering the harmonic drive joint output torque model value and the predicted value mean value to obtain an actual torque value corresponding to the harmonic drive joint output torque model value.
Optionally, the step 1) includes:
by means of the formula (I) and (II),calculating the harmonic transmission deformation angle of the harmonic transmission joint, wherein,
Δθfthe harmonic transmission deformation angle of the harmonic transmission joint is set; q. q.sdPosition signals of the connecting rod end of the harmonic drive joint; q. q.smPosition signals at the motor end; sgn () is a sign function; kωIs the stiffness coefficient of a wave generator in the harmonic drive joint; i isiAs a current of the motor；cωIs the hysteresis coefficient of the wave generator in the harmonic drive joint; l is the reduction ratio of the harmonic reducer in the harmonic drive joint; e is a natural base number; | is a modulo function; thetaerrThe flexible error between the output end of the connecting rod and the output end of the motor is adopted; i is the number of motor current parameters;
by means of the formula (I) and (II),calculating a compliant drive torsion angle, wherein,
delta theta is a flexible transmission torsion angle; delta thetaωA torsion angle that is an input to the wave generator;outputting a torsion angle for the wave generator; l the transmission ratio of the harmonic transmission joint speed changer;
using the formula, τf＝a1Δθ+a2Δθ2+a3Δθ3Calculating the flexible output torque of the harmonic drive joint, and further acquiring a set of model values of the output torque of the harmonic drive joint,
τfthe torque is flexibly output for the harmonic drive joint; a is1Is a preset first parameter; a is2Is a preset second parameter; a is3Is a preset third parameter;
by means of the formula (I) and (II),calculating the output torque model value of the harmonic drive joint, wherein,
τmodeloutputting a torque model value for the harmonic drive joint; tan () is a tangent function; c. CfIs a preset first constant; kfoIs a preset second constant.
Optionally, the step 2) includes:
a: taking the actual output torque value of the harmonic drive joint as a target, dividing the sample data set into a plurality of subsets, and then respectively inputting each subset into each local Gaussian process machine learning model for training, wherein the sample data comprises: the position signal of the motor end of the harmonic drive joint, the position signal of the connecting rod end of the harmonic drive joint, the current data of the motor and the actual torque value of the harmonic drive joint;
b: inputting the position signal of the motor end of the harmonic drive joint, the position signal of the connecting rod end of the harmonic drive joint and the motor current data in the step 1) into the local Gaussian process machine learning model obtained after training, calculating the distance from the sample data to each local Gaussian process machine learning model, and obtaining the local Gaussian process machine learning model corresponding to the maximum value of the distance;
c: acquiring a position factor of the output torque of the harmonic drive joint, and judging whether the maximum value of each distance parameter in the position factor is smaller than a preset threshold value or not;
d: if so, updating the local Gaussian process machine learning model, and taking the central point of the updated local Gaussian process machine learning model as a new fitting point; returning to execute the step B until the judgment result of the step C is negative;
e: and if not, calculating the mean value of the predicted values corresponding to the predicted values of the harmonic drive joint output torque according to the weighted mean value of the predicted values of the harmonic drive joint output torque output by each local Gaussian process machine learning model.
Optionally, step a includes:
a1: dividing the sample data set into a plurality of subsets, and then respectively inputting each subset into each local Gaussian process machine learning model for training, wherein the subsets are determined according to the distance between the position factor of the sample data in the subsets and the central point of the local Gaussian process machine learning model;
a2: aiming at each local Gaussian process machine learning model, by using a formula,calculating the mean of the predicted values, wherein,
f(x*) Is the mean value of the predicted values, y is the predicted value of the local Gaussian process machine learning model of the last iteration, α is the predicted vector, x is*Is sample data;is the variance of gaussian noise; i is a current matrix of the training sample; ()TIs a transposed matrix; k is a covariance matrix in the current iteration;
a3: by means of the formula (I) and (II),and calculating the covariance corresponding to the initial predicted value in the current iteration, wherein,
V(x*) The covariance corresponding to the initial predicted value in the current iteration; k () is a covariance calculation function; k is a radical of*The covariance of the local Gaussian process machine learning model at the last iteration; k is a covariance matrix; x is the number of*Is sample data;is the variance of gaussian noise; i is a current matrix of the training sample; ()TIs a transposed matrix;
a4: judging whether the difference between the covariance and a preset covariance and the difference between the predicted mean value and the actual moment value are both within a preset range;
a5: if so, taking the local Gaussian process machine learning model as a trained local Gaussian process machine learning model;
a6: and if not, updating the parameters of the local Gaussian process machine learning model, and returning to execute the step A2 until the trained local Gaussian process machine learning model is obtained.
Optionally, obtaining a position factor of the output torque of the harmonic drive joint includes:
by means of the formula (I) and (II),calculating the position factor of the output torque of the harmonic drive joint, wherein,
wkoutputting a position parameter of a torque model value relative to a kth local Gaussian process machine learning model for the harmonic drive joint; exp () is an exponential function with a natural base number as the base; x is a subset of input sample data; c. CkMachine learning a position parameter of a center point of the model for each local Gaussian process; ()TIs a transposed matrix; w is a diagonal matrix with the same width as the local Gaussian process machine learning model; k is the number of the machine learning models in the local Gaussian process;
using the formula, w (x, o) ═ w1,...,wk]And obtaining the position factor of the output torque of the harmonic drive joint, wherein,
w (x, o) is a position factor of the harmonic drive joint data moment corresponding to the input sample; x is a subset of input sample data; o is a matrix formed by the central points of the local Gaussian process machine learning models; w is a1Outputting a position parameter of a torque model value relative to a1 st local Gaussian process machine learning model for the harmonic drive joint; w is akOutputting a position parameter of a torque model value relative to a kth local Gaussian process machine learning model for the harmonic drive joint.
Optionally, the updating the local gaussian process machine learning model includes:
and acquiring a new local Gaussian process machine learning model, and taking the new local Gaussian process machine learning model as an updated local Gaussian process machine learning model.
Optionally, step E includes:
by means of the formula (I) and (II),obtaining the probability of the sample data appearing in each local Gaussian process machine learning model, wherein,
p (k | x) is the probability of the sample data x appearing in the kth local Gaussian process machine learning model; x is sample data; k is the number of local Gaussian process machine learning models; m is the number of the obtained local Gaussian process machine learning models corresponding to the harmonic drive joint output torque model value; w is akOutputting a position parameter of a torque model value relative to a kth local Gaussian process machine learning model for the harmonic drive joint;
by means of the formula (I) and (II),calculating the mean value of predicted values corresponding to the model value of the output torque of the harmonic drive joint, wherein,
outputting a predicted value mean value corresponding to the torque model value for the harmonic drive joint;the predicted value of the model is machine-learned for the kth local gaussian process.
The embodiment of the invention provides a device for acquiring an actual torque value of a harmonic drive joint of a robot, which comprises:
the output module is used for acquiring a set of harmonic drive joint output torque model values according to a position signal of a motor end of the harmonic drive joint, a position signal of a connecting rod end of the harmonic drive joint and motor current data by using a harmonic drive joint flexible error model;
the acquisition module is used for acquiring a sample data set, training a machine learning model by using the sample data set, and acquiring a predicted value of the output torque of the harmonic drive joint according to a position signal of a motor end of the harmonic drive joint, a position signal of a connecting rod end of the harmonic drive joint and motor current data;
and the filtering module is used for filtering the harmonic drive joint output torque model value and the predicted value mean value to obtain an actual torque value corresponding to the harmonic drive joint output torque model value.
Optionally, the output module is further configured to:
by means of the formula (I) and (II),calculating the harmonic transmission deformation angle of the harmonic transmission joint, wherein,
Δθfthe harmonic transmission deformation angle of the harmonic transmission joint is set; q. q.sdPosition signals of the connecting rod end of the harmonic drive joint; q. q.smPosition signals at the motor end; sgn () is a sign function; kωIs the stiffness coefficient of a wave generator in the harmonic drive joint; i isiIs the motor current; c. CωIs the hysteresis coefficient of the wave generator in the harmonic drive joint; l is the reduction ratio of the harmonic reducer in the harmonic drive joint; e is a natural base number; | is a modulo function; thetaerrThe flexible error between the output end of the connecting rod and the output end of the motor is adopted; i is the number of motor current parameters;
by means of the formula (I) and (II),calculating a compliant drive torsion angle, wherein,
delta theta is a flexible transmission torsion angle; delta thetaωA torsion angle that is an input to the wave generator;outputting a torsion angle for the wave generator; l the transmission ratio of the harmonic transmission joint speed changer;
using the formula, τf＝a1Δθ+a2Δθ2+a3Δθ3Calculating the flexible output torque of the harmonic drive joint, and further acquiring a set of model values of the output torque of the harmonic drive joint,
τfthe torque is flexibly output for the harmonic drive joint; a is1Is a preset first parameter; a is2Is a preset second parameter; a is3Is a preset third parameter;
by means of the formula (I) and (II),calculating the output torque model value of the harmonic drive joint, wherein,
τmodeloutputting a torque model value for the harmonic drive joint; tan () is a tangent function; c. CfIs a preset first constant; kfoIs a preset second constant.
Optionally, the obtaining module is further configured to:
a: taking the actual output torque value of the harmonic drive joint as a target, dividing the sample data set into a plurality of subsets, and then respectively inputting each subset into each local Gaussian process machine learning model for training, wherein the sample data comprises: the position signal of the motor end of the harmonic drive joint, the position signal of the connecting rod end of the harmonic drive joint, the current data of the motor and the actual torque value of the harmonic drive joint;
b: inputting the position signal of the motor end of the harmonic drive joint, the position signal of the connecting rod end of the harmonic drive joint and the motor current data in the step 1) into the local Gaussian process machine learning model obtained after training, calculating the distance from the sample data to each local Gaussian process machine learning model, and obtaining the local Gaussian process machine learning model corresponding to the maximum value of the distance;
c: acquiring a position factor of the output torque of the harmonic drive joint, and judging whether the maximum value of each distance parameter in the position factor is smaller than a preset threshold value or not;
d: if so, updating the local Gaussian process machine learning model, and taking the central point of the updated local Gaussian process machine learning model as a new fitting point; returning to execute the step B until the judgment result of the step C is negative;
e: and if not, calculating the mean value of the predicted values corresponding to the predicted values of the harmonic drive joint output torque according to the weighted mean value of the predicted values of the harmonic drive joint output torque output by each local Gaussian process machine learning model.
Optionally, the obtaining module is further configured to:
a1: dividing the sample data set into a plurality of subsets, and then respectively inputting each subset into each local Gaussian process machine learning model for training, wherein the subsets are determined according to the distance between the position factor of the sample data in the subsets and the central point of the local Gaussian process machine learning model;
a2: aiming at each local Gaussian process machine learning model, by using a formula,calculating the mean of the predicted values, wherein,
f(x*) Is the mean value of the predicted values, y is the predicted value of the local Gaussian process machine learning model of the last iteration, α is the predicted vector, x is*Is sample data;is the variance of gaussian noise; i is a current matrix of the training sample; ()TIs a transposed matrix; k is a covariance matrix in the current iteration;
a3: by means of the formula (I) and (II),and calculating the covariance corresponding to the initial predicted value in the current iteration, wherein,
V(x*) The covariance corresponding to the initial predicted value in the current iteration; k () is a covariance calculation function; k is a radical of*The covariance of the local Gaussian process machine learning model at the last iteration; k is a covariance matrix; x is the number of*Is sample data;is the variance of gaussian noise; i is a current matrix of the training sample; ()TIs a transposed matrix;
a4: judging whether the difference between the covariance and a preset covariance and the difference between the predicted mean value and the actual moment value are both within a preset range;
a5: if so, taking the local Gaussian process machine learning model as a trained local Gaussian process machine learning model;
a6: and if not, updating the parameters of the local Gaussian process machine learning model, and returning to execute the step A2 until the trained local Gaussian process machine learning model is obtained.
Optionally, the obtaining module is further configured to:
by means of the formula (I) and (II),calculating the position factor of the output torque of the harmonic drive joint, wherein,
wkoutputting a position parameter of a torque model value relative to a kth local Gaussian process machine learning model for the harmonic drive joint; exp () is an exponential function with a natural base number as the baseCounting; x is a subset of input sample data; c. CkMachine learning a position parameter of a center point of the model for each local Gaussian process; ()TIs a transposed matrix; w is a diagonal matrix with the same width as the local Gaussian process machine learning model; k is the number of the machine learning models in the local Gaussian process;
using the formula, w (x, o) ═ w1,...,wk]And obtaining the position factor of the output torque of the harmonic drive joint, wherein,
w (x, o) is a position factor of the harmonic drive joint data moment corresponding to the input sample; x is a subset of input sample data; o is a matrix formed by the central points of the local Gaussian process machine learning models; w is a1Outputting a position parameter of a torque model value relative to a1 st local Gaussian process machine learning model for the harmonic drive joint; w is akOutputting a position parameter of a torque model value relative to a kth local Gaussian process machine learning model for the harmonic drive joint.
Optionally, the obtaining module is further configured to:
and acquiring a new local Gaussian process machine learning model, and taking the new local Gaussian process machine learning model as an updated local Gaussian process machine learning model.
Optionally, the obtaining module is further configured to:
by means of the formula (I) and (II),obtaining the probability of the sample data appearing in each local Gaussian process machine learning model, wherein,
p (k | x) is the probability of the sample data x appearing in the kth local Gaussian process machine learning model; x is sample data; k is the number of local Gaussian process machine learning models; m is the number of the obtained local Gaussian process machine learning models corresponding to the harmonic drive joint output torque model value; w is akFor harmonic transmissionThe dynamic joint output torque model value is relative to the position parameter of the kth local Gaussian process machine learning model;
by means of the formula (I) and (II),calculating the mean value of predicted values corresponding to the model value of the output torque of the harmonic drive joint, wherein,
outputting a predicted value mean value corresponding to the torque model value for the harmonic drive joint;the predicted value of the model is machine-learned for the kth local gaussian process.
Compared with the prior art, the invention has the following advantages:
by applying the embodiment of the invention, the identification and measurement of the actual torque of the harmonic drive joint are realized by fusing the flexible error model and the machine learning method, and compared with the prior art which relies on the fitting parameters of the flexible error model, the calculation precision of the flexible error model for calculating the output torque of the flexible harmonic drive joint is improved.
Drawings
Fig. 1 is a schematic flow chart of a method for acquiring an actual torque value of a harmonic drive joint of a robot according to an embodiment of the present invention;
fig. 2 is a schematic flow chart of machine learning in a method for acquiring an actual torque value of a harmonic drive joint of a robot according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of an apparatus for acquiring an actual torque value of a harmonic drive joint of a robot according to an embodiment of the present invention.
Detailed Description
The following examples are given for the detailed implementation and specific operation of the present invention, but the scope of the present invention is not limited to the following examples.
In order to solve the prior art problems, embodiments of the present invention provide a method and an apparatus for acquiring an actual torque value of a harmonic drive joint of a robot, and first introduce the method for acquiring an actual torque value of a harmonic drive joint of a robot according to embodiments of the present invention.
Fig. 1 is a schematic flow chart of a method for acquiring an actual torque value of a harmonic drive joint of a robot according to an embodiment of the present invention; fig. 2 is a schematic flow chart of machine learning in a method for acquiring an actual torque value of a harmonic drive joint of a robot according to an embodiment of the present invention; as shown in fig. 1 and 2, the method includes:
s101: and acquiring a set of harmonic drive joint output torque model values according to a position signal of a motor end of the harmonic drive joint, a position signal of a connecting rod end of the harmonic drive joint and motor current data by using a harmonic drive joint flexible error model.
In particular, a formula may be used,calculating the harmonic transmission deformation angle of the harmonic transmission joint, wherein,
Δθfthe harmonic transmission deformation angle of the harmonic transmission joint is set; q. q.sdPosition signals of the connecting rod end of the harmonic drive joint; q. q.smPosition signals at the motor end; sgn () is a sign function; kωIs the stiffness coefficient of a wave generator in the harmonic drive joint; i isiIs the motor current; c. CωIs the hysteresis coefficient of the wave generator in the harmonic drive joint; l is harmonic in harmonic drive jointThe reduction ratio of the speed reducer; e is a natural base number; | is a modulo function; thetaerrThe flexible error between the output end of the connecting rod and the output end of the motor is adopted; i is the number of motor current parameters;
by means of the formula (I) and (II),calculating a compliant drive torsion angle, wherein,
delta theta is a flexible transmission torsion angle; delta thetaωA torsion angle that is an input to the wave generator;outputting a torsion angle for the wave generator; l the transmission ratio of the harmonic transmission joint speed changer;
using the formula, τf＝a1Δθ+a2Δθ2+a3Δθ3Calculating the flexible output torque of the harmonic drive joint, and further acquiring a set of model values of the output torque of the harmonic drive joint,
τfthe torque is flexibly output for the harmonic drive joint; a is1Is a preset first parameter; a is2Is a preset second parameter; a is3Is a preset third parameter;
by means of the formula (I) and (II),calculating the output torque model value of the harmonic drive joint, wherein,
τmodeloutputting a torque model value for the harmonic drive joint; tan () is a tangent function; c. CfIs a preset first constant; kfoIs a preset second constant.
In practical application, the flexible joint is a system consisting of a speed reducer and a motor, wherein the speed reducer comprises: one end of the speed reducer is connected with an output shaft of the robot joint, and the other end of the speed reducer is connected with a rotor of a motor in the robot joint; the stator of the motor is connected with the base of the robot joint.
S102: and acquiring a sample data set, training a machine learning model by using the sample data set, and acquiring a predicted value of the output torque of the harmonic drive joint according to a position signal of a motor end of the harmonic drive joint, a position signal of a connecting rod end of the harmonic drive joint and motor current data.
Specifically, the step S102 may include:
a: taking the actual output torque value of the harmonic drive joint as a target, dividing the sample data set into a plurality of subsets, and then respectively inputting each subset into each local Gaussian process machine learning model for training, wherein the sample data comprises: the position signal of the motor end of the harmonic drive joint, the position signal of the connecting rod end of the harmonic drive joint, the current data of the motor and the actual torque value of the harmonic drive joint.
In practical applications, step a may include the following steps:
a1: dividing the sample data set into a plurality of subsets, and then respectively inputting each subset into each local Gaussian process machine learning model for training, wherein the subsets are determined according to the distance between the position factor of the sample data in the subsets and the central point of the local Gaussian process machine learning model;
in practical applications, the local gaussian process machine learning model may be:
wherein,
y is a predicted value of the output torque of the harmonic drive joint; h () is a local Gaussian process machine learning model; k (X, X) is a Western difference matrix;is the variance of gaussian noise; i is a current matrix of the training sample; and X is a matrix formed by a position signal of a motor end of the harmonic drive joint, a position signal of a connecting rod end of the harmonic drive joint and motor current data contained in the training sample.
Additionally, to improve the accuracy of the trained model, the process of training the local gaussian process machine learning model may use three sets of sample data: respectively carrying out actual moment learning training on the joints by using the no-load test acquisition data, the constant load test acquisition data and the variable load test acquisition data, wherein the no-load test acquisition data can comprise 140940 training point data and 5560 measurement point data; the fixed load test collected data can comprise 136220 training point data and 5500 measuring point data; the variable load trial acquisition data may include 135720 training point data and 5000 measurement point data. In order to reduce the calculation amount during training, the input data interval corresponding to the training sample can be divided by the motor input end position signal in the training sample, namely, the input data interval is divided according to whether the distance from the motor input end position signal in the sample data in each control period to the center of the model is larger than a set value or not, and after the distance from the motor input end position signal to the center of the local model is determined, the input data interval can be divided according to wkSample data is assigned to a threshold-tolerant local model for training.
A2: aiming at each local Gaussian process machine learning model, by using a formula,calculating the mean of the predicted values, wherein,
f(x*) The mean value of the predicted values, y is the predicted value of the local Gaussian process machine learning model of the last iteration, α is a predicted vector, x is sample data;is the variance of gaussian noise; i is a current matrix of the training sample; ()TIs a transposed matrix; and K is the covariance matrix at the current iteration.
A3: by means of the formula (I) and (II),and calculating the covariance corresponding to the initial predicted value in the current iteration, wherein,
V(x*) The covariance corresponding to the initial predicted value in the current iteration; k () is a covariance calculation function; k is a radical of*The covariance of the local Gaussian process machine learning model at the last iteration; k is a covariance matrix; x is the number of*Is sample data;is the variance of gaussian noise; i is a current matrix of the training sample; ()TIs a transposed matrix;
a4: judging whether the difference between the covariance and a preset covariance and the difference between the predicted mean value and the actual moment value are both within a preset range;
a5: if so, taking the local Gaussian process machine learning model as a trained local Gaussian process machine learning model;
a6: and if not, updating the parameters of the local Gaussian process machine learning model, and returning to execute the step A2 until the trained local Gaussian process machine learning model is obtained.
B: inputting the position signal of the motor end of the harmonic drive joint, the position signal of the connecting rod end of the harmonic drive joint and the motor current data obtained in the step S102 into the local Gaussian process machine learning model obtained after training, calculating the distance from the sample data to each local Gaussian process machine learning model, and obtaining the local Gaussian process machine learning model corresponding to the maximum value of the distance;
c: acquiring a position factor of the output torque of the harmonic drive joint, and judging whether the maximum value of each distance parameter in the position factor is smaller than a preset threshold value or not;
in practical applications, a formula may be used,calculating the position factor of the output torque of the harmonic drive joint, wherein,
wkoutputting a position parameter of a torque model value relative to a kth local Gaussian process machine learning model for the harmonic drive joint; exp () is an exponential function with a natural base number as the base; x is a subset of input sample data; c. CkMachine learning a position parameter of a center point of the model for each local Gaussian process; ()TIs a transposed matrix; w is a diagonal matrix with the same width as the local Gaussian process machine learning model; k is the number of the machine learning models in the local Gaussian process;
using the formula, w (x, o) ═ w1,...,wk]And obtaining the position factor of the output torque of the harmonic drive joint, wherein,
w (x, o) is a position factor of the harmonic drive joint data moment corresponding to the input sample; x is a subset of input sample data; o is a matrix formed by the central points of the local Gaussian process machine learning models; w is a1Outputting a position parameter of a torque model value relative to a1 st local Gaussian process machine learning model for the harmonic drive joint; w is akOutputting a position parameter of a torque model value relative to a kth local Gaussian process machine learning model for the harmonic drive joint.
D: if so, updating the local Gaussian process machine learning model, and taking the central point of the updated local Gaussian process machine learning model as a new fitting point; returning to execute the step B until the judgment result of the step C is negative;
specifically, a new local gaussian process machine learning model may be obtained, and the new local gaussian process machine learning model may be used as the updated local gaussian process machine learning model. In practical applications, the new gaussian process machine learning model refers to a local gaussian process machine learning model obtained by using new model parameters.
E: and if not, calculating the mean value of the predicted values corresponding to the predicted values of the harmonic drive joint output torque according to the weighted mean value of the predicted values of the harmonic drive joint output torque output by each local Gaussian process machine learning model.
In practical applications, a formula may be used,obtaining the probability of the sample data appearing in each local Gaussian process machine learning model, wherein,
p (k | x) is the probability of the sample data x appearing in the kth local Gaussian process machine learning model; x is sample data; k is the number of local Gaussian process machine learning models; m is the number of the obtained local Gaussian process machine learning models corresponding to the harmonic drive joint output torque model value; w is akOutputting a position parameter of a torque model value relative to a kth local Gaussian process machine learning model for the harmonic drive joint;
by means of the formula (I) and (II),calculating the mean value of predicted values corresponding to the model value of the output torque of the harmonic drive joint, wherein,
outputting a predicted value mean value corresponding to the torque model value for the harmonic drive joint;the predicted value of the model is machine-learned for the kth local gaussian process.
By applying the embodiment of the invention, when the maximum value of each distance parameter in the position factor is larger than the preset threshold value, the local Gaussian process machine learning model is automatically updated, and the sample data enters the new local Gaussian process machine learning model for training. Therefore, the number of the local models is automatically increased when the sample data is complex, and the covariance inverse matrix can be updated through the distribution mode.
S103: and filtering the harmonic drive joint output torque model value and the predicted value mean value to obtain an actual torque value corresponding to the harmonic drive joint output torque model value.
Specifically, the filtering process may be to filter out discrete torque values higher than the set value.
By applying the embodiment shown in the figure 1 of the invention, the identification and measurement of the actual torque of the harmonic drive joint are realized by fusing the flexible error model and the machine learning method, and compared with the prior art which relies on the fitting parameters of the flexible error model, the calculation precision of the flexible error model for calculating the output torque of the flexible harmonic drive joint is improved.
Corresponding to the embodiment of the invention shown in fig. 1, the embodiment of the invention also provides a device for acquiring the actual torque value of the harmonic drive joint of the robot.
Fig. 3 is a schematic structural diagram of an apparatus for acquiring an actual torque value of a harmonic drive joint of a robot according to an embodiment of the present invention, and as shown in fig. 3, the apparatus includes:
the output module 301 is configured to obtain a set of harmonic drive joint output torque model values according to a position signal of a motor end of the harmonic drive joint, a position signal of a link end of the harmonic drive joint, and motor current data, by using a harmonic drive joint flexibility error model;
an obtaining module 302, configured to obtain a sample data set, train a machine learning model using the sample data set, and then obtain a predicted value of the output torque of the harmonic drive joint according to a position signal of a motor end of the harmonic drive joint, a position signal of a link end of the harmonic drive joint, and motor current data;
and the filtering module 303 is configured to perform filtering processing on the harmonic drive joint output torque model value and the predicted value mean value to obtain an actual torque value corresponding to the harmonic drive joint output torque model value.
By applying the embodiment shown in the figure 3 of the invention, the identification and measurement of the actual torque of the harmonic drive joint are realized by fusing the flexible error model and the machine learning method, and compared with the prior art which relies on the fitting parameters of the flexible error model, the calculation precision of the flexible error model for calculating the output torque of the flexible harmonic drive joint is improved.
In a specific implementation manner of the embodiment of the present invention, the output module 301 is further configured to:
by means of the formula (I) and (II),calculating the harmonic transmission deformation angle of the harmonic transmission joint, wherein,
Δθfthe harmonic transmission deformation angle of the harmonic transmission joint is set; q. q.sdPosition signals of the connecting rod end of the harmonic drive joint; q. q.smPosition signals at the motor end; sgn () is a sign function; kωIs the stiffness coefficient of a wave generator in the harmonic drive joint; i isiIs the motor current; c. CωIs the hysteresis coefficient of the wave generator in the harmonic drive joint; l is the reduction ratio of the harmonic reducer in the harmonic drive joint; e is a natural base number; | is a modulo function; thetaerrThe flexible error between the output end of the connecting rod and the output end of the motor is adopted; i is the number of motor current parameters;
by means of the formula (I) and (II),calculating a compliant drive torsion angle, wherein,
Δθa flexible drive torsion angle; delta thetaωA torsion angle that is an input to the wave generator;outputting a torsion angle for the wave generator; l the transmission ratio of the harmonic transmission joint speed changer;
using the formula, τf＝a1Δθ+a2Δθ2+a3Δθ3Calculating the flexible output torque of the harmonic drive joint, and further acquiring a set of model values of the output torque of the harmonic drive joint,
τfthe torque is flexibly output for the harmonic drive joint; a is1Is a preset first parameter; a is2Is a preset second parameter; a is3Is a preset third parameter;
by means of the formula (I) and (II),calculating the output torque model value of the harmonic drive joint, wherein,
τmodeloutputting a torque model value for the harmonic drive joint; tan () is a tangent function; c. CfIs a preset first constant; kfoIs a preset second constant.
In a specific implementation manner of the embodiment of the present invention, the obtaining module 302 is further configured to:
a: taking the actual output torque value of the harmonic drive joint as a target, dividing the sample data set into a plurality of subsets, and then respectively inputting each subset into each local Gaussian process machine learning model for training, wherein the sample data comprises: the position signal of the motor end of the harmonic drive joint, the position signal of the connecting rod end of the harmonic drive joint, the current data of the motor and the actual torque value of the harmonic drive joint;
b: inputting the position signal of the motor end of the harmonic drive joint, the position signal of the connecting rod end of the harmonic drive joint and the motor current data in the step 1) into the local Gaussian process machine learning model obtained after training, calculating the distance from the sample data to each local Gaussian process machine learning model, and obtaining the local Gaussian process machine learning model corresponding to the maximum value of the distance;
c: acquiring a position factor of the output torque of the harmonic drive joint, and judging whether the maximum value of each distance parameter in the position factor is smaller than a preset threshold value or not;
d: if so, updating the local Gaussian process machine learning model, and taking the central point of the updated local Gaussian process machine learning model as a new fitting point; returning to execute the step B until the judgment result of the step C is negative;
e: and if not, calculating the mean value of the predicted values corresponding to the predicted values of the harmonic drive joint output torque according to the weighted mean value of the predicted values of the harmonic drive joint output torque output by each local Gaussian process machine learning model.
In a specific implementation manner of the embodiment of the present invention, the obtaining module 302 is further configured to:
a1: dividing the sample data set into a plurality of subsets, and then respectively inputting each subset into each local Gaussian process machine learning model for training, wherein the subsets are determined according to the distance between the position factor of the sample data in the subsets and the central point of the local Gaussian process machine learning model;
a2: aiming at each local Gaussian process machine learning model, by using a formula,calculating the mean of the predicted values, wherein,
f(x*) Is the mean value of the predicted values, y is the predicted value of the local Gaussian process machine learning model of the last iteration, α is the predicted vector, x is*Is sample data;is the variance of gaussian noise; i is a current matrix of the training sample; ()TIs a transposed matrix; k is a covariance matrix in the current iteration;
a3: by means of the formula (I) and (II),and calculating the covariance corresponding to the initial predicted value in the current iteration, wherein,
v (x) is the covariance corresponding to the initial predicted value in the current iteration; k () is a covariance calculation function; k is the covariance of the local Gaussian process machine learning model at the last iteration; k is a covariance matrix; x is sample data;is the variance of gaussian noise; i is a current matrix of the training sample; ()TIs a transposed matrix;
a4: judging whether the difference between the covariance and a preset covariance and the difference between the predicted mean value and the actual moment value are both within a preset range;
a5: if so, taking the local Gaussian process machine learning model as a trained local Gaussian process machine learning model;
a6: and if not, updating the parameters of the local Gaussian process machine learning model, and returning to execute the step A2 until the trained local Gaussian process machine learning model is obtained.
In a specific implementation manner of the embodiment of the present invention, the obtaining module 302 is further configured to:
by means of the formula (I) and (II),calculating the position factor of the output torque of the harmonic drive joint,wherein,
wkoutputting a position parameter of a torque model value relative to a kth local Gaussian process machine learning model for the harmonic drive joint; exp () is an exponential function with a natural base number as the base; x is a subset of input sample data; c. CkMachine learning a position parameter of a center point of the model for each local Gaussian process; ()TIs a transposed matrix; w is a diagonal matrix with the same width as the local Gaussian process machine learning model; k is the number of the machine learning models in the local Gaussian process;
using the formula, w (x, o) ═ w1,...,wk]And obtaining the position factor of the output torque of the harmonic drive joint, wherein,
w (x, o) is a position factor of the harmonic drive joint data moment corresponding to the input sample; x is a subset of input sample data; o is a matrix formed by the central points of the local Gaussian process machine learning models; w is a1Outputting a position parameter of a torque model value relative to a1 st local Gaussian process machine learning model for the harmonic drive joint; w is akOutputting a position parameter of a torque model value relative to a kth local Gaussian process machine learning model for the harmonic drive joint.
In a specific implementation manner of the embodiment of the present invention, the obtaining module 302 is further configured to:
and acquiring a new local Gaussian process machine learning model, and taking the new local Gaussian process machine learning model as an updated local Gaussian process machine learning model.
In a specific implementation manner of the embodiment of the present invention, the obtaining module 302 is further configured to:
by means of the formula (I) and (II),obtaining the probability of the sample data appearing in each local Gaussian process machine learning model, wherein,
p (k | x) is the probability of the sample data x appearing in the kth local Gaussian process machine learning model; x is sample data; k is the number of local Gaussian process machine learning models; m is the number of the obtained local Gaussian process machine learning models corresponding to the harmonic drive joint output torque model value; w is akOutputting a position parameter of a torque model value relative to a kth local Gaussian process machine learning model for the harmonic drive joint;
by means of the formula (I) and (II),calculating the mean value of predicted values corresponding to the model value of the output torque of the harmonic drive joint, wherein,
outputting a predicted value mean value corresponding to the torque model value for the harmonic drive joint;the predicted value of the model is machine-learned for the kth local gaussian process.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.
Claims (14)
1. A method for acquiring an actual torque value of a harmonic drive joint of a robot is characterized by comprising the following steps:
1) acquiring a set of harmonic drive joint output torque model values according to a position signal of a motor end of a harmonic drive joint, a position signal of a connecting rod end of the harmonic drive joint and motor current data by using a harmonic drive joint flexible error model;
2) acquiring a sample data set, training a machine learning model by using the sample data set, and acquiring a predicted value of the output torque of the harmonic drive joint according to a position signal of a motor end of the harmonic drive joint, a position signal of a connecting rod end of the harmonic drive joint and motor current data;
3) and filtering the harmonic drive joint output torque model value and the predicted value mean value to obtain an actual torque value corresponding to the harmonic drive joint output torque model value.
2. The method for acquiring the actual torque value of the harmonic drive joint of the robot according to claim 1, wherein the step 1) comprises the following steps:
by means of the formula (I) and (II),calculating the harmonic transmission deformation angle of the harmonic transmission joint, wherein,
Δθfthe harmonic transmission deformation angle of the harmonic transmission joint is set; q. q.sdPosition signals of the connecting rod end of the harmonic drive joint; q. q.smPosition signals at the motor end; sgn () is a sign function; kωIs the stiffness coefficient of a wave generator in the harmonic drive joint; i isiIs the motor current; c. CωIs the hysteresis coefficient of the wave generator in the harmonic drive joint; l is the reduction ratio of the harmonic reducer in the harmonic drive joint; e is a natural base number; | is a modulo function; thetaerrThe flexible error between the output end of the connecting rod and the output end of the motor is adopted; i is the number of motor current parameters;
by means of the formula (I) and (II),calculating a compliant drive torsion angle, wherein,
delta theta is a flexible transmission torsion angle; delta thetaωA torsion angle that is an input to the wave generator;outputting a torsion angle for the wave generator; l the transmission ratio of the harmonic transmission joint speed changer;
using the formula, τf＝a1Δθ+a2Δθ2+a3Δθ3Calculating the flexible output torque of the harmonic drive joint, and further acquiring a set of model values of the output torque of the harmonic drive joint,
τfthe torque is flexibly output for the harmonic drive joint; a is1Is a preset first parameter; a is2Is a preset second parameter; a is3Is a preset third parameter;
by means of the formula (I) and (II),calculating the output torque model value of the harmonic drive joint, wherein,
τmodeloutputting a torque model value for the harmonic drive joint; tan () is a tangent function; c. CfIs a preset first constant; kfoIs a preset second constant.
3. The method for acquiring the actual torque value of the harmonic drive joint of the robot according to claim 1, wherein the step 2) comprises the following steps:
a: taking the actual output torque value of the harmonic drive joint as a target, dividing the sample data set into a plurality of subsets, and then respectively inputting each subset into each local Gaussian process machine learning model for training, wherein the sample data comprises: the position signal of the motor end of the harmonic drive joint, the position signal of the connecting rod end of the harmonic drive joint, the current data of the motor and the actual torque value of the harmonic drive joint;
b: inputting the position signal of the motor end of the harmonic drive joint, the position signal of the connecting rod end of the harmonic drive joint and the motor current data in the step 1) into the local Gaussian process machine learning model obtained after training, calculating the distance from the sample data to each local Gaussian process machine learning model, and obtaining the local Gaussian process machine learning model corresponding to the maximum value of the distance;
c: acquiring a position factor of the output torque of the harmonic drive joint, and judging whether the maximum value of each distance parameter in the position factor is smaller than a preset threshold value or not;
d: if so, updating the local Gaussian process machine learning model, and taking the central point of the updated local Gaussian process machine learning model as a new fitting point; returning to execute the step B until the judgment result of the step C is negative;
e: and if not, calculating the mean value of the predicted values corresponding to the predicted values of the harmonic drive joint output torque according to the weighted mean value of the predicted values of the harmonic drive joint output torque output by each local Gaussian process machine learning model.
4. The method for acquiring the actual torque value of the harmonic drive joint of the robot according to claim 3, wherein the step A comprises the following steps:
a1: dividing the sample data set into a plurality of subsets, and then respectively inputting each subset into each local Gaussian process machine learning model for training, wherein the subsets are determined according to the distance between the position factor of the sample data in the subsets and the central point of the local Gaussian process machine learning model;
a2: aiming at each local Gaussian process machine learning model, by using a formula,calculating the mean of the predicted values, wherein,
f(x*) Is the mean value of the predicted values, y is the predicted value of the local Gaussian process machine learning model of the last iteration, α is the predicted vector, x is*Is sample data;is the variance of gaussian noise; i is a current matrix of the training sample; ()TIs a transposed matrix; k is a covariance matrix in the current iteration;
a3: by means of the formula (I) and (II),and calculating the covariance corresponding to the initial predicted value in the current iteration, wherein,
V(x*) The covariance corresponding to the initial predicted value in the current iteration; k () is a covariance calculation function; k is a radical of*The covariance of the local Gaussian process machine learning model at the last iteration; k is a covariance matrix; x is the number of*Is sample data;is the variance of gaussian noise; i is a current matrix of the training sample; ()TIs a transposed matrix;
a4: judging whether the difference between the covariance and a preset covariance and the difference between the predicted mean value and the actual moment value are both within a preset range;
a5: if so, taking the local Gaussian process machine learning model as a trained local Gaussian process machine learning model;
a6: and if not, updating the parameters of the local Gaussian process machine learning model, and returning to execute the step A2 until the trained local Gaussian process machine learning model is obtained.
5. The method for acquiring the actual torque value of the harmonic drive joint of the robot as claimed in claim 3, wherein acquiring the position factor of the output torque of the harmonic drive joint comprises:
by means of the formula (I) and (II),calculating the position factor of the output torque of the harmonic drive joint, wherein,
wkoutputting a position parameter of a torque model value relative to a kth local Gaussian process machine learning model for the harmonic drive joint; exp () is an exponential function with a natural base number as the base; x is a subset of input sample data; c. CkIn machine learning models for respective local Gaussian processesA location parameter of the heart point; ()TIs a transposed matrix; w is a diagonal matrix with the same width as the local Gaussian process machine learning model; k is the number of the machine learning models in the local Gaussian process;
using the formula, w (x, o) ═ w1,...,wk]And obtaining the position factor of the output torque of the harmonic drive joint, wherein,
w (x, o) is a position factor of the harmonic drive joint data moment corresponding to the input sample; x is a subset of input sample data; o is a matrix formed by the central points of the local Gaussian process machine learning models; w is a1Outputting a position parameter of a torque model value relative to a1 st local Gaussian process machine learning model for the harmonic drive joint; w is akOutputting a position parameter of a torque model value relative to a kth local Gaussian process machine learning model for the harmonic drive joint.
6. The method for acquiring the actual torque value of the harmonic drive joint of the robot according to claim 3, wherein the updating the local Gaussian process machine learning model comprises:
and acquiring a new local Gaussian process machine learning model, and taking the new local Gaussian process machine learning model as an updated local Gaussian process machine learning model.
7. The method for acquiring the actual torque value of the harmonic drive joint of the robot as claimed in claim 3, wherein the step E comprises:
by means of the formula (I) and (II),obtaining the probability of the sample data appearing in each local Gaussian process machine learning model, wherein,
p (k | x) is the probability of the sample data x appearing in the kth local Gaussian process machine learning model; x is sample data; k is the number of local Gaussian process machine learning models; m is the output force obtained by the harmonic drive jointThe number of local Gaussian process machine learning models corresponding to the moment model value; w is akOutputting a position parameter of a torque model value relative to a kth local Gaussian process machine learning model for the harmonic drive joint;
by means of the formula (I) and (II),calculating the mean value of predicted values corresponding to the model value of the output torque of the harmonic drive joint, wherein,
outputting a predicted value mean value corresponding to the torque model value for the harmonic drive joint;the predicted value of the model is machine-learned for the kth local gaussian process.
8. An acquisition device for actual torque values of harmonic drive joints of a robot is characterized by comprising:
the output module is used for acquiring a set of harmonic drive joint output torque model values according to a position signal of a motor end of the harmonic drive joint, a position signal of a connecting rod end of the harmonic drive joint and motor current data by using a harmonic drive joint flexible error model;
the acquisition module is used for acquiring a sample data set, training a machine learning model by using the sample data set, and acquiring a predicted value of the output torque of the harmonic drive joint according to a position signal of a motor end of the harmonic drive joint, a position signal of a connecting rod end of the harmonic drive joint and motor current data;
and the filtering module is used for filtering the harmonic drive joint output torque model value and the predicted value mean value to obtain an actual torque value corresponding to the harmonic drive joint output torque model value.
9. The device for acquiring the actual torque value of the harmonic drive joint of the robot according to claim 8, wherein the output module is further configured to:
by means of the formula (I) and (II),calculating the harmonic transmission deformation angle of the harmonic transmission joint, wherein,
Δθfthe harmonic transmission deformation angle of the harmonic transmission joint is set; q. q.sdPosition signals of the connecting rod end of the harmonic drive joint; q. q.smPosition signals at the motor end; sgn () is a sign function; kωIs the stiffness coefficient of a wave generator in the harmonic drive joint; i isiIs the motor current; c. CωIs the hysteresis coefficient of the wave generator in the harmonic drive joint; l is the reduction ratio of the harmonic reducer in the harmonic drive joint; e is a natural base number; | is a modulo function; thetaerrThe flexible error between the output end of the connecting rod and the output end of the motor is adopted; i is the number of motor current parameters;
by means of the formula (I) and (II),calculating a compliant drive torsion angle, wherein,
delta theta is a flexible transmission torsion angle; delta thetaωA torsion angle that is an input to the wave generator;outputting a torsion angle for the wave generator; l the transmission ratio of the harmonic transmission joint speed changer;
using the formula, τf＝a1Δθ+a2Δθ2+a3Δθ3Calculating the flexible output torque of the harmonic drive joint, and further acquiring a set of model values of the output torque of the harmonic drive joint,
τfthe torque is flexibly output for the harmonic drive joint; a is1Is a preset first parameter; a is2Is a preset second parameter; a is3Is a preset third parameter;
by means of the formula (I) and (II),calculating the output torque model value of the harmonic drive joint, wherein,
τmodeloutputting a torque model value for the harmonic drive joint; tan () is a tangent function; c. CfIs a preset first constant; kfoIs a preset second constant.
10. The device for acquiring the actual torque value of the harmonic drive joint of the robot according to claim 8, wherein the acquiring module is further configured to:
a: taking the actual output torque value of the harmonic drive joint as a target, dividing the sample data set into a plurality of subsets, and then respectively inputting each subset into each local Gaussian process machine learning model for training, wherein the sample data comprises: the position signal of the motor end of the harmonic drive joint, the position signal of the connecting rod end of the harmonic drive joint, the current data of the motor and the actual torque value of the harmonic drive joint;
b: inputting the position signal of the motor end of the harmonic drive joint, the position signal of the connecting rod end of the harmonic drive joint and the motor current data in the step 1) into the local Gaussian process machine learning model obtained after training, calculating the distance from the sample data to each local Gaussian process machine learning model, and obtaining the local Gaussian process machine learning model corresponding to the maximum value of the distance;
c: acquiring a position factor of the output torque of the harmonic drive joint, and judging whether the maximum value of each distance parameter in the position factor is smaller than a preset threshold value or not;
d: if so, updating the local Gaussian process machine learning model, and taking the central point of the updated local Gaussian process machine learning model as a new fitting point; returning to execute the step B until the judgment result of the step C is negative;
e: and if not, calculating the mean value of the predicted values corresponding to the predicted values of the harmonic drive joint output torque according to the weighted mean value of the predicted values of the harmonic drive joint output torque output by each local Gaussian process machine learning model.
11. The device for acquiring the actual torque value of the harmonic drive joint of the robot according to claim 10, wherein the acquiring module is further configured to:
a1: dividing the sample data set into a plurality of subsets, and then respectively inputting each subset into each local Gaussian process machine learning model for training, wherein the subsets are determined according to the distance between the position factor of the sample data in the subsets and the central point of the local Gaussian process machine learning model;
a2: aiming at each local Gaussian process machine learning model, by using a formula,calculating the mean of the predicted values, wherein,
f(x*) Is the mean value of the predicted values, y is the predicted value of the local Gaussian process machine learning model of the last iteration, α is the predicted vector, x is*Is sample data;is the variance of gaussian noise; i is a current matrix of the training sample; ()TIs a transposed matrix; k is a covariance matrix in the current iteration;
a3: by means of the formula (I) and (II),and calculating the covariance corresponding to the initial predicted value in the current iteration, wherein,
V(x*) The covariance corresponding to the initial predicted value in the current iteration; k () is a covariance calculation function; k is a radical of*The covariance of the local Gaussian process machine learning model at the last iteration; k is a covariance matrix; x is the number of*Is sample data;is the variance of gaussian noise; i is a current matrix of the training sample; ()TIs a transposed matrix;
a4: judging whether the difference between the covariance and a preset covariance and the difference between the predicted mean value and the actual moment value are both within a preset range;
a5: if so, taking the local Gaussian process machine learning model as a trained local Gaussian process machine learning model;
a6: and if not, updating the parameters of the local Gaussian process machine learning model, and returning to execute the step A2 until the trained local Gaussian process machine learning model is obtained.
12. The device for acquiring the actual torque value of the harmonic drive joint of the robot according to claim 10, wherein the acquiring module is further configured to:
by means of the formula (I) and (II),calculating the position factor of the output torque of the harmonic drive joint, wherein,
wkoutputting a position parameter of a torque model value relative to a kth local Gaussian process machine learning model for the harmonic drive joint; exp () is an exponential function with a natural base number as the base; x is a subset of input sample data; c. CkMachine learning a position parameter of a center point of the model for each local Gaussian process; ()TIs a transposed matrix; w is a diagonal matrix with the same width as the local Gaussian process machine learning model; k is the number of the machine learning models in the local Gaussian process;
using the formula, w (x, o) ═ w1,...,wk]And obtaining the position factor of the output torque of the harmonic drive joint, wherein,
w (x, o) is a position factor of the harmonic drive joint data moment corresponding to the input sample; x is a subset of input sample dataCombining; o is a matrix formed by the central points of the local Gaussian process machine learning models; w is a1Outputting a position parameter of a torque model value relative to a1 st local Gaussian process machine learning model for the harmonic drive joint; w is akOutputting a position parameter of a torque model value relative to a kth local Gaussian process machine learning model for the harmonic drive joint.
13. The device for acquiring the actual torque value of the harmonic drive joint of the robot according to claim 10, wherein the acquiring module is further configured to:
and acquiring a new local Gaussian process machine learning model, and taking the new local Gaussian process machine learning model as an updated local Gaussian process machine learning model.
14. The method for acquiring the actual torque value of the harmonic drive joint of the robot according to claim 10, wherein the acquiring module is further configured to:
by means of the formula (I) and (II),obtaining the probability of the sample data appearing in each local Gaussian process machine learning model, wherein,
p (k | x) is the probability of the sample data x appearing in the kth local Gaussian process machine learning model; x is sample data; k is the number of local Gaussian process machine learning models; m is the number of the obtained local Gaussian process machine learning models corresponding to the harmonic drive joint output torque model value; w is akOutputting a position parameter of a torque model value relative to a kth local Gaussian process machine learning model for the harmonic drive joint;
by means of the formula (I) and (II),calculating the mean value of predicted values corresponding to the model value of the output torque of the harmonic drive joint, wherein,
outputting a predicted value mean value corresponding to the torque model value for the harmonic drive joint;the predicted value of the model is machine-learned for the kth local gaussian process.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811587023.1A CN109623819B (en) | 2018-12-25 | 2018-12-25 | Method and device for acquiring actual torque value of harmonic drive joint of robot |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811587023.1A CN109623819B (en) | 2018-12-25 | 2018-12-25 | Method and device for acquiring actual torque value of harmonic drive joint of robot |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109623819A true CN109623819A (en) | 2019-04-16 |
| CN109623819B CN109623819B (en) | 2020-08-21 |
Family
ID=66077238
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811587023.1A Active CN109623819B (en) | 2018-12-25 | 2018-12-25 | Method and device for acquiring actual torque value of harmonic drive joint of robot |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109623819B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112276935A (en) * | 2019-07-22 | 2021-01-29 | 发那科株式会社 | Position and orientation determination device, position and orientation determination method, and position and orientation determination program |
| CN112388628A (en) * | 2019-08-13 | 2021-02-23 | 罗伯特·博世有限公司 | Apparatus and method for training a gaussian process regression model |
| CN112847365A (en) * | 2021-01-07 | 2021-05-28 | 西安电子科技大学 | Torque estimation method |
| IT202100030044A1 (en) | 2021-11-26 | 2023-05-26 | Ergotech Srl | NEW HARMONIC REDUCER WITH COMPACT STRUCTURE MADE WITH PLASTIC MATERIAL |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101122777A (en) * | 2007-09-18 | 2008-02-13 | 湖南大学 | Large condenser underwater operation environment two-joint robot control method |
| CN102626930A (en) * | 2012-04-28 | 2012-08-08 | 哈尔滨工业大学 | Mechanical arm modular joint with power-off brake and multiple perceptive functions |
| CN106873383A (en) * | 2017-04-17 | 2017-06-20 | 珞石（北京）科技有限公司 | A kind of On-Line Control Method for reducing industrial robot vibration |
| US9694495B1 (en) * | 2013-06-24 | 2017-07-04 | Redwood Robotics Inc. | Virtual tools for programming a robot arm |
| CN106994686A (en) * | 2016-12-01 | 2017-08-01 | 遨博（北京）智能科技有限公司 | The computational methods and device, robot of joint external force torque |
| CN108284442A (en) * | 2017-01-24 | 2018-07-17 | 中国北方车辆研究所 | A kind of mechanical arm flexible joint control method based on fuzzy neural network |
| CN108896304A (en) * | 2018-07-19 | 2018-11-27 | 中科新松有限公司 | Harmonic reducer of robot test device and system |
| CN108994837A (en) * | 2018-08-20 | 2018-12-14 | 哈工大机器人（合肥）国际创新研究院 | A kind of mechanical arm zero-g balance control method of Dynamics Compensation |
-
2018
- 2018-12-25 CN CN201811587023.1A patent/CN109623819B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101122777A (en) * | 2007-09-18 | 2008-02-13 | 湖南大学 | Large condenser underwater operation environment two-joint robot control method |
| CN102626930A (en) * | 2012-04-28 | 2012-08-08 | 哈尔滨工业大学 | Mechanical arm modular joint with power-off brake and multiple perceptive functions |
| US9694495B1 (en) * | 2013-06-24 | 2017-07-04 | Redwood Robotics Inc. | Virtual tools for programming a robot arm |
| CN106994686A (en) * | 2016-12-01 | 2017-08-01 | 遨博（北京）智能科技有限公司 | The computational methods and device, robot of joint external force torque |
| CN108284442A (en) * | 2017-01-24 | 2018-07-17 | 中国北方车辆研究所 | A kind of mechanical arm flexible joint control method based on fuzzy neural network |
| CN106873383A (en) * | 2017-04-17 | 2017-06-20 | 珞石（北京）科技有限公司 | A kind of On-Line Control Method for reducing industrial robot vibration |
| CN108896304A (en) * | 2018-07-19 | 2018-11-27 | 中科新松有限公司 | Harmonic reducer of robot test device and system |
| CN108994837A (en) * | 2018-08-20 | 2018-12-14 | 哈工大机器人（合肥）国际创新研究院 | A kind of mechanical arm zero-g balance control method of Dynamics Compensation |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112276935A (en) * | 2019-07-22 | 2021-01-29 | 发那科株式会社 | Position and orientation determination device, position and orientation determination method, and position and orientation determination program |
| CN112388628A (en) * | 2019-08-13 | 2021-02-23 | 罗伯特·博世有限公司 | Apparatus and method for training a gaussian process regression model |
| CN112847365A (en) * | 2021-01-07 | 2021-05-28 | 西安电子科技大学 | Torque estimation method |
| CN112847365B (en) * | 2021-01-07 | 2022-08-02 | 西安电子科技大学 | Torque estimation method |
| IT202100030044A1 (en) | 2021-11-26 | 2023-05-26 | Ergotech Srl | NEW HARMONIC REDUCER WITH COMPACT STRUCTURE MADE WITH PLASTIC MATERIAL |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109623819B (en) | 2020-08-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109623819B (en) | Method and device for acquiring actual torque value of harmonic drive joint of robot | |
| CN104378038B (en) | Permanent magnet synchronous motor parameter identification method based on artificial neural network | |
| CN109249397B (en) | Six-degree-of-freedom robot dynamics parameter identification method and system | |
| CN103331756A (en) | Mechanical arm motion control method | |
| CN110941183B (en) | Industrial robot dynamics identification method based on neural network | |
| CN110281237B (en) | Series robot joint friction force identification method based on machine learning | |
| CN110962124B (en) | Method for compensating static errors and correcting dynamic stiffness model of cutting machining robot | |
| CN106125548A (en) | Industrial robot kinetic parameters discrimination method | |
| CN113119111A (en) | Mechanical arm and track planning method and device thereof | |
| US11982996B2 (en) | Method and apparatus for configuring processing parameters of production equipment, and computer-readable medium | |
| CN109873587B (en) | Automatic multi-parameter identification method for permanent magnet synchronous motor | |
| CN107861061B (en) | Data-driven induction motor parameter online identification method | |
| CN110276139A (en) | A kind of induction machine key parameter acquisition methods for firmly believing Policy-Gradient algorithm based on depth | |
| CN102252126A (en) | Method for identifying parameters of servo object in electro-hydraulic angular displacement servo system | |
| CN115741718B (en) | Robot complete zero-force control method and system | |
| CN103679639B (en) | Image denoising method and device based on non-local mean value | |
| CN113285647A (en) | Permanent magnet synchronous motor feedback adjustment method and device and permanent magnet synchronous motor | |
| CN103760827A (en) | Saltus constrained off-line planning method for numerical control machining feed rate | |
| CN114260892B (en) | Elastic joint moment control method and device, readable storage medium and robot | |
| CN116659803A (en) | Method for acquiring aerodynamic load of continuous wind tunnel based on balance zero point on-line monitoring | |
| CN109764876B (en) | Multi-mode fusion positioning method of unmanned platform | |
| CN111103846A (en) | Numerical control machine tool state prediction method based on time sequence | |
| CN109150049A (en) | A kind of rest frame motor distributed parameter model method for building up | |
| CN108015761B (en) | Single-connecting-rod flexible mechanical arm control method and system | |
| CN102141172A (en) | Device and method for identifying parameter of actuating mechanism in electrohydraulic linear displacement servo system |
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 | ||
| PP01 | Preservation of patent right |
Effective date of registration: 20240626 Granted publication date: 20200821 |
|
| PP01 | Preservation of patent right |