CN112327938A - Robot near-zero following error control method based on data driving - Google Patents
Robot near-zero following error control method based on data driving Download PDFInfo
- Publication number
- CN112327938A CN112327938A CN202011089607.3A CN202011089607A CN112327938A CN 112327938 A CN112327938 A CN 112327938A CN 202011089607 A CN202011089607 A CN 202011089607A CN 112327938 A CN112327938 A CN 112327938A
- Authority
- CN
- China
- Prior art keywords
- robot
- track
- data
- neural network
- following
- 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
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D1/00—Control of position, course or altitude of land, water, air, or space vehicles, e.g. automatic pilot
- G05D1/12—Target-seeking control
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention discloses a robot near-zero following error control method based on data driving, which belongs to the field of industrial robot control and comprises the following steps: extracting target attributes influencing the following errors of the robot from the online process data, inputting the target attributes into an online deep neural network model, obtaining a position compensation value corresponding to each discrete point on a reference track at the tail end of the robot in each control period, and mapping the position compensation values into angle compensation quantities of each joint of the robot; compensating the input of the variable parameter self-adaptive controller by using the angle compensation quantity to obtain a target control quantity, and sending the target control quantity to the robot so as to enable the robot to move along a reference track and complete the control of the current control period; and in the whole track following process of the robot, dynamically adjusting the parameters of the online deep neural network model according to a preset structure updating rule. The invention can adapt to the change of working conditions and process the track following data stream in real time, achieves the control effect of near-zero following error and effectively improves the precision of the track control of the robot.
Description
Technical Field
The invention belongs to the field of industrial robot control, and particularly relates to a data-driven robot near-zero following error control method.
Background
When complex multi-curved-surface workpieces are ground, polished, welded and the like, in order to obtain a good machining effect, related machining processes are often completed by means of robots. Because the workpieces are frequently used under the working conditions of humidity, corrosion, high temperature, high speed or variable load, the processing quality of the workpieces is an important factor influencing the working life and the working limit of the workpieces, and the track precision of the robot is a key influencing the processing quality in the process of processing by using the robot. Take laser beam machining as an example, complicated many curved surfaces work piece is because one shot forming difficulty and quality are not good, and welding is used commonly to process manufacturing, in order to guarantee welding quality, avoids welding hot crack and hole, often comes the strict control welding input through laser welding robot, accomplishes robot machining. In the welding process of the laser welding robot, the track precision of the robot is the key influencing the welding quality of the robot, but due to the influences of factors such as working environment interference, load change, operation joint aging, driver saturation and the like, and in addition, the track deviation error caused by workpiece thermal deformation and the cascade error generated by joint angle following, the running track of a welding gun has larger position deviation with a welding seam, and the phenomena of welding opacity and welding inaccuracy occur, so that the yield of the workpieces and the service performance drop suddenly.
When a robot is used for processing a complex multi-curved-surface workpiece, an internal control mechanism is a fundamental factor for determining the control precision of a robot controller, and the realization of high-precision control generally comprises the design of the controller and a compensator. The internal structure design of the controller and the adaptive optimization of the control parameters can ensure that the system is in an optimal or suboptimal working state, and the method is a common method for improving the control precision. However, the method has certain limitations, for example, adaptive control generally cannot achieve control indexes, and although iterative learning control has a simple structure, effectiveness is only limited to repetitive tasks and learning gain is difficult to adapt to external disturbance. In addition, even if the controller can correct the control rule or the controller parameters according to the information obtained by comparing the required performance index with the performance index of the actual system under the interference effect, the optimal or suboptimal working state is ensured, asymptotic steady-state errors generated by high-frequency and low-frequency unmodeled dynamics still exist at the track boundary, the processing effect is worsened, and the service life of the workpiece is further reduced. Under the condition that the following error can not be effectively reduced after the parameters of the controller are continuously optimized and the control rule is modified, the mode of adopting the additional compensator becomes an effective means for further improving the track precision. However, some existing compensation methods still have great limitations, for example, a model-based compensation method is highly dependent on models and expert experience, and improper compensation easily causes system oscillation or compensation failure.
In general, for the application occasions requiring high-precision track control, such as complex multi-curved surface workpiece processing, the track precision obtained by the existing control method still needs to be further improved.
Disclosure of Invention
Aiming at the defects and the improvement requirements of the prior art, the invention provides a data-driven robot near-zero following error control method, which aims to improve the precision of robot track control.
To achieve the above object, according to an aspect of the present invention, there is provided a data-driven-based near-zero following error control method for a robot, including:
extracting target attributes of the robot from the online process data, inputting the target attributes into an online deep neural network model, obtaining a position compensation value corresponding to each discrete point on a reference track at the tail end of the robot in each control period, and mapping the position compensation values into angle compensation quantities of each joint of the robot; compensating the input of the variable parameter self-adaptive controller by using the angle compensation quantity to obtain a target control quantity, and sending the target control quantity to the robot so as to enable the robot to move along a reference track and complete the control of the current control period;
in the whole track following process of the robot, dynamically adjusting the parameters of the online deep neural network model according to a preset structure updating rule;
the target attribute is the attribute which can most represent the influence on the following error in the process data for controlling the robot track to follow; the online deep neural network model is a multi-input multi-output deep neural network and is used for predicting a position compensation value corresponding to each discrete point on the reference track according to the attribute of the robot; according to the structure updating rule, the parameters of the online deep neural network model dynamically change along with the change of the working condition and the data quantity accumulated in the following process.
The invention predicts the position compensation value corresponding to each discrete point on the reference track of the robot tail end by using the online deep neural network model, and in the following process of the robot track, the parameters of the online deep neural network model can dynamically change along with the change of the working condition and the data amount accumulated in the following process, thereby being capable of adapting to the change of the working condition and processing the track following data stream in real time, always keeping the tail end of the robot to be consistent with the reference track, achieving the near-zero following error control effect, solving the problem of processing position deviation caused by uncertain factors such as controller saturation, temperature change, environment change, operation joint aging and the like, and effectively improving the precision of the robot track control.
And further, dynamically adjusting parameters of the online deep neural network model, and completing the adjustment through a chicken flock optimization algorithm.
Compared with other parameter optimization methods such as a particle swarm optimization algorithm and a bat optimization algorithm, the chicken flock optimization algorithm has stronger global search capability; the parameters of the on-line deep neural network model are dynamically adjusted through the chicken flock optimization algorithm, so that the global optimal solution can be obtained.
Further, the parameters of the online deep neural network model are dynamically adjusted through a chicken flock optimization algorithm, so that the constraint of the Lyapunov energy function is met;
the lyapunov energy function is defined as follows:
wherein x ise、ye、zeIs the positional deviation of the tail end of the robot,
is the derivative of V with respect to time; if and only if xe＝ye＝zeWhen the content is equal to 0, the content,
the addition of the compensation value can bring certain influence to the stability of the system, and particularly under the condition that the track of the curved surface processing is time-varying with the working condition, the stability of the system is ensured by constructing a proper Lyapunov energy function and automatically adjusting the parameters of an online deep neural network model, so that the instability states of oscillation, divergence and the like are not generated while the robot realizes high-precision track following.
Further, the obtaining mode of the structure updating rule includes:
testing the track following effects of different robots on various different tracks to obtain an online deep neural network model structure optimization method for processing track following dynamic data streams under each track, and determining the optimization method capable of obtaining the optimal control compensation value as a structure updating rule.
According to the invention, the structure updating rule is determined by testing the track following effects of different robots on various different tracks, the structure updating rule applicable to different working conditions and different tracks can be obtained, and the adaptability of the control method to data streams is ensured.
Further, the method for controlling the near-zero following error of the robot based on data driving provided by the invention further comprises the following steps before sending the target control quantity to the robot: constructing a decoupling model between the position and the speed of the tail end of the robot, and determining an advance corresponding to a target control quantity by using the decoupling model;
the timing for transmitting the target control amount to the robot is determined by the lead amount.
According to the method, the corresponding lead amount of the compensation value is obtained according to the decoupling model between the position and the speed, so that time lag errors and lead errors caused by abrupt change of the track curvature can be avoided, the situation of lagging following or overshooting is avoided, and the track following precision is further ensured.
Further, the decoupling model satisfies the following relationship:
δF＝g(f(s),qE,J,δG)；
wherein, deltaFRepresenting the amount of lead, f(s) being the reference track, qEFor the pose of the robot tip, J is the Jacobian matrix describing the transformation between the robot tip velocity and joint velocity, deltaGAnd g is a lead quantity constraint relation for a compensation value output by the online deep neural network.
In some alternative embodiments, the following linear combination relationship is satisfied between the input and the output of the lead amount constraint relationship g:
δF＝Adf(s)+BdqE+CdJ+DδG；
where D represents the difference between the current control period and the previous control period, and A, B, C and D are 4 coefficients in a linear combination relationship.
Further, the process of obtaining the control quantity is controlled in a variable parameter self-adaptive mode, and the constraint of the Lyapunov energy function is met;
the lyapunov energy function is defined as follows:
wherein x ise、ye、zeIs the positional deviation of the tail end of the robot,
is the derivative of V with respect to time; if and only if xe＝ye＝zeWhen the content is equal to 0, the content,
the addition of the compensation value can bring certain influence to the stability of the system, and particularly under the condition that the track of the curved surface processing is time-varying with the working condition, the stability of the system is ensured by constructing a proper Lyapunov energy function and optimizing a variable parameter self-adaptive control algorithm, so that the instability states of oscillation, divergence and the like are not generated while the robot realizes high-precision track following.
Further, the method for controlling the near-zero following error of the robot based on data driving provided by the invention further comprises the following steps: correcting the coefficients in the mapping function by combining an experimental method; and the mapping function is used for mapping the position compensation values corresponding to the discrete points on the reference track into angle compensation quantities of each joint of the robot.
Due to joint cascade and error coupling, the mapping relation between the position compensation value and each joint angle compensation value obtained by a theoretical derivation mode still has the problem of under-compensation or over-compensation.
According to another aspect of the present invention, there is provided a computer-readable storage medium, which includes a stored computer program, and when the computer program is executed by a processor, the computer-readable storage medium controls an apparatus to execute the method for controlling a near-zero following error of a robot based on data driving according to the present invention.
Generally, by the above technical solution conceived by the present invention, the following beneficial effects can be obtained:
(1) the invention predicts the position compensation value corresponding to each discrete point on the reference track of the robot tail end by using the online deep neural network model, and in the following process of the robot track, the parameters of the online deep neural network model can dynamically change along with the change of the working condition and the data amount accumulated in the following process, thereby being capable of adapting to the change of the working condition and processing the track following data stream in real time, always keeping the tail end of the robot to be consistent with the reference track, achieving the near-zero following error control effect, solving the problem of processing position deviation caused by uncertain factors such as controller saturation, temperature change, environment change, operation joint aging and the like, and effectively improving the precision of the robot track control.
(2) According to the method, the corresponding lead amount of the compensation value is obtained according to the decoupling model between the position and the speed, so that time lag errors and lead errors caused by abrupt change of the track curvature can be avoided, the situation of lagging following or overshooting is avoided, and the track following precision is further ensured.
(3) According to the invention, by constructing a proper Lyapunov energy function, the parameters of the online deep neural network model are automatically adjusted and the variable parameter adaptive control algorithm is optimized to ensure the stability of the system, and the robot can be ensured not to generate unstable states such as oscillation and divergence while realizing high-precision track following.
(4) According to the method, the correction coefficient is introduced to correct the mapping function in combination with an experimental method, so that the difference between the theoretical mapping relation and the actual mapping relation can be reduced, and the track following precision is further improved.
Drawings
Fig. 1 is a schematic diagram of a near-zero following error control method for a robot based on data driving according to an embodiment of the present invention;
fig. 2 is a schematic diagram of a dynamic data stream processing process according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
In the present application, the terms "first," "second," and the like (if any) in the description and the drawings are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order.
Aiming at the technical problem that the precision of the existing track following control method is not high, the invention provides a data-driven robot near-zero following error method, which has the overall thought that: considering that a large amount of key information related to the operation process and the self state is constantly generated and stored in the robot processing process, the controller is ensured to work in the optimal or suboptimal state, and meanwhile, the online process data flow is used for driving and generating an effective compensation value so as to accurately track and predict errors in advance; the high precision in the track following process is ensured by compensating the input quantity, the near-zero following error control of each joint is realized, the near-zero following is realized by cascading, the influence of local defocusing or machining position deviation brought by the following error to the machining of a complex curved surface is greatly reduced, and the machining quality is improved.
Without loss of generality, the following embodiments take the track following of a laser welding robot as an example, and further explain the technical scheme of the invention. The following are examples.
Example 1:
a method for controlling a near-zero following error of a robot based on data driving, as shown in fig. 1, includes:
extracting target attributes of the robot from the online process data, inputting the target attributes into an online deep neural network model, obtaining a position compensation value corresponding to each discrete point on a reference track at the tail end of the robot in each control period, and mapping the position compensation values into angle compensation quantities of each joint of the robot; compensating the input of the variable parameter self-adaptive controller by using the angle compensation quantity to obtain a target control quantity, and sending the target control quantity to the robot so as to enable the robot to move along a reference track and complete the control of the current control period;
in the whole track following process of the robot, dynamically adjusting the parameters of the online deep neural network model according to a preset structure updating rule;
the target attribute is the attribute which can most represent the influence on the following error in the process data for controlling the robot track to follow; the online deep neural network model is a multi-input multi-output deep neural network and is used for predicting a position compensation value corresponding to each discrete point on the reference track according to the attribute of the robot; according to the structure updating rule, the parameters of the online deep neural network model dynamically change along with the change of the working condition and the data quantity accumulated in the following process;
in the embodiment, the position compensation value corresponding to each discrete point on the reference track of the tail end of the robot is predicted by using the online deep neural network model, and in the following process of the robot, the parameters of the online deep neural network model can dynamically change along with the change of the working condition and the data quantity accumulated in the following process, so that the change of the working condition can be adapted, the track following data stream can be processed in real time, the tail end of the robot is always kept to be consistent with the reference track, the near-zero following error control effect is achieved, the problem of processing position deviation caused by uncertain factors such as controller saturation, temperature change, environment change, operation joint aging and the like is solved, and the precision of robot track control is effectively improved; meanwhile, the embodiment integrates the online deep neural network model and the variable parameter adaptive control to realize the near-zero following error control of the robot, and the variable parameter adaptive control method used in the embodiment is a commonly used adaptive control method in the robot following control, and a specific control principle thereof will not be described herein again.
Optionally, in this embodiment, the on-line process data generated by the welding robot during the welding process may be collected by using a conventional data collection device; the target attribute input into the online deep neural network model, namely the attribute which can most characterize and influence the following error in the online process data, can be determined by using a Principal Component Analysis (PCA) method, and the specific process is as follows:
firstly, normalizing related original data influencing the following error of the welding robot, such as a reference track of the tail end of the robot, parameters of a controller, real-time angles, speeds, accelerations and the like of joints of the robot, the temperature of a welding gun, parameters of a motor (such as time constants and torque fluctuation coefficients), parameters of a driver (such as proportional gain and acceleration and deceleration time), environmental changes (such as temperature, humidity and pressure) and the like, so as to ensure that dimensions of different attribute data are uniform, and a calculation formula is as follows:
wherein x isiIs a column vector representing data for attribute items numbered i (e.g., real-time angle x)1)；
Is a scalar quantity representing the mean value of the attribute item data numbered i; sigmaiIs a scalar quantity representing the standard deviation of the attribute item data numbered i; x is the number ofi' is a column vector, which is the result of the calculation after normalization.
Then, a covariance matrix among the attribute items is calculated, and an eigenvalue and an eigenvector of the obtained covariance matrix are extracted, wherein the calculation formula is as follows:
|λX-CX|＝0
(λXI-CX)vX＝0
wherein n is the total number of attributes influencing the following error of the welding robot; x ═ X1 x2 … xn]，XTIs the transpose of X; cXIs a covariance matrix, I is a unit matrix, λXIs an eigenvalue, v, of a covariance matrixXIs the eigenvector of the covariance matrix.
Finally, the obtained eigenvalue λ will be calculatedXSorting, and selecting the first k (k is more than 0 and less than or equal to n) maximum attribute items as the final input items of the online deep neural network model, namely the target attributes of the welding robot;
in the embodiment, only the attribute which can most represent the influence of the following error is taken as the input item of the online deep neural network model, the redundant items of the track following precision influence factors can be eliminated, the data dimension of the input layer of the online deep neural network model is simplified, the consumption of data processing and data storage of the online deep neural network model is reduced, the response characteristic of the system is improved, and the weld joint is prevented from being heated at high temperature continuously due to time lag;
after determining the parameters of the input online deep neural network model based on the above method, a process of dynamically processing a data stream using the online deep neural network model is shown in fig. 2.
As an optional implementation manner, in order to avoid gradient disappearance, invalid output caused by processing delay, and the like, the parameters of the online deep neural network model are dynamically adjusted, such as the selection of the initial values and the number of nodes of the network node, and the adjustment of the number of network layers and the structural weight, and the adjustment is completed through a chicken flock optimization algorithm;
compared with other parameter optimization methods such as a particle swarm optimization algorithm and a bat optimization algorithm, the chicken flock optimization algorithm has stronger global search capability; in the embodiment, the parameters of the online deep neural network model are dynamically adjusted through a chicken flock optimization algorithm, so that a global optimal solution is obtained.
The addition of the compensation value can bring certain influence on the stability of the system, especially under the condition that the track of curved surface processing and the working condition are time-varying; in order to ensure the stability of the system, as an optimal implementation manner, in this embodiment, a process of dynamically adjusting parameters of an online deep neural network model through a chicken flock optimization algorithm satisfies the constraint of a lyapunov energy function;
the lyapunov energy function is defined as follows:
wherein x ise、ye、zeIs the positional deviation of the tail end of the robot,
is the derivative of V with respect to time; if and only if xe＝ye＝zeWhen the content is equal to 0, the content,
according to the embodiment, the stability of the system is ensured by constructing a proper Lyapunov energy function and automatically adjusting the parameters of the online deep neural network model, and unstable states such as oscillation and divergence are not generated when the robot realizes high-precision track following.
The high efficiency of control is kept through the structural optimization of the online deep neural network model, and the overall welding precision is maintained; as an optional implementation manner, in this embodiment, the obtaining manner of the structure update rule includes:
testing the track following effects of different robots on various different tracks to obtain an online deep neural network model structure optimization method for processing track following dynamic data streams under each track, and determining the optimization method capable of obtaining the optimal control compensation value as a structure updating rule;
in the embodiment, the structure updating rules are determined by testing the track following effects of different robots on various different tracks, so that the structure updating rules applicable to different working conditions and different tracks can be obtained, and particularly, in the initial track following stage, the data volume is small, and a network model with a simple structure and a simple hierarchy is selected to avoid the local optimal problem caused by time lag and supersaturation; along with the accumulation of data quantity, the original model learning effect is reduced and the performance is insufficient, the optimization and adjustment rule of the network structure is determined according to the historical learning effect, the adaptability of the control method to the data flow is ensured,
in order to further ensure the following accuracy, as a preferred embodiment, the method for controlling the near-zero following error of the robot based on data driving provided by this embodiment further includes, before sending the target control amount to the robot: constructing a decoupling model between the position and the speed of the tail end of the robot, and determining an advance amount corresponding to a target control amount by using the decoupling model, namely a compensation opportunity;
and, the time for transmitting the target control amount to the robot is determined by the lead amount;
in this embodiment, the process of constructing the decoupling model between the position and the speed is as follows:
firstly, a robot terminal reference track and a jacobian matrix are obtained, the robot terminal reference track selected in the embodiment is a fourth-order spatial Bezier curve track, and the expression of the four-order spatial Bezier curve track is as follows:
wherein, PiIs the coordinate of the ith control point, Ji,4(t) is the mixed basis function of the Bezier curve,
the number of combinations of i elements is taken from the n number; and f(s) is a reference track having coordinate components of three axes of X, Y and Z in a Cartesian coordinate system.
The jacobian matrix calculation formula of this embodiment is as follows:
wherein X, Y, Z is a 6-dimensional column vector and represents the position of the six-degree-of-freedom laser welding robot; thetaX、θY、θZThe six-dimensional column vector represents the posture of the six-degree-of-freedom laser welding robot; q is a joint position vector, and all the embodiments are joint angles; j is a Jacobian matrix used for describing the transformation relation between the speed of the tail end of the robot and the speed of the joint;
finally, in this embodiment, the decoupling model satisfies the following relationship:
δF＝g(f(s),qE,J,δG)；
wherein, deltaFRepresenting the amount of lead, f(s) being the reference track, qEFor the pose of the robot tip, J is the Jacobian matrix describing the transformation between the robot tip velocity and joint velocity, deltaGThe compensation value is output by the online deep neural network, and g is a lead quantity constraint relation;
the specific decoupling model needs to be combined with a corresponding welding robot system to perform relationship setting, and optionally, in this embodiment, the following linear combination relationship is satisfied between the input and the output of the advance constraint relationship g:
δF＝Adf(s)+BdqE+CdJ+DδG；
wherein d represents a difference between the current control period and the previous control period, e.g., df(s) represents a difference between two discrete points on the reference trajectory corresponding to the current control period and the previous control period; A. b, C and D are 4 coefficients in a linear combination;
according to the embodiment, the corresponding lead amount of the compensation value is obtained according to the decoupling model between the position and the speed, so that the time lag error and the lead error caused by abrupt change of the track curvature can be avoided, the situation of lagging following or overshooting is avoided, and the track following precision is further ensured.
In order to further ensure the stability of the system, in the embodiment, the process of obtaining the control quantity is controlled by utilizing variable parameter self-adaptation, so that the constraint of the Lyapunov energy function is met; the definition of the lyapunov energy function herein is the same as the lyapunov energy function defined during the dynamic adjustment of the model parameters, and will not be repeated here.
In order to further improve the precision of trajectory tracking, as an optional implementation manner, the method for controlling the near-zero tracking error of the robot based on data driving provided in this embodiment further includes: correcting the coefficients in the mapping function by combining an experimental method; the mapping function is used for mapping the position compensation values corresponding to the discrete points on the reference track into angle compensation quantities of joints of the robot;
due to joint cascade and error coupling, the mapping relation between the position compensation value and each joint angle compensation value obtained by a theoretical derivation mode still has the problem of under-compensation or over-compensation.
In other complex multi-curved surface workpiece processing applications, such as grinding and polishing of complex multi-curved surface workpieces, the embodiment of the method for controlling the near-zero following error of the robot based on data driving provided by the invention is similar to the trajectory following control of a welding robot, and will not be listed here.
Example 2:
a computer-readable storage medium, comprising a stored computer program, which, when executed by a processor, controls an apparatus in which the computer-readable storage medium is located to execute the method for controlling a near-zero following error of a robot based on data driving provided in embodiment 1 above.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (10)
1. A robot near-zero following error control method based on data driving is characterized by comprising the following steps:
extracting target attributes of the robot from online process data, inputting the target attributes into an online deep neural network model, obtaining a position compensation value corresponding to each discrete point on a reference track at the tail end of the robot in each control period, and mapping the position compensation values into angle compensation quantities of joints of the robot; compensating the input of the variable parameter self-adaptive controller by using the angle compensation quantity to obtain a target control quantity, and sending the target control quantity to the robot so as to enable the robot to move along the reference track and complete the control of the current control period;
in the whole track following process of the robot, dynamically adjusting the parameters of the online deep neural network model according to a preset structure updating rule;
the target attribute is the attribute which can most represent the influence on the following error in the process data for controlling the robot track following; the online deep neural network model is a multi-input multi-output deep neural network and is used for predicting a position compensation value corresponding to each discrete point on the reference track according to the attribute of the robot; and according to the structure updating rule, the parameters of the online deep neural network model dynamically change along with the change of the working condition and the data quantity accumulated in the following process.
2. The data-driven-based near-zero following error control method for the robot as claimed in claim 1, wherein the parameters of the online deep neural network model are dynamically adjusted and are completed through a chicken flock optimization algorithm.
3. The data-driven-based near-zero following error control method for the robot is characterized in that the constraint of a Lyapunov energy function is met through a process of dynamically adjusting parameters of the online deep neural network model through a chicken flock optimization algorithm;
the lyapunov energy function is defined as follows:
wherein x ise、ye、zeIs the positional deviation of the tail end of the robot,
is the derivative of V with respect to time; if and only if xe＝ye＝zeWhen the content is equal to 0, the content,
4. the data-driven-based near-zero following error control method for the robot as claimed in claim 1, wherein the structure updating rule is obtained by:
and testing the track following effects of different robots on various different tracks to obtain an online deep neural network model structure optimization method for processing track following dynamic data streams under each track, and determining the optimization method capable of obtaining the optimal control compensation value as the structure updating rule.
5. The data-driven-based near-zero following error control method for a robot according to any one of claims 1 to 4,
before sending the target control quantity to the robot, the method further includes: constructing a decoupling model between the position and the speed of the tail end of the robot, and determining an advance corresponding to the target control quantity by using the decoupling model;
and, the time for transmitting the target control amount to the robot is determined by the lead amount.
6. The data-driven-based near-zero following error control method for the robot as claimed in claim 5, wherein the decoupling model satisfies the following relationship:
δF＝g(f(s),qE,J,δG)；
wherein, deltaFRepresenting said lead amount, f(s) being said reference trajectory, qEFor the end pose of the robot, J is a Jacobian matrix describing the transformation relationship between the robot end velocity and joint velocity, δGAnd g is a lead amount constraint relation for a compensation value output by the online deep neural network.
7. The data-driven-based near-zero following error control method for the robot as claimed in claim 6, wherein the input and the output of the lead amount constraint relation g satisfy the following linear combination relation:
δF＝Adf(s)+BdqE+CdJ+DδG；
where D represents the difference between the current control period and the previous control period, and A, B, C and D are 4 coefficients in a linear combination relationship.
8. The data-driven-based near-zero following error control method for the robot as claimed in any one of claims 1 to 4, wherein the process of obtaining the control quantity is adaptively controlled by using variable parameters, and the constraint of the Lyapunov energy function is satisfied;
the lyapunov energy function is defined as follows:
wherein x ise、ye、zeIs the positional deviation of the tail end of the robot,
is the derivative of V with respect to time; if and only if xe＝ye＝zeWhen the content is equal to 0, the content,
9. the data-driven-based near-zero following error control method for the robot according to any one of claims 1 to 4, further comprising: correcting the coefficients in the mapping function by combining an experimental method; and the mapping function is used for mapping the position compensation values corresponding to the discrete points on the reference track into angle compensation quantities of all joints of the robot.
10. A computer-readable storage medium, comprising a stored computer program, which, when executed by a processor, controls an apparatus to perform the method of any one of claims 1 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011089607.3A CN112327938B (en) | 2020-10-13 | 2020-10-13 | Robot near-zero following error control method based on data driving |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011089607.3A CN112327938B (en) | 2020-10-13 | 2020-10-13 | Robot near-zero following error control method based on data driving |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112327938A true CN112327938A (en) | 2021-02-05 |
| CN112327938B CN112327938B (en) | 2021-11-02 |
Family
ID=74314847
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011089607.3A Active CN112327938B (en) | 2020-10-13 | 2020-10-13 | Robot near-zero following error control method based on data driving |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112327938B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113043277A (en) * | 2021-04-01 | 2021-06-29 | 中联重科股份有限公司 | Multi-joint mechanism trajectory planning method and device, electronic equipment and storage medium |
| CN113589753A (en) * | 2021-07-12 | 2021-11-02 | 西安电子科技大学 | Three-dimensional model surface path planning method, system, equipment, terminal and application |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| FR2668625A1 (en) * | 1990-10-30 | 1992-04-30 | Thomson Csf | Automatic learning process and multilayer connectional network for the implementation of this process |
| CN109176525A (en) * | 2018-09-30 | 2019-01-11 | 上海神添实业有限公司 | A kind of mobile manipulator self-adaptation control method based on RBF |
| CN109739090A (en) * | 2019-01-15 | 2019-05-10 | 哈尔滨工程大学 | A kind of autonomous type underwater robot neural network intensified learning control method |
| CN110802589A (en) * | 2019-10-23 | 2020-02-18 | 山东科技大学 | Hysteresis compensation method for single-joint servo control of industrial robot |
| US20200070346A1 (en) * | 2018-08-28 | 2020-03-05 | Kabushiki Kaisha Toshiba | Robot control device, robot control parameter adjustment method, and non-transitory storage medium storing program |
| CN111203890A (en) * | 2020-02-28 | 2020-05-29 | 中国科学技术大学 | Position error compensation method of robot |
| CN111618864A (en) * | 2020-07-20 | 2020-09-04 | 中国科学院自动化研究所 | Robot model prediction control method based on adaptive neural network |
| CN111679644A (en) * | 2020-07-07 | 2020-09-18 | 南京航空航天大学 | Uncertain industrial robot motion control method considering system delay |
-
2020
- 2020-10-13 CN CN202011089607.3A patent/CN112327938B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| FR2668625A1 (en) * | 1990-10-30 | 1992-04-30 | Thomson Csf | Automatic learning process and multilayer connectional network for the implementation of this process |
| US20200070346A1 (en) * | 2018-08-28 | 2020-03-05 | Kabushiki Kaisha Toshiba | Robot control device, robot control parameter adjustment method, and non-transitory storage medium storing program |
| CN109176525A (en) * | 2018-09-30 | 2019-01-11 | 上海神添实业有限公司 | A kind of mobile manipulator self-adaptation control method based on RBF |
| CN109739090A (en) * | 2019-01-15 | 2019-05-10 | 哈尔滨工程大学 | A kind of autonomous type underwater robot neural network intensified learning control method |
| CN110802589A (en) * | 2019-10-23 | 2020-02-18 | 山东科技大学 | Hysteresis compensation method for single-joint servo control of industrial robot |
| CN111203890A (en) * | 2020-02-28 | 2020-05-29 | 中国科学技术大学 | Position error compensation method of robot |
| CN111679644A (en) * | 2020-07-07 | 2020-09-18 | 南京航空航天大学 | Uncertain industrial robot motion control method considering system delay |
| CN111618864A (en) * | 2020-07-20 | 2020-09-04 | 中国科学院自动化研究所 | Robot model prediction control method based on adaptive neural network |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113043277A (en) * | 2021-04-01 | 2021-06-29 | 中联重科股份有限公司 | Multi-joint mechanism trajectory planning method and device, electronic equipment and storage medium |
| CN113043277B (en) * | 2021-04-01 | 2022-06-07 | 中联重科股份有限公司 | Multi-joint mechanism trajectory planning method and device, electronic equipment and storage medium |
| CN113589753A (en) * | 2021-07-12 | 2021-11-02 | 西安电子科技大学 | Three-dimensional model surface path planning method, system, equipment, terminal and application |
| CN113589753B (en) * | 2021-07-12 | 2022-09-09 | 西安电子科技大学 | Three-dimensional model surface path planning method, system and equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112327938B (en) | 2021-11-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Pane et al. | Reinforcement learning based compensation methods for robot manipulators | |
| CN100562823C (en) | A kind of contour outline control method of complicated track | |
| CN109782601B (en) | Design method of self-adaptive neural network synchronous robust controller of coordinated mechanical arm | |
| Dixon | Adaptive regulation of amplitude limited robot manipulators with uncertain kinematics and dynamics | |
| CN112327938B (en) | Robot near-zero following error control method based on data driving | |
| Zhang et al. | Design and implementation of hybrid force/position control for robot automation grinding aviation blade based on fuzzy PID | |
| Ogbemhe et al. | Towards achieving a fully intelligent robotic arc welding: a review | |
| Zhang et al. | Robotic constant-force grinding control with a press-and-release model and model-based reinforcement learning | |
| Dachang et al. | Constant force PID control for robotic manipulator based on fuzzy neural network algorithm | |
| Zhang et al. | The hybrid force/position anti-disturbance control strategy for robot abrasive belt grinding of aviation blade base on fuzzy PID control | |
| Lin et al. | Guidance law design by adaptive fuzzy sliding-mode control | |
| Tian et al. | Constrained motion control of flexible robot manipulators based on recurrent neural networks | |
| Kumar et al. | Adaptive control systems in CNC machining processes--a review | |
| Duong et al. | Contour error pre-compensation for five-axis high speed machining: offline gain adjustment approach | |
| Wen et al. | A novel 3D path following control framework for robots performing surface finishing tasks | |
| Zhang et al. | A review of surface quality control technology for robotic abrasive belt grinding of aero-engine blades | |
| CN114310914A (en) | Fuzzy self-adaptive iterative trajectory tracking control method and system for multi-degree-of-freedom mechanical arm | |
| He et al. | High-speed high-accuracy spatial curve tracking using motion primitives in industrial robots | |
| Dai et al. | Constant force control for aluminum wheel hub grinding based on ESO+ backstepping | |
| CN113043266A (en) | Adaptive force tracking control method based on iterative learning | |
| Dai et al. | Research on constant force grinding control of aero-engine blades based on extended state observer | |
| Liu et al. | A new approach to the pre-compensation of contour errorsfor three-axis machine tools using an adaptive cross-coupled controller | |
| Handler et al. | Multidimensional path tracking with global least squares solution | |
| Dai et al. | Adaptive friction compensation control of robotic pneumatic end-effector based on LuGre model | |
| CN114800529A (en) | Industrial robot positioning error online compensation method based on fixed-length memory window incremental learning and incremental model reconstruction |
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 |