CN115229798A - Force impedance control method integrating feedforward compensation and variable damping modeling - Google Patents

Force impedance control method integrating feedforward compensation and variable damping modeling Download PDF

Info

Publication number
CN115229798A
CN115229798A CN202211004966.3A CN202211004966A CN115229798A CN 115229798 A CN115229798 A CN 115229798A CN 202211004966 A CN202211004966 A CN 202211004966A CN 115229798 A CN115229798 A CN 115229798A
Authority
CN
China
Prior art keywords
force
layer
representing
variable damping
robot
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.)
Pending
Application number
CN202211004966.3A
Other languages
Chinese (zh)
Inventor
党选举
林智武
原翰玫
李晓
伍锡如
张向文
张斌
季运佳
邹水中
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guilin University of Electronic Technology
Original Assignee
Guilin University of Electronic Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Guilin University of Electronic Technology filed Critical Guilin University of Electronic Technology
Priority to CN202211004966.3A priority Critical patent/CN115229798A/en
Publication of CN115229798A publication Critical patent/CN115229798A/en
Pending legal-status Critical Current

Links

Images

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
    • B25J9/00—Programme-controlled manipulators
    • B25J9/16—Programme controls
    • B25J9/1602—Programme controls characterised by the control system, structure, architecture
    • B25J9/161—Hardware, e.g. neural networks, fuzzy logic, interfaces, processor
    • 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
    • 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
    • B25J9/1694—Programme controls characterised by use of sensors other than normal servo-feedback from position, speed or acceleration sensors, perception control, multi-sensor controlled systems, sensor fusion

Abstract

The invention discloses a force impedance control method integrating feedforward compensation and variable damping modeling, which is characterized in that under the framework of position impedance control, adaptive feedforward PI compensation is integrated with a variable damping model of a deep neural network, wherein the adaptive PI control effectively compensates reference position information through feedforward control, and reduces force steady-state errors; the variable damping model of the deep neural network adjusts the damping coefficient on line in real time so as to dynamically adapt to the unknown contact environment and realize the effective control of the contact force.

Description

Force impedance control method integrating feedforward compensation and variable damping modeling
Technical Field
The invention relates to the technical field of robots, in particular to a force impedance control method integrating feedforward compensation and variable damping modeling.
Background
The robot technology enters an intelligent stage and plays an important role in the fields of manufacturing industry, transportation, service industry and the like. When a robot performs operations such as grinding, deburring, holding an article, precision assembly, human-computer interaction, etc., it is often necessary to control the contact force within a certain desired range. But because of different tasks, the change of rigidity and position along with environmental information and the like, the difficulty in keeping good tracking of contact force and controlling dynamic response indexes is increased. Since most industrial robots have a high positional accuracy, the force control of the robot contacting an unknown environment is controlled by means of position control, and the existing methods thereof are: 1) Directly predict the reference trajectory or estimate environmental information. There are documents in which an estimation formula of a reference trajectory and an external force is obtained from a necessary condition of a force steady-state error and an environment model, and environment information is estimated in real time by using Lyapunov stability. There is a document that a reference position adaptive law is constructed according to a difference value between a current position and a reference position, and a PID (Proportion integration-inverse-Derivative) is introduced into impedance control to form a new impedance relation. The literature introduces the idea of fractional order into an adaptive impedance control framework, which exhibits compliance to the contact environment. The document establishes an impedance relation with a self-adaptive damping coefficient from two aspects of environmental rigidity and unknown contact position, and proves the stability of the method. 2) Intelligent control techniques are used for the adjustment of the impedance coefficient. The literature introduces fuzzy control on the basis of self-adaptive impedance control, adjusts the impedance coefficient in real time according to force measurement information, and improves the dynamic response capability of the system. There is literature proposing a fuzzy adaptive hybrid impedance control strategy to compensate for force control errors caused by real-time changes in stiffness of the contact environment. There are documents that use neural networks to compensate the reference trajectory so that the final system effect exhibits impedance characteristics. The document introduces a wavelet neural network to adjust a damping term on line, compensate unknown terms of impedance relation and improve the robustness of a control system. Although research has been conducted to improve the force control performance in terms of predicting a reference trajectory, estimating environmental information, adjusting an impedance coefficient to compensate for system uncertainty, etc., steady-state performance and dynamic characteristics of the robot when contacting the environment are not comprehensively considered, which directly affects further improvement of the force control performance.
Disclosure of Invention
The invention aims to solve the problem that the dynamic change of random and unknown contact environment cannot be effectively adapted in the existing robot manual impedance control, so that the static and dynamic characteristics of industrial robot force control are further improved, and provides a force impedance control method integrating feedforward compensation and variable damping modeling.
In order to solve the problems, the invention is realized by the following technical scheme:
the force impedance control method integrating feedforward compensation and variable damping modeling comprises the following steps:
step 1, collecting contact force F of the robot through a force sensor e And according to a set desired force F r Contact force F with the pick-up e Obtaining a force deviation Δ F, wherein Δ F = F r -F e ；
Step 2, constructing a deep neural network variable damping model, wherein the deep neural network variable damping model is composed of an input layer, a hidden layer, a convolution layer and an output layer, the input of the deep neural network variable damping model is the absolute value | delta F | of force deviation, and the output of the deep neural network variable damping model is a damping coefficient B d ；
Step 3, the absolute value | delta F | of the force deviation is sent to the deep neural network variable damping model constructed in the step 2 to learn the network parameters, and a damping coefficient B is obtained d ；
Step 4, calculating a position error delta x by using impedance models of the robot and the environment; the impedance model of the robot and the environment is as follows:
Figure BDA0003808363600000021
step 5, at the initial reference position x r0 On the basis of the current reference position x, the feedforward PI is used for compensating the current reference position x on line r Namely:
x r ＝x r0 +k p ΔF+k i ∫ΔFdt
step 6, obtaining the position error delta x according to the step 4 and the current reference position x obtained in the step 5 r Obtaining the position control quantity x of the robot c Wherein x is c ＝x r +Δx；
Step 7, acquiring the actual position x of the robot through a position sensor, and controlling the position x obtained in the step 6 c Deviation x from the actual position x of acquisition c After PID control is carried out on the x, control torque of the robot is obtained to carry out position control on the robot, and the purpose of controlling force is achieved;
in the formula, Δ F represents a force deviation; b d Denotes the damping coefficient, M d Denotes the mass coefficient, K d Representing a stiffness coefficient; the position error is represented by deltax,
Figure BDA0003808363600000022
and
Figure BDA0003808363600000023
first and second differentials respectively representing the position error Δ x; x is the number of r Representing the current reference position, x r0 Denotes the initial reference position, k p Denotes the scale parameter, k i Representing the integration parameter.
The deep neural network variable damping model constructed in the step 2 is as follows:
the inputs to the input layer are:
|ΔF|
the activation function of the hidden layer is:
Figure BDA0003808363600000024
the convolution results for the convolutional layer are:
Figure BDA0003808363600000025
the output of the output layer is:
Figure BDA0003808363600000026
where, | Δ F | represents the absolute value of the force deviation; h (i) denotes the activation function of the ith neuron of the hidden layer, c i And b i Respectively representing the height and width of the ith neuron activation function of the hidden layer; a is a j Represents the convolution result of the jth neuron of the convolution layer, B j Represents the bias of the jth neuron of the convolutional layer;
Figure BDA0003808363600000031
representing the sharing weight vector of the hidden layer and the convolutional layer, h (j) representing the activation function of the jth neuron of the hidden layer, h (j + 1) representing the activation function of the jth +1 neuron of the hidden layer, and h (j + 2) representing the activation function of the jth +2 neuron of the hidden layer; w is a j Represents the weight of the jth neuron of the convolutional layer to the output layer, B d Representing a damping coefficient; i =1,2, \8230, I is the number of hidden layer neurons; j =1,2, \8230, J is the number of convolutional layer neurons.
As a refinement, in step 5, the ratio parameter k p Is a fixed value; integral parameter k i Is an adaptive value in a non-linear relationship with the force deviation Δ F, wherein the integral parameter k i Comprises the following steps:
Figure BDA0003808363600000032
in the formula,. DELTA.F represents a force deviation, h 1 Representing the set nonlinear function height and w representing the set nonlinear function width.
Compared with the prior art, the invention has the following characteristics:
1. an introduced deep neural network variable damping model is obtained according to a force error delta F and a damping coefficient B in the deep neural network variable damping model d Mechanism characteristics, active function of designed hidden layer, and useful characteristic of extracting data by convolution layer using shared weight and local connection modeCompared with other neural networks, the method has the advantages that weight parameters are fewer, data characteristics of a hidden layer can be better extracted, the parameter updating speed is increased, and meanwhile, the model complexity is reduced.
2. The feedforward control of a PI structure is introduced to compensate the reference position quantity on line, a nonlinear function with force deviation as an independent variable is designed to change the integral intensity, the self-adaptive adjustment of integral in the feedforward PI compensation is realized, and the aim of dynamically correcting the reference position information is fulfilled.
3. The online acquisition of the variable damping coefficient and the adaptive feedforward PI compensation improve the static and dynamic tracking performance of force control, reduce the overshoot of the force control and realize the flexible control of the robot.
Drawings
FIG. 1 is a schematic diagram of a robot impedance control system according to the present invention.
Fig. 2 is a graph of the activation function of an implied layer.
FIG. 3 is a structural diagram of a deep neural network variable damping model.
Fig. 4 is a graph of a non-linear function.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to specific examples.
The position-based force tracking impedance control is to equate a mass-damping-spring model to a robot tail end and environment contact force model, when force deviation exists, corresponding position error exists at the robot tail end, and the relation between the two errors is adjusted by the impedance coefficient of the model, so that the contact force is indirectly and flexibly controlled.
The impedance model of the robot and the environment is as follows:
Figure BDA0003808363600000033
wherein Δ x is a position error, Δ x = x c -x r ，x c Is the position control quantity, x, of the robot r Is the reference position to which the desired force corresponds,
Figure BDA0003808363600000041
and
Figure BDA0003808363600000042
first and second differentials of the position error Δ x, respectively; m d Is a known mass coefficient, B d Is the damping coefficient, K d Is a known stiffness coefficient; f r Is the desired force, F e It is the contact force that is obtained by the force sensor.
The force sensor describes:
Figure BDA0003808363600000043
wherein x is the actual position, x e Is the environmental location, k e Is the ambient stiffness.
1) Steady state performance
Let force deviation Δ F = F r -F e As known in the art, when the system is stable, the force deviation Δ F corresponds to the steady state error e fss Comprises the following steps:
Figure BDA0003808363600000044
to achieve a steady state error of 0, the following condition is satisfied:
Figure BDA0003808363600000045
equation (4) illustrates: 1) The premise of accurate force control is that the accurate reference position x is obtained in real time through environmental information r Or coefficient of stiffness K d =0. When K is d =0, impedance control degenerates to damping control; 2) When referring to the position x r The contact environment is constant, and the system generates larger force deviation without adaptive adjustment.
In practice, the information of the contact environment is not known precisely, i.e. right afterThe degree and the position of the contact surface are unknown and dynamically changing, which makes x r The time is changed, and the accurate x is obtained in advance r It is very difficult.
2) Dynamic performance
The damping ratio of the robot and the environment can be known by an impedance model
Figure BDA0003808363600000046
Comprises the following steps:
Figure BDA0003808363600000047
natural frequency omega n Comprises the following steps:
Figure BDA0003808363600000048
according to the damping ratio and the influence factor of natural frequency, when the contact environment is changed, the environmental rigidity k e Change in real time, resulting in damping ratio
Figure BDA0003808363600000049
And natural frequency omega n And the change directly influences the dynamic performance of the system. In addition, because of the environmental stiffness k e And a coefficient of stiffness K d All determined by the characteristics of the system, and cannot be adjusted manually. So that the target damping coefficient B is adjusted in real time d The purposes of indirectly adjusting the damping ratio and reducing the influence of environmental change on the dynamic performance of the system can be achieved.
Under the position impedance control framework, the feedforward compensation and variable damping modeling combined force impedance control method provided by the invention is as shown in fig. 1, the set value is obtained by self-adaptive feedforward compensation, and the position error delta x is obtained by obtaining a damping ratio B through a deep neural network variable damping model d Substituting the formula (1) and solving the formula (1) to obtain the product.
(1) Deep neural network variable damping model
In robot force impedance control, force deviation Delta F and damping coefficient B d Mechanism of (1)The characters are as follows: 1) When the force deviation Δ F is large, the damping coefficient B is desired d Become small in order to respond quickly; 2) When the force deviation Δ F is small, the damping coefficient B is desired d And increasing and reducing the overshoot of the system so as to maintain stability. For this purpose, the force deviation Δ F and the damping coefficient B are used d And (3) constructing a deep neural network for describing the characteristics of the mechanism, wherein the activation function of a designed hidden layer is as follows:
Figure BDA0003808363600000051
where h (i) is the activation function of the ith neuron of the hidden layer, which is used to express the force deviation Δ F and the damping coefficient B d H (i) is shown in FIG. 2; c. C i And b i Height and width of the ith neuron activation function of the hidden layer, respectively; Δ F is force deviation, Δ F = F r -F e ，F r Is the desired force, F e Is the contact force.
Considering the complexity of the network structure and the computation amount, the deep neural network variable damping model designed by the invention is composed of an input layer, a hidden layer, a convolution layer and an output layer, as shown in fig. 3. The hidden layer describes the force deviation Δ F and the damping coefficient B by activating a function h (i) d The convolution layer can extract useful characteristics of data by using a shared weight and a local connection mode, and the weight parameters are fewer, so that the quick learning of model weighting is facilitated. In the context of figure 3 of the drawings,
Figure BDA0003808363600000052
is the neural network input variable, i.e. the force deviation Δ F; c. C i Is the activation function height of the hidden layer, b i Is the activation function width of the hidden layer, h (i) is the nonlinear activation function of the ith neuron of the hidden layer;
Figure BDA0003808363600000053
is the shared weight vector of the hidden layer and the convolution layer; z is a radical of formula j Is the j-th neuron activation function value of the convolutional layer, a j Is the jth of the convolutional layerConvolution results of neurons; w is a j Is the weight from the jth neuron of the convolutional layer to the output layer;
Figure BDA0003808363600000054
is the neural network output, i.e. the damping coefficient B d . I =1,2, \8230, and I is the number of hidden layer neurons. J =1,2, \ 8230;, J, J is the number of convolutional layer neurons.
The network optimization process comprises a forward propagation part and a parameter learning part, wherein the forward propagation part calculates an output value, and the parameter learning part updates related parameters by adopting a gradient descent method to approach a target value.
(1.1) Forward propagation of networks
1) An input layer:
Figure BDA0003808363600000055
2) Hidden layer:
Figure BDA0003808363600000056
3) A convolutional layer:
Figure BDA0003808363600000057
Figure BDA0003808363600000061
in the formula (I), the compound is shown in the specification,
Figure BDA0003808363600000062
B j is the bias of the jth neuron of the convolutional layer.
4) An output layer:
Figure BDA0003808363600000063
(1.2) network parameter learning
Let the performance index function be:
Figure BDA0003808363600000064
and respectively obtaining an updating increment formula of the weighting coefficients of each layer of the neural network according to the steepest descent method.
1) Weight increment from output layer to convolutional layer:
Figure BDA0003808363600000065
2) Hidden layer to convolutional layer shared weight increment:
Figure BDA0003808363600000066
in the formula:
Figure BDA0003808363600000067
3) Convolutional layer bias increment:
Figure BDA0003808363600000068
4) Hidden layer activation function width parameter delta:
Figure BDA0003808363600000069
Figure BDA00038083636000000610
Figure BDA00038083636000000611
Figure BDA00038083636000000612
Figure BDA0003808363600000071
Figure BDA0003808363600000072
in the formula:
Figure BDA0003808363600000073
5) Hidden layer activation function height parameter delta:
Figure BDA0003808363600000074
Figure BDA0003808363600000075
Figure BDA0003808363600000076
Figure BDA0003808363600000077
Figure BDA0003808363600000078
Figure BDA0003808363600000079
in the formula:
Figure BDA00038083636000000710
then in the neural network iterative process, the weight coefficient updating formula is as follows
w j (k)＝w j (k-1)+Δw j +α[w j (k-1)-w j (k-2)] (29)
Figure BDA00038083636000000711
B j (k)＝B j (k-1)+ΔB j +α[B j (k-1)-B j (k-2)] (31)
b i (k)＝b i (k-1)+Δb i +α[b i (k-1)-b i (k-2)] (32)
c i (k)＝c i (k-1)+Δc i +α[c i (k-1)-c i (k-2)] (33)
In the formula: w is a j (k) The weight of the jth neuron of the hidden layer to the output layer at the current k moment,
Figure BDA00038083636000000712
is the m-th shared weight of convolutional layer at the current k time, B j (k) For the current time k convolutional layer j neuron bias, b i (k) Width of activation function for the ith neuron of the hidden layer at the current time k, c i (k) The activation function height of the ith neuron of the hidden layer at the current k moment is shown, eta is the learning rate, and alpha is the momentum factor. k-1 is the time immediately preceding the current time k.
(2) Adaptive feed-forward PI compensation
At an initial reference position x r0 Introducing a feedforward PI online compensation reference position x r And directly adjusting the reference position through the force deviation to enable the force steady-state error to approach 0. The feedforward PI compensation reference positions are as follows:
x r ＝x r0 +k p (F r -F e )+k i ∫(F r -F e )dt (34)
in the formula, k p Is a proportional parameter, k i Is an integration parameter.
The fixed proportionality parameter k is a fluctuating system contact process, taking into account the changing contact environment p And an integral parameter k i Cannot be adapted, so the present invention uses a non-linear function F (Δ F) as the integration parameter k i So that the integral parameter k i The speed is dynamically adjusted along with the stress tracking condition in the contact process, the integral action is increased when the force error is small, and the integral action is weakened when the force error is large, so that the variable contact environment can be better adapted, and the stability of a control system is improved. Wherein the nonlinear function F (Δ F) is adjusted by the following rule: when the force deviation is large, F (Δ F) should be reduced to prevent system integral saturation, causing large overshoot; when the force deviation is small, F (Δ F) is increased to eliminate the steady state error, as shown in fig. 4. From the target curve in fig. 4, the design force deviation is an independent variable, and F (Δ F) is a non-linear function of the dependent variable:
Figure BDA0003808363600000081
in the formula, h 1 Representing the set nonlinear function height and w representing the set nonlinear function width. When the force deviation becomes smaller and F (delta F) gradually approaches 1, the integral action becomes stronger; on the contrary, when F (Δ F) gradually approaches 0, the integration function becomes weak, and the dynamic adjustment k is achieved i The purpose of speed is to better reduce the force steady-state error.
When the impedance of the reference position is compensated by the adaptive feedforward PI
x r ＝x r0 +k p ΔF+f(ΔF)∫ΔFdt (36)
Based on the above analysis, the invention provides a force impedance control method integrating feedforward compensation and variable damping modeling, which comprises the following steps:
step 1, collecting contact force F of the robot through a force sensor e And according to a set desired force F r Contact force F with the pick-up e Obtaining a force deviation Δ F, wherein Δ F = F r -F e ；
Step 2, constructing a deep neural network variable damping model, wherein the deep neural network variable damping model is composed of an input layer, a hidden layer, a convolution layer and an output layer, the input of the deep neural network variable damping model is the absolute value | delta F | of force deviation, and the output of the deep neural network variable damping model is a damping coefficient B d ；
The inputs to the input layer are:
|ΔF|
the activation function of the hidden layer is:
Figure BDA0003808363600000082
the convolution results for the convolutional layer are:
Figure BDA0003808363600000091
the output of the output layer is:
Figure BDA0003808363600000092
step 3, the absolute value | delta F | of the force deviation is sent to the deep neural network variable damping model constructed in the step 2 to learn the network parameters, and a damping coefficient B is obtained d ；
Step 4, calculating a position error delta x by using the impedance model of the robot and the environment; the impedance model of the robot and the environment is as follows:
Figure BDA0003808363600000093
step 5, at the initial reference position x r0 On the basis, the current reference position x is compensated on line by using the self-adaptive feedforward PI r Namely:
x r ＝x r0 +k p ΔF+k i ∫ΔFdt
step 6, obtaining the position error delta x according to the step 4 and the current reference position x obtained in the step 5 r Obtaining the position control quantity x of the robot c Wherein x is c ＝x r +Δx；
Step 7, collecting the actual position x of the robot through a position sensor, and controlling the position x obtained in the step 6 c Deviation x from the actual position x of acquisition c After PID control is carried out on the x, the control torque of the robot is obtained to carry out position control on the robot, and the purpose of controlling force is achieved;
where, | Δ F | represents the absolute value of the force deviation; h (i) denotes the activation function of the ith neuron of the hidden layer, c i And b i Respectively representing the height and width of the ith neuron activation function of the hidden layer; a is j Represents the convolution result of the jth neuron of the convolution layer, B j Represents the bias of the jth neuron of the convolutional layer;
Figure BDA0003808363600000094
representing the shared weight vector of the hidden layer and the convolutional layer, h (j) representing the activation function of the jth neuron of the hidden layer, h (j + 1) representing the activation function of the jth +1 neuron of the hidden layer, and h (j + 2) representing the activation function of the jth +2 neuron of the hidden layer; w is a j Represents the weight of the jth neuron of the convolutional layer to the output layer, B d Representing a damping coefficient; i =1,2, \8230, I is the number of hidden layer neurons; j =1,2, \ 8230, J, J is the number of convolutional layer neurons; Δ F represents force deviation; m d Denotes the mass coefficient, K d Representing a stiffness coefficient; the position error is represented by deltax,
Figure BDA0003808363600000095
and
Figure BDA0003808363600000096
first and second differentials respectively representing the position error Δ x; x is a radical of a fluorine atom r Representing the current reference position, x r0 Denotes the initial reference position, k p The set proportion parameter is a set fixed value. k is a radical of formula i Indicating a settingThe integral parameter may be a fixed value or an adaptive value having a nonlinear relationship with the force deviation Δ F, that is, the integral parameter may be a set fixed value
Figure BDA0003808363600000097
h 1 Representing the set nonlinear function height and w representing the set nonlinear function width.
The invention discloses an industrial robot force impedance control method based on self-adaptive feedforward PI compensation and deep neural network variable damping model fusion under a position impedance control framework. The self-adaptive PI control effectively compensates reference position information through feedforward control, and reduces force steady-state errors; the deep neural network variable damping model adjusts the damping coefficient in real time on line so as to dynamically adapt to an unknown contact environment and realize effective control of the contact force.
It should be noted that, although the above-mentioned embodiments of the present invention are illustrative, the present invention is not limited thereto, and therefore, the present invention is not limited to the above-mentioned specific embodiments. Other embodiments, which can be made by those skilled in the art in light of the teachings of the present invention, are considered to be within the scope of the present invention without departing from its principles.

Claims (4)

1. The force impedance control method integrating feedforward compensation and variable damping modeling is characterized by comprising the following steps of:
step 1, collecting contact force F of the robot through a force sensor e And according to a set desired force F r Contact force F with the pick-up e Obtaining a force deviation Δ F, wherein Δ F = F r -F e ；
Step 2, constructing a deep neural network variable damping model, wherein the deep neural network variable damping model is composed of an input layer, a hidden layer, a convolution layer and an output layer, the input of the deep neural network variable damping model is the absolute value | delta F | of force deviation, and the output is a damping coefficient B d ；
Step 3, the absolute value | delta F | of the force deviation is sent to the deep neural network variable damping model constructed in the step 2 to learn the network parameters, and the damping is obtainedCoefficient B d ；
Step 4, calculating a position error delta x by using impedance models of the robot and the environment; the impedance model of the robot and the environment is as follows:
Figure FDA0003808363590000011
step 5, at the initial reference position x r0 On the basis of the current reference position x, the feedforward PI is used for compensating the current reference position x on line r Namely:
x r ＝x r0 +k p ΔF+k i ∫ΔFdt
step 6, obtaining the position error delta x according to the step 4 and the current reference position x obtained in the step 5 r Obtaining the position control quantity x of the robot c Wherein x is c ＝x r +Δx；
Step 7, acquiring the actual position x of the robot through a position sensor, and controlling the position x obtained in the step 6 c After PID control is carried out on the deviation of the collected actual position x, the control torque of the robot is obtained to carry out position control on the robot, and the purpose of controlling force is achieved;
wherein Δ F represents a force deviation; b d Representing the damping coefficient, M d Denotes the mass coefficient, K d Representing a stiffness coefficient; the position error is represented by deltax,
Figure FDA0003808363590000012
and
Figure FDA0003808363590000013
first and second differentials respectively representing the position error Δ x; x is a radical of a fluorine atom r Indicates the current reference position, x r0 Denotes the initial reference position, k p Denotes the scale parameter, k i Representing the integration parameter.
2. The feedforward compensation and variable damping modeling fused force impedance control method according to claim 1, wherein the deep neural network variable damping model constructed in step 2 is:
the inputs to the input layer are:
|ΔF|
the activation function of the hidden layer is:
Figure FDA0003808363590000014
the convolution results for the convolutional layers are:
Figure FDA0003808363590000015
the output of the output layer is:
Figure FDA0003808363590000021
where, | Δ F | represents the absolute value of the force deviation; h (i) denotes the activation function of the ith neuron of the hidden layer, c i And b i Respectively representing the height and width of the ith neuron activation function of the hidden layer; a is a j Represents the convolution result of the jth neuron of the convolution layer, B j Represents the bias of the jth neuron of the convolutional layer;
Figure FDA0003808363590000022
representing the shared weight vector of the hidden layer and the convolutional layer, h (j) representing the activation function of the jth neuron of the hidden layer, h (j + 1) representing the activation function of the jth +1 neuron of the hidden layer, and h (j + 2) representing the activation function of the jth +2 neuron of the hidden layer; w is a j Represents the weight of the jth neuron in the convolutional layer to the output layer, B d Representing a damping coefficient; i =1,2, \8230, I is the number of hidden layer neurons; j =1,2, \8230, J is the number of convolutional layer neurons.
3. Feed forward compensation and variable damping according to claim 1The modeling fused force impedance control method is characterized in that in step 5, a proportional parameter k p Is a fixed value; integral parameter k i Is an adaptive value that has a non-linear relationship with the force deviation deltaf.
4. The feedforward compensation and variable damping modeling fused force impedance control method of claim 3, wherein an integral parameter k i Comprises the following steps:
Figure FDA0003808363590000023
in the formula,. DELTA.F represents a force deviation, h 1 Representing the set nonlinear function height and w representing the set nonlinear function width.
CN202211004966.3A 2022-08-22 2022-08-22 Force impedance control method integrating feedforward compensation and variable damping modeling Pending CN115229798A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211004966.3A CN115229798A (en) 2022-08-22 2022-08-22 Force impedance control method integrating feedforward compensation and variable damping modeling

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211004966.3A CN115229798A (en) 2022-08-22 2022-08-22 Force impedance control method integrating feedforward compensation and variable damping modeling

Publications (1)

Publication Number Publication Date
CN115229798A true CN115229798A (en) 2022-10-25

Family

ID=83680603

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211004966.3A Pending CN115229798A (en) 2022-08-22 2022-08-22 Force impedance control method integrating feedforward compensation and variable damping modeling

Country Status (1)

Country Link
CN (1) CN115229798A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117532623A (en) * 2024-01-10 2024-02-09 南京鼎臻自动化科技有限公司 Mechanical arm external torque estimation method

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117532623A (en) * 2024-01-10 2024-02-09 南京鼎臻自动化科技有限公司 Mechanical arm external torque estimation method
CN117532623B (en) * 2024-01-10 2024-03-29 南京鼎臻自动化科技有限公司 Mechanical arm external torque estimation method

Similar Documents

Publication Publication Date Title
CN108710303B (en) Spacecraft relative attitude control method containing multi-source disturbance and actuator saturation
CN111941432B (en) Artificial intelligence output feedback control method for high-performance mechanical arm
CN110376886B (en) Model prediction control algorithm based on expansion state Kalman filter
Wu et al. Adaptive terminal sliding mode control for hypersonic flight vehicles with strictly lower convex function based nonlinear disturbance observer
CN110908281A (en) Finite-time convergence reinforcement learning control method for attitude motion of unmanned helicopter
CN111872937B (en) Control method for uncertain mechanical arm in task space
Durmaz et al. Sliding mode control for non-linear systems with adaptive sliding surfaces
CN113110048B (en) Nonlinear system output feedback adaptive control system and method adopting HOSM observer
Wang et al. Extended state observer-based fixed-time trajectory tracking control of autonomous surface vessels with uncertainties and output constraints
Qi et al. Adaptive dynamic programing design for the neural control of hypersonic flight vehicles
Pan et al. Online data‐driven composite adaptive backstepping control with exact differentiators
CN115229798A (en) Force impedance control method integrating feedforward compensation and variable damping modeling
CN111273544A (en) Radar pitching motion control method based on prediction RBF feedforward compensation type fuzzy PID
Zhang et al. Low-level control technology of micro autonomous underwater vehicle based on intelligent computing
Karami et al. Adaptive neural observer-based nonsingular terminal sliding mode controller design for a class of nonlinear systems
Xu et al. Terminal sliding mode control using adaptive fuzzy-neural observer
CN111240201B (en) Disturbance suppression control method
CN117226613A (en) Robot constant force control polishing method based on PPO reinforcement learning
CN113741183B (en) Industrial robot self-adaptive admittance control method based on damping ratio model
Li et al. Adaptive optimal trajectory tracking control of AUVs based on reinforcement learning
CN114397820B (en) Self-adaptive control method of piezoelectric micro-motion platform based on Hopfield neural network estimator
Weng et al. SMC-based model-free tracking control of unknown autonomous surface vehicles
Wang et al. A novel prescribed performance controller with unknown dead-zone and impactive disturbance
CN115128951A (en) Double-loop high-performance control method based on expected track limited optimization
CN114637318A (en) Sliding mode control method for hypersonic aircraft

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