Puncture robot based on neural network and nonlinear observer
By adopting a control system based on neural networks and nonlinear observers in the puncture robot, the control accuracy problem of the puncture robot in an uncertain environment is solved, and higher control accuracy and stability are achieved.
Patent Information
- Application Number
- CN202510321586.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2024-03-29
- Filing Date
- 2025-03-18
- Publication Date
- 2025-06-27
AI Technical Summary
The control accuracy of the piercing robot in the actual working environment is affected by various uncertain factors, especially under the huge damping changes at the moment when the damping layer is pierced, the existing control technologies such as PID control and sliding mode control are not ideal.
The control system based on neural network and nonlinear observer is adopted, including dynamic model construction, decomposition module, robust controller construction and interference observer design, and the radial basis neural network is used to predict puncture errors and correct the controller to improve control accuracy.
It effectively improves the control accuracy of the puncture robot in a nonlinear time-varying environment, reduces the impact of disturbances and external interference, and improves the stability and accuracy of the puncture process.
Smart Images
Figure CN120206514A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of robots, and particularly to a puncture robot based on a neural network and a nonlinear observer. Background Art
[0002] In the actual working environment of a puncture robot, its control accuracy is often affected by various uncertain factors. For example, the wear of robot joint parts, the change of lubrication environment, the error of dynamic modeling, and the movement of the punctured object. Especially, a huge damping change will occur when the puncture robot pierces the damping layer. Some of these uncertain factors come from the disturbances inside the robot system, and some come from the external disturbances caused by the change of the external environment. These disturbances will have a great impact.
[0003] For the control of a puncture robot, which is a nonlinear time-varying system, there are already various control technologies such as PID control technology and sliding mode control technology. PID control has been widely used because of its low dependence on the dynamic model. However, in the case of nonlinear friction and external disturbances, the PID control effect is not ideal. Sliding mode control is prone to chattering, which affects the control accuracy. Summary of the Invention
[0004] The present invention discloses a puncture robot based on a neural network and a nonlinear observer. The control system of the biopsy robot includes: a dynamic model construction module, a dynamic model decomposition module, a first robust controller construction module, a puncture error prediction module, a second robust controller construction module, and a third robust controller control module;
[0005] The dynamic model construction module simplifies the puncture robot and establishes a dynamic model of the puncture robot;
[0006] The dynamic model decomposition module decomposes the dynamic model to obtain a nominal control term, a disturbance control term, and an external puncture disturbance term;
[0007] The first robust controller construction module constructs a first robust controller for the puncture robot according to the nominal control term and the disturbance control term;
[0008] The puncture error prediction module uses the first robust controller to control the puncture robot to perform a simulation damping layer test puncture to obtain sample data of the control error during puncture; trains a radial basis neural network model with the error sample data, and predicts the puncture error with the trained radial basis neural network model;
[0009] The second robust controller construction module constructs a second robust controller by combining the first robust controller and the puncture error;
[0010] The third robust controller control module designs a disturbance observer to correct the second robust controller, and obtains a third robust controller for controlling the puncture robot.
[0011] The present invention also discloses a factory setting method for a puncture robot based on a neural network and a nonlinear observer, and the specific method is as follows:
[0012] Simplify the puncture robot and establish a dynamic model of the puncture robot;
[0013] Decompose the dynamic model to obtain a nominal control term, a disturbance control term, and an external puncture disturbance term;
[0014] Construct a first robust controller for the puncture robot according to the nominal control term and the disturbance control term;
[0015] Use the first robust controller to control the puncture robot to perform a simulation damping layer test puncture, and obtain sample data of the control error during puncture; train a radial basis neural network model with the error sample data, and predict the puncture error with the trained radial basis neural network model;
[0016] Combine the first robust controller and the puncture error to construct a second robust controller;
[0017] Design a disturbance observer to correct the second robust controller, and obtain a third robust controller for controlling the puncture robot.
[0018] Furthermore, the puncture robot is simplified into five sequentially connected active joints, and the dynamic model of the puncture robot is constructed as follows:
[0019]
[0020] where t ∈ R is time, q ∈ R n is the position vector, is the velocity vector, is the acceleration vector, represents the uncertain parameter, the set ∑ represents the possible bounds of the unknown σ, τ(t) ∈ R n is the control input vector; in addition, M(q, σ, t) represents the inertia matrix, represents the Coriolis / centrifugal term matrix, G(q, σ, t) represents the gravity, represents the friction between joints, and the functions M(·), V(·), G(·), F(·) are continuous.
[0021] Furthermore, decompose the dynamic model to obtain the nominal control term and the disturbance control term, specifically as follows:
[0022]
[0023] where is the nominal part, ΔM, ΔV, ΔG, ΔF are the disturbance terms, and τ d Puncture external disturbance.
[0024] Furthermore, a first robust controller for the puncture robot is constructed as follows:
[0025]
[0026]
[0027] P = diag[k pi 5×5
[0028] D = diag[k vi 5×5
[0029] where τ1 is the output control torque, k pi 、k vi are the proportional coefficient and the differential coefficient based on the traditional PID controller, γ is a constant greater than 0, and ∈ > 0 is an arbitrary design parameter.
[0030] Furthermore, the training of the radial basis neural network model is as follows:
[0031] During the test puncture, the position, velocity, and force feedback information collected by the tip sensor of the puncture robot are input into the input layer;
[0032] The hidden layer further processes the output of the input layer using the Gaussian function as the activation function;
[0033] The output layer linearly outputs the output of the hidden layer neurons and outputs the result.
[0034] Furthermore, a second robust controller is constructed as follows:
[0035]
[0036] Furthermore, the puncture error ε is regarded as part of the mixed disturbance, and a disturbance observer for the puncture error is designed as follows:
[0037]
[0038] where z ∈ R 2 , is the estimate of the error ε, and the nonlinear function is the gain matrix.
[0039] Furthermore, a third robust controller is constructed as follows:
[0040]
[0041] Similarly, the present invention also discloses a storage medium storing a number of instructions, and a processor loads the instructions to execute the factory setting method of the puncture robot based on the neural network and the non-linear observer as described above.
[0042] Other advantages, objectives, and features of the present invention will be described to some extent in the subsequent specification, and to some extent, will be obvious to those skilled in the art based on the study of the following text, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the following specification. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] The drawings of the present invention are described as follows.
[0044] Figure 1 It is a simplified structural schematic diagram of the puncture robot.
[0045] Figure 2 It is a structure diagram of the radial basis neural network.
[0046] Figure 3 It is a schematic diagram of the control input of the final torque.
[0047] Figure 4 It is a three-dimensional schematic diagram of the puncture robot. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0048] The present invention will be further described below with reference to the drawings and embodiments.
[0049] A factory setting method for a puncture robot based on a neural network and a non-linear observer, the specific steps are as follows:
[0050] S1. Simplify the puncture robot and establish a dynamic model of the puncture robot;
[0051] In step S1, the puncture robot is as Figure 4 shown, and the simplified model is as Figure 1 shown. The robot joints are respectively represented by 1, 2, 3, 4, 5, and the components connected to the joints are respectively represented by A, B, C, D, E. Among them, XYZ is a rectangular coordinate system based on point O, d1 is the displacement of point A, d2 is the displacement of point B, d3 is the displacement of point C, h1 is the distance between D and C, h2 is the vertical distance from D to P, θ4 is the angular displacement of joint 4, and θ5 is the angular displacement of joint 5.
[0052] S2. Decompose the dynamic model to obtain a nominal control term, a disturbance control term, and an external puncture disturbance term.
[0053] According to the principle of Lagrange's equation, the kinetic energy of each joint during the movement of the robot can be expressed as
[0054] K = K1 + K2 + K3 + K4 + K5
[0055] The mass of component A is m1, and the displacement is d1. The kinetic energy K1 of component A can be expressed as
[0056]
[0057] The mass of component B is m2, and the displacement is d2. Given the centroid coordinates of component B (d1, d2, 0), the kinetic energy K3 of component C can be obtained
[0058]
[0059] To reduce the complexity of the equation, let the centroid of component D be on the central axis, then the moment of inertia I4 = 0. Given the mass m4 of component D, from the centroid coordinates of component D (d1, d2, d3 + h1), find its kinetic energy K4
[0060]
[0061] Calculate the kinetic energy of component E. To reduce the amount of calculation with minimal impact on the result, assume its moment of inertia is I E , and the mass of component E is m5. Then
[0062]
[0063] The velocity v of the centroid of component E E can be obtained from the following formula
[0064]
[0065] Substituting the above formula, the kinetic energy K of the whole robot can be solved as
[0066]
[0067] Calculate the overall potential energy: Taking the lower plane of the base of the acupuncture robot as the zero potential energy surface, the overall potential energy P is
[0068] P = P1 + P2 + P3 + P4 + P5
[0069] Where: P1 = 0, P2 = 0, P3 = m3gd3, P4 = m4g(d3 + h1), P5 = mg(h1 + h2 + d3 + lS4)
[0070] Simplify to get
[0071] P = (m3 + m4 + m5)gd3 + (m3 + m4)gh1 + m5gh2 + lm5gS4
[0072] Substituting into the Lagrange equation, the dynamic equation of the designed MR-compatible acupuncture robot can be obtained as
[0073]
[0074] Or rewrite the above equation to make the physical meaning more explicit as
[0075]
[0076] where
[0077]
[0078] In the formula: q = [d1, d2, d3, θ4, θ5] T ; m i (i = 1, 2, 3, 4, 5) represents the mass of each module; d1, d2, d3, θ4, θ5 represent the displacements of each module; S i represents sin(θ i ), C i represents cos(θ i ).
[0079] Introduce the uncertain parameter σ and time t to establish the dynamic model of the puncture robot. The mathematical equation of the dynamic model is expressed as:
[0080]
[0081] where t ∈ R is time, q ∈ R n is the position vector, is the velocity vector, is the acceleration vector, represents the uncertain parameter (the set ∑ represents the possible bounds of the unknown σ), τ(t) ∈ R n is the control input vector; in addition, M(q, σ, t) represents the inertia matrix, represents the Coriolis / centrifugal term matrix, G(q, σ, t) represents the gravity, represents the friction between joints, and the functions M(·), V(·), G(·), F(·) are continuous.
[0082] S3. According to the nominal control term and the disturbance control, construct the first robust controller of the puncture robot.
[0083] In step S3, separate the uncertain terms in the dynamic model:
[0084]
[0085] where is the nominal part, ΔM, ΔV, ΔG, ΔF are the perturbation terms, τ d Puncture external perturbation.
[0086] Define the system trajectory tracking error:
[0087] e(t) = q(t) - q d (t)
[0088] Furthermore
[0089]
[0090] where q(t) is the actual trajectory, q d (t) is the ideal trajectory, is the actual velocity, is the ideal velocity, is the actual acceleration, is the ideal acceleration
[0091] Define Φ as the uncertain perturbation vector, then
[0092]
[0093] Furthermore, define the uncertain perturbation boundary as
[0094]
[0095] where S > 0 and is a constant, ||·|| is the Euclidean norm, ρ is the mixed perturbation boundary, and when Φ ≡ 0, it is regarded as no perturbation. Thus, construct the first robust controller
[0096]
[0097] P = diag[k pi 5×5
[0098] D = diag[k vi 5×5
[0099] where τ1 is the output control torque, k pi , k vi are the proportional coefficient and the differential coefficient based on the traditional PID controller, γ is a constant greater than 0, and ∈ > 0 is an arbitrary design parameter.
[0100] S4. Use the first robust controller to control the puncture robot to perform a test puncture, and obtain the sample data of the control error during puncture.
[0101] S5. Train the radial basis neural network model with the error sample data, such as Figure 2 As shown, the trained radial basis neural network model is used to predict the puncture error.
[0102] In step S5, the radial basis neural network is trained according to the position, velocity signal information and force feedback signal information obtained in real time during the puncture process, and an estimation model is constructed for the uncertain disturbance ρ to compensate for the system uncertainty of the puncture robot. The designed RBF neural network is a three-layer neural network composed of an input layer, a hidden layer and an output layer. The specific processing process of the radial basis neural network for signals is as follows:
[0103] Input layer: The position, velocity and force feedback signals are obtained in real time through the needle tip sensor, and the signals are introduced into the neural network
[0104] Input signal is the neural network input of the i-th joint
[0105] where e is the position error, is the velocity error, is the acceleration error, f is the soft tissue contact force, is the derivative of the soft tissue contact force
[0106] Hidden layer, using the Gaussian function as the activation function to further process the signal
[0107] The Gaussian function h(x) is selected as the activation function of the radial basis function
[0108]
[0109] where b j is the width of the radial basis function; c j is the center point of the neuron; m is the number of neurons in the hidden layer.
[0110] Output layer, linearly combining and outputting the outputs of the hidden layer neurons
[0111] The output value of the radial basis neural network is set to
[0112] where W T =[W0,W1,…,W i-1 ,W i T is the neural network output weight vector, and ε is the estimation error.
[0113] S6. Combine the first robust controller and the puncture error to construct a second robust controller.
[0114]
[0115] S7. Design an interference observer, modify the second robust controller, and obtain a third robust controller for the control of the puncture robot, as Figure 3 shown.
[0116] In step S7, the error ε is regarded as part of the mixed disturbance, and a nonlinear disturbance observer is designed for the error ε.
[0117] Construct an auxiliary variable z:
[0118] where z ∈ R 2 , is an estimate of the error ε, and the nonlinear function is to be determined.
[0119] Furthermore, let Then:
[0120]
[0121] where is the gain matrix.
[0122] Furthermore, the nonlinear interference observer is designed as follows:
[0123]
[0124] Furthermore, integrating gives
[0125]
[0126] where K is an invertible matrix to be solved.
[0127] Furthermore, the gain matrix can be taken as:
[0128]
[0129] Then
[0130]
[0131] where is a more accurate estimate of ρ.
[0132] The third robust controller is as follows:
[0133]
[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: it is still possible to modify the specific embodiments of the present invention or make equivalent substitutions, and any modification or equivalent substitution that does not depart from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.
Claims
1. A puncture robot based on a neural network and a nonlinear observer, characterized in that: The control system of the live picking robot includes: a dynamic model building module, a dynamic model decomposition module, a first robust controller building module, a puncture error prediction module, a second robust controller building module and a third robust controller control module; The dynamic model building module simplifies the puncture robot and establishes a dynamic model of the puncture robot; The dynamic model decomposition module decomposes the dynamic model to obtain the nominal control term, the disturbance control term and the external puncture disturbance term; The first robust controller building module builds a first robust controller of the puncture robot according to the nominal control term and the disturbance control term; The puncture error prediction module controls the puncture robot to perform a simulated damping layer test puncture with a first robust controller to obtain control error sample data during puncture; trains a radial basis function neural network model with the error sample data, and predicts the puncture error with the trained radial basis function neural network model; The second robust controller construction module combines the first robust controller and the puncture error to construct a second robust controller; The third robust controller control module designs a disturbance observer, corrects the second robust controller, and obtains a third robust controller for controlling the puncture robot.
2. A factory setting method for a puncture robot based on a neural network and a nonlinear observer, characterized in that: The specific method is as follows: Simplify the puncture robot and establish its dynamic model; Decompose the dynamic model to obtain the nominal control term, disturbance control term and external puncture disturbance term; According to the nominal control term and disturbance control term, the first robust controller of the puncture robot is constructed; The puncture robot is controlled by the first robust controller to perform a simulated damping layer test puncture, and control error sample data during puncture is obtained; the radial basis function neural network model is trained by the error sample data, and the puncture error is predicted by the trained radial basis function neural network model; Combining the first robust controller with the puncture error, a second robust controller is constructed; A disturbance observer is designed, the second robust controller is modified, and the third robust controller is obtained for controlling the puncture robot.
3. The factory setting method of the puncture robot based on the neural network and the nonlinear observer as claimed in claim 2, characterized in that: The puncture robot is simplified into five movable joints connected in sequence, and the puncture robot power model is constructed as follows: Where t∈R is the time, q∈R n is the position vector, is the velocity vector, is the acceleration vector, represents the uncertain parameters, the set ∑ represents the possible bounds of the unknown σ, τ(t)∈R n is the control input vector; in addition, M(q,σ,t) represents the inertia matrix, represents the Coriolis / centrifugal term matrix, G(q,σ,t) represents gravity, Represents the friction force between joints. The functions M(·), V(·), G(·), and F(·) are continuous.
4. The factory setting method of the puncture robot based on the neural network and the nonlinear observer as claimed in claim 3, characterized in that: Decompose the dynamic model to obtain the nominal control term and disturbance control term as follows: in is the nominal part, ΔM, ΔV, ΔG, ΔF are disturbance terms, τ d Piercing external disturbances.
5. The factory setting method of the puncture robot based on the neural network and the nonlinear observer as claimed in claim 4, characterized in that: Construct the first robust controller of the puncture robot as follows: P=diag[k pi ] 5×5 D=diag[k vi ] 5×5 Where τ1 is the output control torque, k pi , k vi is the proportional coefficient and differential coefficient based on the traditional PID controller, γ is a constant greater than 0, and ∈>0 is an arbitrary design parameter.
6. The factory setting method of the puncture robot based on the neural network and the nonlinear observer as claimed in claim 4, characterized in that: The training of the radial basis neural network model is as follows: During the test puncture, the position, speed and force feedback information collected by the needle tip sensor of the puncture robot are input into the input layer; The hidden layer further processes the output of the input layer by using the Gaussian function as the activation function; The output layer linearly outputs the output of the hidden layer neurons and outputs the result.
7. The factory setting method of the puncture robot based on the neural network and the nonlinear observer as claimed in claim 4, characterized in that: Construct the second robust controller as follows:
8. The factory setting method of the puncture robot based on the neural network and the nonlinear observer as claimed in claim 4, characterized in that: The puncture error ε is regarded as part of the mixed disturbance, and the disturbance observer of the puncture error is designed as follows: Where z∈R 2 , is an estimate of the error ε, a nonlinear function is the gain matrix.
9. The factory setting method of the puncture robot based on the neural network and the nonlinear observer as claimed in claim 4, characterized in that: Construct the third robust controller as follows:
10. A storage medium, characterized in that: A plurality of instructions are stored, and the processor loads the instructions to execute the factory setting method of the puncture robot based on the neural network and the nonlinear observer as described in any one of claims 1 to 9.