Intelligent sliding mode control method for puncture surgery robot based on fuzzy neural network

Abstract

The intelligent sliding mode control method for puncture robot based on fuzzy neural network belongs to the technical field of robot control, and comprises the following steps: 1) collecting the joint position information and joint speed information of the robot in real time, and inputting the normalized information to a fuzzy wavelet neural network as input signals; 2) training the input signals by the fuzzy wavelet neural network to obtain an estimation model; 3) constructing a second-order non-singular fast terminal sliding mode controller according to the collected signal data and the estimation model; and 4) inputting the torque signal generated by the second-order non-singular fast terminal sliding mode controller to the robot to adjust the joint position and joint speed of the puncture robot in real time. The present application can significantly improve the puncture accuracy of the puncture robot under the condition of uncertainty, ensure that the puncture robot system is not sensitive to external interference, and improve the tracking speed of the puncture robot on the preset puncture trajectory.

Description

Technical Field

[0001] This invention belongs to the field of robot control technology, specifically relating to an intelligent sliding mode control method for puncture surgery robots based on fuzzy neural networks. Background Technology

[0002] In the actual working environment of a puncture robot, its control performance is frequently affected by various uncertainties. For example, friction between different mechanical structures and wear of mechanical parts can cause parameter perturbations in the puncture robot, affecting the dynamic modeling during operation and leading to system uncertainties. Furthermore, puncture robots are susceptible to interference from environmental factors such as external measurement noise, and due to their physical characteristics, their actual control torque requires a smooth and continuous input signal. If the puncture robot operates under non-ideal conditions, its control performance deteriorates due to various factors, puncture accuracy and speed decrease, internal mechanical components are damaged, and the puncture robot system may fail rapidly. Therefore, it is essential to address the uncertainties in puncture robot systems appropriately, as this helps improve puncture accuracy, ensures a smooth and successful puncture process, and reduces harm to the patient.

[0003] As a typical nonlinear system, puncture robots have attracted considerable attention in recent years due to the development of various control techniques to optimize their performance. Among these, sliding mode control has proven to be one of the most effective nonlinear control methods, drawing significant research interest. The difference between sliding mode control and traditional control methods lies in its non-fixed "structure." Instead, it can be purposefully modified based on the system's current state during dynamic processes, forcing the system to move up and down along a predetermined trajectory with small amplitudes and high frequencies under certain characteristics. This is known as sliding mode or "sliding motion." This sliding mode can be designed artificially and is independent of system parameters and disturbances, ensuring excellent robustness. It is precisely this strong robustness to external disturbances that makes sliding mode control widely used in robot control.

[0004] In recent years, the development of neural network algorithms has provided opportunities to address the uncertainty problem of puncture robots. Commonly used neural networks include radial basis function neural networks, wavelet neural networks, recurrent neural networks, and convolutional neural networks. By utilizing the powerful estimation capabilities of neural network algorithms for nonlinear functions, it is possible to effectively construct estimation models of the system uncertainties of puncture robots. With the estimation model based on neural network feedback, the controller can compensate for the system uncertainties of the puncture robot, thereby improving its control performance. However, this method is easily limited by the estimation performance of neural network algorithms. The estimation capabilities exhibited by traditional neural networks are insufficient to cope with the complex working conditions faced by puncture robots, thus failing to ensure that the puncture robot achieves the desired puncture accuracy. Therefore, designing a neural network algorithm with strong estimation capabilities and integrating it with sliding mode control into a single controller framework remains a challenge. Summary of the Invention

[0005] To address the aforementioned problems in the existing technology, the present invention aims to provide a control method based on a fuzzy wavelet neural network (FWNN) and a second-order non-singular fast terminal sliding mode controller (SONFTSMC), which is applied to a medical puncture robot system. This method solves the system uncertainty problem of the puncture robot to improve puncture accuracy and ensures that the puncture needle quickly and accurately tracks the preset puncture trajectory, thus smoothly completing the entire puncture process.

[0006] This invention provides the following technical solution:

[0007] The intelligent sliding mode control method for puncture surgery robots based on fuzzy neural networks mainly includes the following steps:

[0008] 1) Install intelligent sensors at the puncture needle of the puncture robot to collect the position and velocity signals of the needle tip in real time, and convert them into the corresponding position and velocity signals of each joint of the puncture robot according to coordinate transformation. Then, send this information to the fuzzy wavelet neural network (FWNN) and the second-order non-singular fast terminal sliding mode controller (SONFTSMC), respectively.

[0009] 2) The Fuzzy Wavelet Neural Network (FWNN) will be trained based on the obtained position and velocity signals to construct an estimation model to compensate for the system uncertainties of the puncture robot. This FWNN consists of two hidden layers. Each hidden layer has a feedback term to store the output of the previous step and, according to the designed weights, feeds it back as the input signal for the next step. The presence of the feedback term not only ensures that the FWNN has strong memory capabilities but also helps improve its estimation accuracy. Furthermore, the output layer's result is also fed back to the input layer according to given weights to ensure that the input signal of the FWNN contains more information. The specific signal processing procedure of the FWNN is as follows:

[0010] 2.1) Input layer: Tan-Sigmoid is selected as the activation function to normalize the input signal.

[0011]

[0012] Where x i It is the i-th term of X, X = [x1, ..., x]. n ] T It is the input signal, φ i It is the i-th term of φ, φ = [φ1, ..., φ] n ] T It is the output signal, O y It is the feedback signal of the output layer, ω roi It is ω ro The i-th term, ω ro =[ω ro1 ,…,ω ron ] T It is the weight vector connecting the input layer and the output layer, i = 1, 2, ..., n.

[0013] 2.2) Blurring layer: This layer will blur the signal.

[0014]

[0015] Where ψ ji It is a fuzzy membership function, ω rji This is the self-feedback weight of this layer. It is the output signal of the previous step in this layer, c ji and b ji Let c represent the translation parameter and the dilation parameter, respectively. For ease of description, we set c = [c1, c2, ..., c...]. L ], c j =[c j1 ,c j2,…,c jn ] T b = [b1, b2, ..., b L ], b j =[b j1 ,b j2 ,…,b jn ] T ω r =[ω r1 ,ω r2 ,…,ω rL ],ω rj =[ω rj1 ,ω rj2 ,…,ω rjn ] T , j = 1, 2, ..., L.

[0016] 2.3) Wavelet layer: This layer will use the Mexican hat function to process the signal.

[0017]

[0018]

[0019] Where h j It is the output of the wavelet layer. ω rmji It is a self-feedback weight. This is the output signal from the previous step. Then we set h = [h1, h2, ..., h... L ] T and ω rmj =[ω rmj1 ,ω rmj2 ,…,ω rmjn ] T .

[0020] 2.4) Output layer: This layer sums all the input signals according to the weights and feeds the result back to the input layer.

[0021] f(X)=ω T h

[0022] Where f(X) is the output of the neural network, ω = [ω1, ω2, ..., ω L ] T This is the weight vector connecting the wavelet layer and the output layer. Finally, the fuzzy wavelet neural network (FWNN) is used to obtain an estimation model for the uncertainty of the puncture robot.

[0023]

[0024] in and These are estimated values ​​of the parameters of the fuzzy wavelet neural network (FWNN).

[0025] 3) The pre-planned puncture path is transformed using coordinates to obtain the reference trajectory acting on the puncture robot system. A second-order non-singular fast termination sliding mode controller (SONFTSMC) is designed by combining the real-time position and velocity information of the puncture robot after coordinate transformation with the estimation model generated by the fuzzy wavelet neural network (FWNN). This controller adjusts the actual position and velocity signals of the puncture robot to achieve rapid tracking of the reference trajectory. The use of second-order sliding mode control technology accelerates the tracking speed of the puncture robot, enhances the system's robustness against external disturbances, and avoids the degradation of control performance caused by chattering. The specific design process of the second-order non-singular fast termination sliding mode controller SONFTSMC is as follows:

[0026] 3.1) Establish a dynamic model by combining the mechanical structure and physical characteristics of the puncture robot.

[0027]

[0028] Where θ is the position vector. It is a velocity vector. H(θ) is the acceleration vector, and H(θ) is the inertia matrix of the robotic arm system. Here, G(θ) is the matrix of Coriolis force and centrifugal force, G(θ) is the gravity vector, and D is the matrix of viscous friction coefficient. It is the vector of static friction force. τ d It is a bounded external disturbance.

[0029] 3.2) Construct an error system model based on the reference trajectory signal and the actual position and velocity signals of the puncture robot:

[0030]

[0031] Where x1 = ε is the position error. It is a speed error.

[0032] 3.3) A second-order sliding surface is designed to ensure rapid convergence of the error system while avoiding singularity problems. This sliding surface consists of two layers: the first layer is established based on the error signal to achieve rapid convergence of the trajectory tracking error of the puncture robot; the second layer is established based on the sliding variables of the first layer, which avoids singularity problems and ensures that the sliding variables of the first layer converge to zero rapidly within a fixed time by using a nonlinear piecewise function. This sliding surface design can effectively improve the tracking performance of the puncture robot system. The specific sliding surface structure is shown below.

[0033]

[0034]

[0035] Where r = [r1, ..., r n ] T K1 and K2 are positive definite symmetric matrices, γ1 is a positive number, and s = [s1, ..., s2]. n ] T K3 is a positive definite symmetric matrix. It is constructed as follows:

[0036]

[0037] in γ2 and γ3 are positive numbers, γ2 > γ3 > 1. It is a positive number.

[0038] 3.4) A second-order non-singular fast terminal sliding mode controller (SONFTSMC) is constructed by combining the designed second-order sliding mode surface and the estimation model generated by the fuzzy wavelet neural network (FWNN). The specific controller design is as follows:

[0039] τ=τ e +τ a

[0040]

[0041]

[0042] Where K a It is a positive definite matrix, K t It is a positive definite matrix, υ1 and γ4 are positive numbers, 0 < γ4 < 1.

[0043] It is constructed as follows:

[0044]

[0045] in

[0046] 4) The control torque generated by the second-order non-singular fast terminal sliding mode controller (SONFTSMC) is used to adjust the actual position and speed signals of the puncture robot in real time until the puncture needle of the puncture robot tracks the preset puncture trajectory.

[0047] Furthermore, the Fuzzy Wavelet Neural Network (FWNN) can effectively establish an estimation model for the system uncertainty of the puncture robot. This is specifically described in steps 2.1)–2.4). The FWNN includes a fuzzification layer and a wavelet layer as hidden layers, and a feedback loop is designed for each hidden layer to store the results of the previous step. This design not only gives the FWNN a memory function but also greatly enhances the estimation capability of the neural network, ensuring that the constructed estimation model can better compensate for the system uncertainty of the puncture robot.

[0048] The second-order sliding surface design in step 3.3) not only avoids singularity problems by using a nonlinear piecewise function, but also ensures that the error system of the puncture robot can converge quickly. The second-segment sliding surface design ensures that the sliding variables of the first layer can converge to zero in a fixed time, which effectively improves the control performance of the designed second-order nonsingular fast terminal sliding mode controller SONFTSMC.

[0049] The second-order nonsingular fast terminal sliding mode controller (SONFTSMC) enables faster tracking speeds, higher tracking accuracy, and chatter-free control torque. This design not only helps improve the puncture accuracy of the puncture robot but also ensures a smooth and rapid completion of the entire puncture process, thereby reducing damage to the patient's body.

[0050] By employing the above-described technology, the beneficial effects of the present invention compared to the prior art are as follows:

[0051] 1) The present invention designs a new second-order non-singular fast terminal sliding mode controller (SONFTSMC) based on fuzzy wavelet neural network (FWNN) to adjust the position and velocity signals of each joint of the puncture robot, so as to ensure that the puncture needle of the puncture robot can quickly and accurately track the preset puncture trajectory.

[0052] 2) In the method of the present invention, the novel fuzzy wavelet neural network (FWNN) is designed to construct an estimation model for the system uncertainty of the puncture robot. Its hidden layer includes a fuzzification layer and a wavelet layer, and each layer is designed with a feedback loop to store the result of the previous step. This design not only enables the fuzzy wavelet neural network (FWNN) to have a memory function, but also greatly enhances the estimation ability of the neural network.

[0053] 3) In the method of the present invention, the singularity problem is overcome by designing a second-order sliding surface and the convergence performance of the puncture robot system is improved; on this basis, the second-order non-singular fast terminal sliding mode controller SONFTSMC designed can provide a chatter-free control input signal and ensure that the puncture robot has a faster tracking speed and higher tracking accuracy. Attached Figure Description

[0054] Figure 1 This is a flowchart of the method of the present invention;

[0055] Figure 2 This is the position tracking response of the two links in this embodiment of the invention;

[0056] Figure 3 This is the speed tracking response of the two links in this embodiment of the invention;

[0057] Figure 4 This is the response curve of the tracking error in an embodiment of the present invention;

[0058] Figure 5 This is the response curve of the control torque in an embodiment of the present invention;

[0059] Figure 6 This is an embodiment of the invention illustrating the response process of the sliding surface;

[0060] Figure 7 This is the adaptive adjustment curve of the translation parameter c in this embodiment of the invention;

[0061] Figure 8 This is the adaptive adjustment curve of the expansion parameter b in this embodiment of the invention;

[0062] Figure 9 This is the adaptive adjustment curve of the weight ω in this embodiment of the invention;

[0063] Figure 10 In this embodiment of the invention, the weight ω r The adaptive adjustment curve;

[0064] Figure 11 In this embodiment of the invention, the weight ω rm The adaptive adjustment curve;

[0065] Figure 12 In this embodiment of the invention, the weight ω ro The adaptive adjustment curve;

[0066] Figure 13 This is a tracking and comparison of position vectors in an embodiment of the present invention;

[0067] Figure 14 This is a tracking and comparison of velocity vectors in an embodiment of the present invention;

[0068] Figure 15 This is a comparison of position tracking errors in an embodiment of the present invention;

[0069] Figure 16 This is a comparison of control torque in an embodiment of the present invention. Detailed Implementation

[0070] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and simulation examples on Matlab / Simulink software. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0071] Conversely, this invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of the invention as defined in the claims. Furthermore, to provide a better understanding of the invention, certain specific details are described in detail below. However, those skilled in the art will fully understand the invention even without these detailed descriptions.

[0072] For ease of demonstration, this invention will use a two-link rigid robotic arm as an example for simulation to verify the feasibility of the designed second-order non-singular fast end-of-line sliding mode control scheme based on a fuzzy wavelet neural network (FWNN). Specifically, it includes the following steps:

[0073] S1: The dynamic equations of the two-link rigid manipulator are constructed as follows.

[0074]

[0075]

[0076] H 21 =m2l2 2 +m2l1l2 cos(θ2)

[0077]

[0078] B 22 =0

[0079] G 11 =(m1+m2)gl1 cos(θ1)+m2gl2 cos(θ1+θ2)G 21 =m2gl2 cos(θ1+θ2)

[0080]

[0081] Wherein, it is assumed that the external disturbance τ d =[τ d1 ,τ d2 ] T It is modeled as follows.

[0082]

[0083] S2: Set the initial state values ​​of the robotic arm system to θ1(0) = 0.2, θ2(0) = 2.1. and The viscous friction coefficient matrix is ​​D = diag{d1, d2} = diag{0.5, 0.5}, and the static friction coefficient is f. c1 =2.5 and f c2 =2.5.

[0084] The specific physical parameters of the robotic arm can be seen in Table 1.

[0085] Table 1 Physical parameters of a two-link rigid robotic arm

[0086]

[0087] S3: The reference trajectory obtained after transforming the preset puncture path using coordinates is given as follows:

[0088]

[0089] The parameters of the second-order nonsingular fast terminal sliding mode controller (SONFTSMC) are set as follows: γ1 = 7 / 3, γ2 = 11 / 5, γ3 = 9 / 5, γ4 = 1 / 3, α = 1.5. v1=0.5, K1=diag{25,25}, K2=diag{5,5}, K3=diag{2,2}, K t =diag{50,50}, K a =diag{20,20}. The parameters of the fuzzy wavelet neural network (FWNN) are set as η1=10, η2=15, η3=0.5, η4=10, η5=45, η6=15. Furthermore, to evaluate the robustness of the robotic arm system when the designed controller is applied, the effect of measurement noise is incorporated into the simulation experiment. It is assumed that the measurement noise is a Gaussian white noise signal, with its mean and variance set to 0 and 0.01, respectively.

[0090] S4: Numerical simulations were performed using Matlab / Simulink software, and the specific simulation results are shown in... Figure 2-12 In the middle. Among them:

[0091] Figure 2 and Figure 3 This demonstrates how two links track a reference trajectory, and the corresponding tracking errors are shown below. Figure 4 As shown. The results demonstrate that, under the proposed controller, the two-link robotic arm can quickly and accurately track the reference trajectory. This also indicates that by placing the puncture needle at the end effector of the robotic arm and combining coordinate transformation with the setting of the reference trajectory, rapid tracking of the puncture needle along the desired puncture path can be achieved. Furthermore, as... Figure 2 and Figure 3 As shown, the effect of measurement noise did not degrade the tracking performance of the robotic arm. This means that the proposed second-order nonsingular fast end-of-line sliding mode controller SONFTSMC is highly robust to measurement noise. The control torque of the two links is as follows: Figure 5 As shown, it can be clearly observed that the joint driving torque is a chatter-free control signal. Figure 6 The response curve of the sliding mode variable over time is plotted, showing that the sliding mode variable can converge to zero quickly. The parameter adjustment process of the fuzzy wavelet neural network (FWNN) over time is illustrated. Figure 7-12 The estimated network parameters are adaptively adjusted from arbitrary initial values ​​until they converge to zero. Ultimately, the parameters of the fuzzy wavelet neural network (FWNN) converge quickly to a constant value, and the tuning process is performed very well. The simulation results above have confirmed the feasibility of the second-order nonsingular fast terminal sliding mode control scheme. Furthermore, the excellent tracking results also demonstrate that the fuzzy wavelet neural network can effectively compensate for the system uncertainties of the robotic arm.

[0092] S5: The second-order non-singular fast terminal sliding mode controller based on a fuzzy wavelet neural network (FWNN) is compared with several other existing sliding mode controllers to verify the superiority of the proposed controller. The controllers compared are: the sliding mode adaptive controller SMAC based on a radial basis function neural network (RBFNN); the non-singular fast terminal sliding mode controller NFTSM based on an RBFNN; the non-singular fast terminal sliding mode controller based on a wavelet neural network (WNN); the non-singular fast terminal sliding mode controller based on a fuzzy wavelet neural network (FWNN); and the non-singular fast terminal sliding mode controller employing a boundary layer method (BLM) and a fuzzy wavelet neural network (FWNN). Specific comparison results are presented in [the table / section]. Figure 13-16 In the middle. Among them:

[0093] Figure 13 and Figure 14 The trajectory tracking response of two links under five different sliding mode controllers is described. Clearly, the proposed control method outperforms the other four methods in tracking speed, whether for joint position tracking or joint velocity tracking. The convergence time of SONFTSMC is less than 0.7 s, while the convergence time of the other four controllers is greater than 1.3 s. Furthermore, the corresponding position tracking errors are as follows: Figure 15 As shown in the figure, this also demonstrates that the sliding mode control strategy based on the fuzzy wavelet neural network (FWNN) has the characteristics of fast response speed and high tracking accuracy. The superior tracking performance indicates that, compared with traditional neural networks, the designed FWNN can more effectively estimate lumped disturbances. Figure 16 The comparison results of five control torques are presented. From Figure 16As can be seen from the enlarged image, the second-order nonsingular fast terminal sliding mode controller (SONFTSMC) exhibits no chattering at its control input, while the chattering problem is not well addressed in the other four methods. This result indicates that the SONFTSM control algorithm has a good chattering suppression effect.

[0094] In summary, the second-order non-singular fast terminal sliding mode controller (SONFTSMC) designed in this invention not only possesses advantages such as fast convergence speed, high tracking accuracy, and no chattering, but also exhibits strong robustness against external disturbances. Therefore, it is feasible to consider using a second-order non-singular fast terminal sliding mode controller based on a fuzzy wavelet neural network to control a puncture robot to complete the entire puncture process.

[0095] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent sliding mode control method for a puncture surgical robot based on fuzzy neural networks, characterized by... Includes the following steps: 1) Based on the tracking behavior of the puncture robot on the preset puncture trajectory, the measurement signals of the robot's joint position and speed are collected in real time and normalized. 2) An estimation model for compensating for uncertainties in the puncture robot system is obtained by training a fuzzy wavelet neural network (FWNN). 3) A second-order non-singular fast end-of-line sliding mode controller (SONFTSMC) is constructed by combining joint position measurement signals, velocity measurement signals, and estimation models; 4) The control torque generated by the second-order non-singular fast terminal sliding mode controller (SONFTSMC) is used to adjust the joint position and joint speed of the puncture robot in real time. The specific design process of the second-order non-singular fast terminal sliding mode controller SONFTSMC is as follows: 2-1) Establish a dynamic model by combining the mechanical structure and physical characteristics of the puncture robot; ； in It is a position vector. It is a velocity vector. It is an acceleration vector. It is the inertia matrix of the robotic arm system. It is the matrix of Coriolis force and centrifugal force. It is the gravity vector. It is the matrix of viscous friction coefficients. It is the vector of static friction force; It is a bounded external disturbance; 2-2) Construct an error system model based on the reference trajectory signal and the actual position and velocity signals of the puncture robot; ； in It is a positional error. It is a speed error; 2-3) Design a second-order sliding surface to ensure fast convergence of the error system while avoiding singularity problems; The second-order sliding surface consists of two layers: the first layer is established based on the error signal to achieve rapid convergence of the trajectory tracking error of the puncture robot; the second layer is established based on the sliding variables of the first layer, which avoids singularity problems and ensures that the sliding variables of the first layer converge to zero quickly within a fixed time by using a nonlinear piecewise function; the specific sliding surface structure is shown below: ； ； in , and It is a positive definite symmetric matrix. It is a positive number. , It is a positive definite symmetric matrix; It is constructed as follows: ； in , and It is a positive number. , , , It is a positive number; 2-4) The second-order non-singular fast terminal sliding mode controller (SONFTSMC) is constructed by combining the designed second-order sliding mode surface and the estimation model generated by the fuzzy wavelet neural network (FWNN); the specific controller design is as follows: ； in It is a positive definite matrix. It is a positive definite matrix. and It is a positive number. ; It is constructed as follows: ； in .

Images