A dynamic hysteresis modeling method for pneumatic soft robotic arm
Through the combination of the Bouc-Wen model and the Hopfield neural network, a dynamic hysteresis model of the pneumatic soft robot arm was established, which solved the problem of hysteresis curve changes, and achieved the improvement of the real-time accuracy and motion performance of the hysteresis model.
Patent Information
- Application Number
- CN202211545170.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-11-30
- Publication Date
- 2024-12-13
- Estimated Expiration
- 2042-11-30
AI Technical Summary
Because the soft robotic arms are composed of soft materials, there is a hysteresis nonlinear phenomenon, which affects the motion performance. The hysteresis curve changes with wear, making it difficult to achieve precise control.
The Bouc-Wen model combined with the Hopfield neural network is used to establish a dynamic hysteresis model of the pneumatic soft robotic arm, and real-time adjustment is made through the parameter identification matrix to ensure the accuracy of the hysteresis model.
It effectively solves the problem of hysteresis curve changes in the soft robotic arm, ensures the real-time accuracy of the hysteresis model, and improves the control reliability of motion performance.
Smart Images
Figure CN116184819B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of hysteresis modeling of pneumatic soft robot arms, and in particular relates to a dynamic hysteresis modeling method of a pneumatic soft robot arm. Background Art
[0002] Soft robots made of soft materials such as silicone have the disadvantage of prominent hysteresis. The hysteresis nonlinear phenomenon will affect the motion performance of the soft robot and bring difficulties to further model control. Hysteresis often occurs in nonlinear systems, and soft robots are made of soft materials and are typical nonlinear systems. The main manifestation is that the output of the system is not only related to the current input value, but also to the historical state. The hysteresis phenomenon brings great difficulties to the precise control of soft robots. Based on the above soft robotic arm model, the hysteresis is mainly caused by the following reasons.
[0003] (1) The designed soft robotic arm is mainly composed of soft materials such as silicone. Compared with rigid materials, soft materials have prominent nonlinear characteristics and are difficult to control. Therefore, material properties are the main reason for the hysteresis of the soft robotic arm. (2) The coupling of the designed soft robotic arm has the result of blocking and changing stiffness. Friction and gaps between particles, changes in stiffness, etc. are prone to cause hysteresis. (3) When preparing the soft robotic arm, tiny bubbles exist in the silicone solution used.
[0004] Common hysteresis models include the Preisach model, the Prandtl-Ishlinskii (PI) model, the Maxwell model, the Bouc-Wen model, etc. For the Preisach model, error accumulation may occur, and the Maxwell model is relatively complex.
[0005] In addition, when the soft robotic arm is worn, the hysteresis curve will change accordingly, and the accuracy of the obtained hysteresis model will be reduced. Summary of the invention
[0006] Technical problem solved: The present invention discloses a method for modeling dynamic hysteresis of a pneumatic soft robotic arm. The Bouc-Wen method is used to establish a dynamic hysteresis model of a soft robotic arm, and a Hopfield neural network is constructed to identify parameters of the model. This method can solve the technical problem that the hysteresis curve will change due to wear of the soft robotic arm, and ensure the real-time accuracy of the hysteresis model.
[0007] Technical solution:
[0008] A pneumatic soft robot arm dynamic hysteresis modeling method, the pneumatic soft robot arm includes a pneumatic drive structure and a variable stiffness structure of the soft robot arm; the dynamic hysteresis modeling method includes the following steps:
[0009] S1, use MATLAB / Simulink software to build the Bouc-Wen hysteresis model:
[0010]
[0011] Among them, u(t) is the hysteresis between the air pressure and the end displacement of the pneumatic soft manipulator with uncoupled variable stiffness structure, and is the input air pressure value of the soft manipulator; is the derivative of u(t) with respect to time t; f d (t) is the output of the model, that is, the output displacement value of the soft robot arm; α, β, ρ, γ are unknown parameters that affect the shape of the model; the Bouc-Wen hysteresis model is composed of the addition of a linear part and a hysteresis part, the linear part corresponds to u(t), its weight is ρ, and the nonlinear hysteresis part corresponds to H(t);
[0012] S2, analyze the influence of each parameter change on the shape change of the Bouc-Wen curve; build a Hopfield neural network to obtain the parameter identification weight matrix of the Bouc-Wen hysteresis model;
[0013] S3, the parameter identification weight matrix of the Bouc-Wen hysteresis model is substituted into the Hopfield neural network, and the calculation is performed in MATLAB. After stabilization, the Bouc-Wen hysteresis model parameters to be identified are obtained.
[0014] Furthermore, in theoretical simulation and control, the Bouc-Wen hysteresis model is converted into a discrete model:
[0015]
[0016] Among them, ΔH(k)=H(k)-H(k-1), Δu(k)=u(k)-u(k-1).
[0017] Furthermore, in step S2, a simulation block diagram of the Bouc-Wen model is established using MATLAB or simulink, and the effects of the changes in the parameters α, β, ρ and γ in the simulation block diagram are observed:
[0018] As α increases, the shape of the entire hysteresis loop becomes wider and rotates counterclockwise; when β continues to increase, the entire hysteresis loop also becomes wider, and when β continues to increase, it rotates clockwise; the parameter γ has no effect on the shape of the hysteresis loop, and as ρ increases, the hysteresis loop rotates clockwise; the parameter ρ has no effect on the shape of the hysteresis loop, and as ρ increases, the hysteresis loop rotates counterclockwise.
[0019] Furthermore, in step S2, the Bouc-Wen hysteresis model is inverted to obtain a corresponding inverse hysteresis model for feedforward control of the soft robotic arm; specifically:
[0020] The Bouc-Wen hysteresis model can be rewritten as:
[0021]
[0022] By finding the inverse function of the rewritten Bouc-Wen hysteresis model, we can obtain the inverse hysteresis model of the Bouc-Wen hysteresis model:
[0023]
[0024]
[0025]
[0026] Furthermore, in step S2, the process of building a Hopfield neural network and obtaining a parameter identification weight matrix of the Bouc-Wen hysteresis model includes the following steps:
[0027] For the unknown parameters α, β, ρ, γ, a Hopfield neural network consisting of 4 neurons is constructed; the input-output relationship function of each neuron is:
[0028]
[0029] Where i = 1, 2, 3, 4, w ij is the weight coefficient of the jth neuron to the ith neuron, v i is the output of the neuron, I i is the input of the neuron, u i is the state of the neuron, g(·) is the activation function of the neuron, which is set to a hyperbolic function:
[0030] The energy function of the Bouc-Wen hysteresis model is set to:
[0031]
[0032] The objective function of the Bouc-Wen hysteresis model is set as:
[0033]
[0034] For the Bouc-Wen hysteresis model, the four output values v 1 , …, v 4 For α, β, ρ, γ, take partial derivatives of these four parameters, and make them all zero at the same time, we get:
[0035]
[0036] Let the corresponding terms be equal to obtain the matrix coefficients, and for α(k) we have the following solution equation:
[0037]
[0038] Solve to get w 11 , w 12 , w 13 , w 14 , I 1 The corresponding matrix coefficients;
[0039] Through the equations corresponding to β(k), ρ(k), and γ(k), we can get w 21 , w 22 , w 23 , w 24 , I 2 、w 31 , w 32 , w 33 , w 34 , I 3 、w 41 , w 42 , w 43 , w 44 , I 4 The corresponding matrix coefficients;
[0040] The matrix coefficients are calculated:
[0041]
[0042] Further, in step S3, the parameter identification weight matrix of the Bouc-Wen hysteresis model is substituted into the Hopfield neural network, and the calculation is performed in MATLAB. After stabilization, the process of obtaining the parameters of the Bouc-Wen hysteresis model to be identified includes the following steps:
[0043] S31, build a Hopfield neural network in matlab. The Hopfield neural network parameter identification module corresponds to 6 inputs, namely Δu(k), H(k-1), Δu(k)*|Δu(k)|, H(k-1)*|H(k-1)|, u(k) and y(k); among them, u(k) is the model input at time k, Δu(k) is the model input at time k minus the model input at time k-1; H(k-1) is an intermediate variable, H(k-1)=H(k-2)+ΔH(k), which assigns an initial value to H(k); y(k) is the output of the soft robot arm at time k;
[0044] S32, substituting the six inputs into the parameter identification weight matrix to obtain the output values of α, β, γ, and ρ at the current moment;
[0045] S33, using the output values of α, β, γ, and ρ at the current moment to update the Bouc-Wen model, update the output of the Bouc-Wen model and the six inputs of the parameter identification module at the next moment, and return to step S32 until the output accuracy of the Bouc-Wen model meets the accuracy requirements.
[0046] Beneficial effects:
[0047] The pneumatic soft robotic arm dynamic hysteresis modeling method of the present invention uses the Bouc-Wen method to establish a soft robotic arm dynamic hysteresis model, and builds a Hopfield neural network to perform parameter identification on the model, which can solve the technical problem that the hysteresis curve will change due to wear of the soft robotic arm, and ensure the real-time accuracy of the hysteresis model. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] Figure 1 It is a structural diagram of the Bouc-Wen model;
[0049] Figure 2 This is the MATLAB / simulink simulation block diagram of the Bouc-Wen model;
[0050] Figure 3 is the effect of parameters on the shape of the hysteresis loop, where (a) is the effect of parameter α on the hysteresis loop, (b) is the effect of parameter β on the hysteresis loop, (c) is the effect of parameter γ on the hysteresis loop, and (d) is the effect of parameter ρ on the hysteresis loop;
[0051] Figure 4 It is a schematic diagram of the inverse model structure of the Bouc-Wen model;
[0052] Figure 5 A schematic diagram of the Hopfield neural network structure consisting of four neurons;
[0053] Figure 6 A schematic diagram of the structure of a single neuron in the Hopfield neural network;
[0054] Figure 7 It is a schematic diagram of the parameter identification principle of Bouc-Wen model based on Hopfield neural network;
[0055] Figure 8 The flowchart of Bouc-Wen modeling based on Hopfield neural network;
[0056] Fig. 9 This is a schematic diagram of the open-loop control system structure of the soft robotic arm;
[0057] Fig.10 This is a schematic diagram of the air pressure control experimental platform;
[0058] Fig.11 Schematic diagram of the hysteresis phenomenon between air pressure and end displacement of a pneumatic manipulator;
[0059] Fig.12 This is a schematic diagram of the pressure-displacement modeling results based on the Bouc-Wen model;
[0060] Fig.13 This is the principle diagram of the experimental test of the end displacement and end contact force of the pneumatic soft robot arm;
[0061] Fig.14 This is a schematic diagram of the soft robot arm displacement and end force experimental platform;
[0062] Fig.15 Schematic diagram of Bouc-Wen modeling based on Hopfield at 0 kPa when verifying the displacement and end contact force hysteresis model of the pneumatic soft robot arm, where (a) is the Bouc-Wen modeling curve based on Hopfield, and (b) is the Bouc-Wen modeling error based on Hopfield.
[0063] Fig.16 Schematic diagram of Bouc-Wen modeling based on Hopfield at 40 kPa when verifying the displacement and end contact force hysteresis model of the pneumatic soft robot arm, where (a) is the Bouc-Wen modeling curve based on Hopfield, and (b) is the Bouc-Wen modeling error based on Hopfield;
[0064] Fig.17 Schematic diagram of the overall structure of the soft robotic arm, (a) is a top view, and (b) is a front view. DETAILED DESCRIPTION
[0065] The following examples will enable those skilled in the art to more fully understand the present invention, but are not intended to limit the present invention in any way.
[0066] The structural design of the soft robotic arm includes the pneumatic drive structure and the variable stiffness structure of the soft robotic arm. The overall structure of the soft robotic arm is described below.
[0067] The overall structure of the software driver is as follows Fig.17 As shown, Fig.17 (a) is a top view of the entire soft robotic arm. It can be seen from the top view that the entire robotic arm consists of three air cavities and a cavity coupled with a variable stiffness structure. The cavity of the blocked variable stiffness structure is located at the center of the soft robotic arm, which is used to reserve an assembly position for the variable stiffness structure. The three cavities are located in the deformation layer and are distributed at 120°. Fig.17(b) in the figure shows a front view of the soft robotic arm. There are three inflation pipes and one vacuum pipe at the end of the robotic arm. When inflated, the cavity pressure changes, thereby causing movement. There is an exhaust pipe. When the pneumatic soft robotic arm needs to change its stiffness, the vacuum degree of the blocking structure is changed by exhausting air, thereby changing the stiffness.
[0068] The embodiment of the present invention discloses a dynamic hysteresis modeling method for a pneumatic soft robot arm, wherein the pneumatic soft robot arm includes a pneumatic drive structure and a variable stiffness structure of the soft robot arm; the dynamic hysteresis modeling method includes the following steps:
[0069] S1, use MATLAB / Simulink software to build the Bouc-Wen hysteresis model:
[0070]
[0071] Among them, u(t) is the hysteresis between the air pressure and the end displacement of the pneumatic soft manipulator with uncoupled variable stiffness structure, and is the input air pressure value of the soft manipulator; is the derivative of u(t) with respect to time t; f d (t) is the output of the model, that is, the output displacement value of the soft robot arm; α, β, ρ, γ are unknown parameters that affect the shape of the model; the Bouc-Wen hysteresis model is composed of the addition of a linear part and a hysteresis part, the linear part corresponds to u(t), its weight is ρ, and the nonlinear hysteresis part corresponds to H(t);
[0072] S2, analyze the influence of each parameter change on the shape change of the Bouc-Wen curve; build a Hopfield neural network to obtain the parameter identification weight matrix of the Bouc-Wen hysteresis model;
[0073] S3, the parameter identification weight matrix of the Bouc-Wen hysteresis model is substituted into the Hopfield neural network, and the calculation is performed in MATLAB. After stabilization, the Bouc-Wen hysteresis model parameters to be identified are obtained.
[0074] The Bouc-Wen model is a differential equation expression with unknown parameters, as shown in formula (4.1):
[0075]
[0076] Where u(t) is the hysteresis between the air pressure and the end displacement of the pneumatic soft robot arm with uncoupled variable stiffness structure, and is the input air pressure value of the soft robot arm. is the derivative of u(t) with respect to time t, f d (t) is the output of the model, that is, the displacement value output by the soft robotic arm, and α, β, ρ, and γ are unknown parameters that affect the shape of the model.
[0077] The structure of the Bouc-Wen model can be viewed as Figure 1 The physical model shown in Figure 4 is composed of a linear part and a hysteresis part. The linear part corresponds to u(t) in formula (4.1), and its weight is ρ. The nonlinear hysteresis part corresponds to H(t) in formula (4.1). H(t) is obtained by get, is the derivative of H(t) with respect to time t.
[0078] The model of formula (4.1) is a continuous model. It needs to be converted into a discrete model during theoretical simulation and control. Formula (4.1) is converted into a discrete model as shown in formula 4.2: where ΔH(k) = H(k) - H(k-1), Δu(k) = u(k) - u(k-1):
[0079]
[0080] In order to study the influence of the parameters that need to be identified on the model, the simulation block diagram of the Bouc-Wen model is established using MATLAB / simulink, as shown in Figure 2 By changing the parameters α, β, ρ, γ in the simulation module, the influence of the parameters on the hysteresis loop can be observed.
[0081] The input signal y = sin(2t) is used as the input of the simulation. By changing the values of different parameters, the influence of the parameters on the hysteresis loop is studied. The results are as follows: Figure 3 As shown in the figure, as α increases, the shape of the entire hysteresis loop becomes wider and rotates counterclockwise. As β increases, the entire hysteresis loop also becomes wider, but contrary to the increase of α, it rotates clockwise as β increases. When the parameter γ increases, the width of the hysteresis loop does not change significantly, but there is a certain clockwise rotation as a whole. The parameter ρ does not change the shape of the hysteresis loop, but as ρ increases, the hysteresis loop rotates counterclockwise.
[0082] Similar to the PI model, after the hysteresis model is established, the model must be inverted to obtain the inverse hysteresis model, which is used for the feedforward control of the soft robot arm. Regarding the Bouc-Wen hysteresis model, for the Bouc-Wen hysteresis model shown in formula (4.1), it is rewritten as follows, where About Function:
[0083]
[0084] By inverting formula (4.3), we can get the Bouc-Wen inverse hysteresis model:
[0085]
[0086]
[0087]
[0088] In Simulink, the inverse model of the Bouc-Wen model can be directly established, such as Figure 4 shown.
[0089] In this embodiment, a Hopfield neural network is built in MATLAB, and the Hopfield neural network is used to perform online identification of the parameters of the Bouc-Wen hysteresis model. The hysteresis parameters of the Bouc-Wen model can be dynamically changed according to the hysteresis curve, thereby modeling the asymmetric hysteresis phenomenon in the pneumatic soft robot arm.
[0090] Hopfield neural network is a recursive neural network invented by John Hopfield. It can be used to solve optimization combinatorial problems. Hopfield neural network consists of multiple interconnected neural units, such as Figure 5 The figure shows a Hopfield neural network composed of 4 neurons. The Bouc-Wen hysteresis model based on this embodiment has 4 unknown parameters that need to be identified, so the Hopfield neural network composed of 4 neurons can complete the Bouc-Wen parameter identification work.
[0091] Since the neuron structures of Hopfield neural networks are similar, we take one neuron for analysis, such as Figure 6 Shown is the i-th neuron of the Hopfield neural network.
[0092] The input-output relationship can be established in the form of a differential equation:
[0093]
[0094] Where i = 1, 2, ..., n, n is the number of neurons, w ij is the weight coefficient of the jth neuron to the ith neuron. vi is the output of the neuron, I i is the input of the neuron, u i is the state of the neuron, g(·) is the activation function of the neuron, which is generally set to a hyperbolic function in the form of formula (4.8):
[0095]
[0096] The Lyapunov energy function of the Hopfield neural network is:
[0097]
[0098] Assume that the weight matrix w is a symmetric structure, that is, w ij =w ji ,but
[0099]
[0100] Because of w ij =w ji , we can get
[0101]
[0102] In the formula, c i > 0, g(s) is a hyperbolic function, and its inverse function is also a monotonically increasing function, so we can get therefore That is, the energy function has a negative gradient only when It can be seen that the output vector of the neural network is the equilibrium point of the network, that is, E n The smallest point.
[0103] The main purpose of using Bouc-Wen to model the soft robotic arm is to solve the four parameters of the Bouc-Wen model, α, β, ρ, and γ. Therefore, a Hopfield neural network consisting of four neurons is used. The principle diagram of Bouc-Wen model parameter identification is as follows: Figure 7 shown.
[0104] For the Bouc-Wen model, according to the Lyapunov energy function of the Hoprield neural network, the energy function is set as:
[0105]
[0106] Set the objective function of the model to
[0107]
[0108] For the Bouc-Wen model, the four output values v 1 , …, v 4 For α, β, ρ, γ, we can use formulas (4.12) and (4.13) to find partial derivatives of these four parameters and set them to 0 at the same time, and we can get
[0109]
[0110] Let the corresponding terms of formula (4.14) be equal, then we can get the matrix coefficients. For α(k), we have
[0111]
[0112] From equation (4.15), we can get w 11 , w 12 , w 13 , w 14 , I 1 The corresponding matrix coefficients for w 21 , w 22 , w 23 , w 24 , I 2 、w 31 , w 32 , w 33 , w 34 , I 3 、w 41 , w 42 , w 43 , w 44 , I 4 It can be obtained through the equations corresponding to β(k), ρ(k), and γ(k).
[0113] Final matrix coefficients
[0114]
[0115]
[0116] Among them, A and Figure 5 R i，j (i=1,2,3,4,j=1,2,3,4) corresponds to I k with I i (i=1, 2, 3, 4) Correspondingly, Formulas 4.16 and 4.17 can be used to build a Hopfield neural network based on Bouc-Wen parameter identification.
[0117] According to formulas (4.16) and (4.17), a Hopfield neural network can be built in MATLAB. In formula (4.16), u(k) is the model input at time k, Δu(k) is the model input at time k minus the model input at time k-1; H(k-1) is an intermediate variable, as shown in formula (4.2), H(k-1) = H(k-2) + ΔH(k), and H(k) is initialized at the beginning of the program.
[0118] The Bouc-Wen modeling flow chart based on Hopfield neural network is as follows Figure 8As shown, the Hopfield neural network parameter identification module has 6 inputs, namely Δu(k), H(k-1), Δu(k)*|Δu(k)|, H(k-1)*|H(k-1)|, u(k) and y(k). The first five outputs have the same meaning as in the previous paragraph, and y(k) is the output of the soft robot at time k. Using these 6 inputs combined with the matrices shown in formulas (4.16) and (4.17), the next output values of α, β, γ, and ρ can be obtained.
[0119] The Bouc-Wen module updates the Bouc-Wen model according to the four outputs α, β, γ, and ρ of the parameter identification module, and outputs the modeling output of the model, as well as the parameters Δu(k), H(k-1), Δu(k)*|Δu(k)|, H(k-1)*|H(k-1)|, and u(k) that the parameter identification module needs to input at the next moment.
[0120] The dynamic hysteresis modeling method of the Bouc-Wen model based on the Hopfield neural network proposed in this embodiment first introduces the Bouc-Wen model, builds the Bouc-Wen hysteresis model using MATLAB / Simulink software, and analyzes the influence of various parameter changes on the shape change of the Bouc-Wen curve. Then, the Hopfield neural network principle is introduced, and the stability of the neural network is proved by the Lyapunov principle, and the Bouc-Wen parameter identification weight matrix is obtained. The matrix is introduced into the neural network model, and the calculation is performed in MATLAB. After stabilization, the parameters to be identified can be obtained.
[0121] Performance Experiment
[0122] The control block diagram of the software robot arm is as follows: Fig. 9 As shown, the power module provides power for the control board and the air pump. The air pressure value in the robot arm is measured by the air pressure sensor and sent to the Arduino control board to adjust the air pressure. The PC / MATLAB host computer is connected to the Arduino control board. The PC transmits the control information to the Arduino control board, and then drives the solenoid valve through the relay. The control signal controls the solenoid switch valve to control the air pressure of the pneumatic actuator through the PWM duty cycle. For this article, due to the small number of experimental acquisition points and the large acquisition interval, the air pressure value after inverse hysteresis compensation corresponding to the expected displacement is first calculated in matlab, as shown in Fig. 9 The principle shown is then sent to the control board through the serial port to verify the effectiveness of hysteresis compensation.
[0123] According to the experimental principle, the corresponding experimental platform is built. The experiment based on the pressure and displacement in the air cavity has certain requirements on the control accuracy of the air pressure, so a reasonable air pressure controller is very important. The air pressure controller used is as follows Fig.10 As shown in the figure, the air pressure controller model is SCB-18PRO, and the air pressure adjustment range is -70-100kPa. Fig.10 As shown, the air pressure controller has an air inlet port connected to the air pump, an air outlet port that can be connected to the robotic arm, and a communication interface connected to the PC. In addition, there are I / O interfaces, status indicator lights and other devices. The control panel can be used to easily adjust the air pressure.
[0124] Through experimental testing, the relationship between air pressure and terminal elongation is obtained as follows: Fig.11 As shown, it can be seen that there is a large hysteresis between the rising stage and the falling stage.
[0125] For the Hopfield neural network principle and modeling simulation steps, the simulation results are as follows Fig.12 As shown in Figure 12, the blue curve is the measured hysteresis curve, and the red curve is the Bouc-Wen modeling curve. From Figure 12 (a), it can be seen that the Bouc-Wen model obtained based on the Hopfield neural network identification has a basically consistent trend with the hysteresis model. Fig.12 From the error diagram (b), it can be seen that the model error based on neural network parameter identification is in the range of [-1.1mm, 1.3mm] after stabilization.
[0126] The experimental test principle between the end displacement and end contact force of the pneumatic soft manipulator with coupled blocking variable stiffness structure is as follows Fig.13 As shown, the stiffness of the variable stiffness module of the soft robot arm is controlled by a vacuum controller, and the vacuum degree inside the variable stiffness module is controlled to 0kPa and 40kPa. The end of the soft robot arm is caused to produce the same displacement by a stepper motor, the stepper motor module is controlled by a guide rail controller, and the displacement sensor and the end force sensor respectively measure the end displacement and end force of the soft robot arm. By recording the displacement and end force data respectively, the hysteresis relationship between the end force and end displacement of the pneumatic soft robot arm can be obtained.
[0127] In order to measure the hysteresis relationship between the end displacement and the end contact force under the coupled blocked variable stiffness structure state, the schematic diagram of the end displacement and end contact force experimental platform is shown in the figure. Fig.14 The entire experimental platform consists of a soft robotic arm coupled with a blocking mechanism, a linear guide rail, a pressure sensor, a vacuum control device, a guide rail controller, and a stepper motor.
[0128] For the Hopfield neural network, the simulation modeling hysteresis curve results are as follows: Fig.15 As shown in (a) of Fig.15The error diagram (b) shows that the modeling error range is between [0.86N, 0.91N] after stabilization. Based on the Hopfield neural network, the simulation results at 40kPa are as follows Fig.16 As shown by Fig.16 From the modeling error diagram (b), it can be seen that the modeling error range after stabilization is between [-0.077N, 0.082N].
[0129] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.
Claims
1. A pneumatic soft robot arm dynamic hysteresis modeling method, characterized in that: The pneumatic soft robot arm meter includes a pneumatic drive structure and a variable stiffness structure of the soft robot arm; the dynamic hysteresis modeling method includes the following steps: S1, use MATLAB / Simulink software to build the Bouc-Wen hysteresis model: Among them, u(t) is the hysteresis between the air pressure and the end displacement of the pneumatic soft manipulator with uncoupled variable stiffness structure, and is the input air pressure value of the soft manipulator; is the derivative of u(t) with respect to time t; f d (t) is the output of the model, that is, the output displacement value of the soft robot arm; α, β, ρ, γ are unknown parameters that affect the shape of the model; the Bouc-Wen hysteresis model is composed of the addition of a linear part and a hysteresis part, the linear part corresponds to u(t), its weight is ρ, and the nonlinear hysteresis part corresponds to H(t); S2, analyze the influence of each parameter change on the shape change of the Bouc-Wen curve; build a Hopfield neural network to obtain the parameter identification weight matrix of the Bouc-Wen hysteresis model; S3, the parameter identification weight matrix of the Bouc-Wen hysteresis model is substituted into the Hopfield neural network, and the calculation is performed in MATLAB. After stabilization, the Bouc-Wen hysteresis model parameters to be identified are obtained.
2. The pneumatic soft robot arm dynamic hysteresis modeling method according to claim 1, characterized in that: In theoretical simulation and control, the Bouc-Wen hysteresis model is converted into a discrete model: Among them, ΔH(k)=H(k)-H(k-1), Δu(k)=u(k)-u(k-1).
3. The pneumatic soft robot arm dynamic hysteresis modeling method according to claim 1, characterized in that: In step S2, a simulation block diagram of the Bouc-Wen model is established using MATLAB or simulink. By changing the parameters α, β, ρ and γ in the simulation block diagram, the influence of each parameter change on the hysteresis loop is observed: As α increases, the shape of the entire hysteresis loop becomes wider and rotates counterclockwise; when β continues to increase, the entire hysteresis loop also becomes wider, and when β continues to increase, it rotates clockwise; the parameter γ has no effect on the shape of the hysteresis loop, and as ρ increases, the hysteresis loop rotates clockwise; the parameter ρ has no effect on the shape of the hysteresis loop, and as ρ increases, the hysteresis loop rotates counterclockwise.
4. The pneumatic soft robot arm dynamic hysteresis modeling method according to claim 1, characterized in that: In step S2, the Bouc-Wen hysteresis model is inverted to obtain the corresponding inverse hysteresis model for the feedforward control of the soft robot arm; specifically: The Bouc-Wen hysteresis model can be rewritten as: By finding the inverse function of the rewritten Bouc-Wen hysteresis model, we can obtain the inverse hysteresis model of the Bouc-Wen hysteresis model:
5. The pneumatic soft robot arm dynamic hysteresis modeling method according to claim 1, characterized in that: In step S2, the process of building a Hopfield neural network and obtaining a parameter identification weight matrix of the Bouc-Wen hysteresis model includes the following steps: For the unknown parameters α, β, ρ, γ, a Hopfield neural network consisting of 4 neurons is constructed; the input-output relationship function of each neuron is: Where i = 1, 2, 3, 4, w ij is the weight coefficient of the jth neuron to the ith neuron, v i is the output of the neuron, I i is the input of the neuron, u i is the state of the neuron, g(·) is the activation function of the neuron, which is set to a hyperbolic function: The energy function of the Bouc-Wen hysteresis model is set to: The objective function of the Bouc-Wen hysteresis model is set as: For the Bouc-Wen hysteresis model, the four output values v1, ..., v4 are α, β, ρ, γ. Taking the partial derivatives of these four parameters and making them all zero at the same time, we get: Let the corresponding terms be equal to obtain the matrix coefficients, and for α(k) we have the following solution equation: Solve to get w 11 , w 12 , w 13 , w 14 , the matrix coefficient corresponding to I1; Through the equations corresponding to β(k), ρ(k), and γ(k), we can get w 21 , w 22 , w 23 , w 24 , I2, w 31 , w 32 , w 33 , w 34 , I3, w 41 , w 42 , w 43 , w 44 , the matrix coefficient corresponding to I4; The matrix coefficients are calculated:
6. The pneumatic soft robot arm dynamic hysteresis modeling method according to claim 5, characterized in that: In step S3, the parameter identification weight matrix of the Bouc-Wen hysteresis model is substituted into the Hopfield neural network, and the calculation is performed in MATLAB. The process of obtaining the parameters of the Bouc-Wen hysteresis model to be identified after stabilization includes the following steps: S31, build a Hopfield neural network in matlab. The Hopfield neural network parameter identification module corresponds to 6 inputs, namely Δu(k), H(k-1), Δu(k)*|Δu(k)|, H(k-1)*|H(k-1)|, u(k) and y(k); among them, u(k) is the model input at time k, Δu(k) is the model input at time k minus the model input at time k-1; H(k-1) is an intermediate variable, H(k-1)=H(k-2)+ΔH(k), which assigns an initial value to H(k); y(k) is the output of the soft robot arm at time k; S32, substituting the six inputs into the parameter identification weight matrix to obtain the output values of α, β, γ, and ρ at the current moment; S33, using the output values of α, β, γ, and ρ at the current moment to update the Bouc-Wen model, update the output of the Bouc-Wen model and the six inputs of the parameter identification module at the next moment, and return to step S32 until the output accuracy of the Bouc-Wen model meets the accuracy requirements.