Backstepping control system of suspension system based on radial basis function neural network

By designing an inverse step control system based on radial-based neural network in the semi-active suspension system, combining a first-order low-pass filter and input saturation function module, the unknown body mass and vibration interference problems in the semi-active suspension system are solved. , improves the smoothness of the vehicle and ensures the stability of the system through the Lyapunov function analysis.

CN119078431APending Publication Date: 2024-12-06YANCHENG INST OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410406309.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-04-07
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

Traditional reverse step control methods cannot effectively solve the problems of unknown body mass, vibration interference and input limitation in semi-active suspension systems.

Method used

A reverse step control system based on radial-based neural network is designed to estimate spring-loaded mass and optimize suspension by defining state variables, introducing first-order low-pass filters and input saturation function modules, combining adaptive methods and auxiliary systems. Input damping force of the frame system.

Benefits of technology

This system can effectively solve the problems of unknown spring-loaded mass, vibration interference and input control force exceeding the limit value of the suspension system, improve the smoothness of the vehicle, and prove that the consistency of the system is finally achieved through the stability analysis of the Lyapunov function. The boundary is stable.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119078431A_ABST
    Figure CN119078431A_ABST
Patent Text Reader

Abstract

The invention discloses a backstepping control system of a suspension system based on a radial basis function neural network, which comprises a backstepping controller, the output of the backstepping controller is used as the input of a first-order low-pass filter, and the output of the first-order low-pass filter is used as the input of an input saturation function module. And the output of the input saturation function module is used as the input damping force of the suspension system. According to the method, the backstepping controller based on the radial basis function neural network is designed to solve the problem that the sprung mass of the suspension system is unknown. And then, a first-order low-pass filter is added on the basis of the backstepping controller, so that the problem of vibration interference of the suspension system is solved. And finally, an input saturation function module is added on the basis of the first-order low-pass filter, so that the problem that the input control force of the system exceeds a limit value is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of vehicle suspension system control, in particular to a backstepping control system of a suspension system based on a radial basis function neural network. Background Art

[0002] The function of the suspension is to transmit force and torque between the wheels and the frame to ensure the smooth driving of the motor vehicle. The suspension system mainly includes passive, semi-active and active suspension systems. Although the passive suspension system has a simple structure, its damping force is uncontrollable and cannot achieve the improvement of ride comfort and handling stability at the same time; the active suspension system has a complex structure, high cost and high power consumption, so its market promotion is greatly limited; in contrast, the semi-active suspension system provides ideal performance under moderate power consumption and cost, and is therefore widely used. Among the many semi-active suspensions, the magnetorheological semi-active suspension has excellent characteristics such as fast response, large capacity and low power consumption. It can provide excellent robustness in semi-active control and improve the ride comfort of drivers and passengers. Therefore, the research on magnetorheological semi-active suspension systems has received more and more attention. The traditional backstepping control method cannot solve the problems of unknown body mass, vibration interference, and limited input in the semi-active suspension system. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a backstepping control system of a suspension system based on a radial basis function neural network.

[0004] The technical solution adopted by the present invention to solve the technical problem is: a backstepping control system of a suspension system based on a radial basis neural network, including a backstepping controller, defining a state variable: x1 = z s , x3=z us , The state space expression of the semi-active suspension system of a quarter vehicle is:

[0005]

[0006]

[0007] Among them, m s is the sprung mass, m us is the unsprung mass, z s is the displacement of the sprung mass, z us is the displacement of the unsprung mass; z t is the disturbance of the road surface; k is the suspension stiffness coefficient, k t is the tire stiffness coefficient, c is the suspension damping coefficient, c t is the tire damping coefficient; u is the input damping force of the suspension system;

[0008] The output of the backstepping controller is used as the input of a first-order low-pass filter, the output of the first-order low-pass filter is used as the input of an input saturation function module, and the output of the input saturation function module is used as the input damping force of the suspension system;

[0009] The design method of the backstepping controller is as follows:

[0010] 1) Design tracking error e1: e1 = x1-x 1r , x 1r is a continuously differentiable reference trajectory;

[0011] 2) Introduce virtual controller α: t1 is a positive constant greater than zero;

[0012] 3) Filter signal of first-order low-pass filter γ1 is a constant greater than zero;

[0013] 4) Define filtering error

[0014] 5) Design auxiliary signal tracking error

[0015] 6) Use radial basis function neural network to approximate the sprung mass m s Function And an adaptive method is used to estimate the weights in the neural network:

[0016]

[0017] in, is the estimated value of H1(X1), ε is the approximate error value, X1 is the input of the radial basis neural network; vector w1 is the ideal weight value of the neural network, is the estimated value of w1, and the basis function vector φ1(X1) of the radial basis neural network selects the Gaussian basis function;

[0018] 7) Design the actual output u1 of the backstepping controller:

[0019]

[0020] F z =F k +F c ;

[0021] t2 is a preset parameter, t2>0.

[0022] Preferably, the backstepping controller is an adaptive backstepping controller, and the adaptive law of the backstepping controller is: Λ1 is a positive definite symmetric matrix, r1 is a preset parameter, and r1>0.

[0023] Preferably, the input damping force u of the suspension system satisfies the following saturation function by inputting a saturation function module:

[0024] u max It is the maximum active control force of the suspension system.

[0025] Preferably, an auxiliary system is introduced into the input saturation function module:

[0026]

[0027] η is the function of the auxiliary system, is the derivative of the auxiliary system function,

[0028] Wherein, p1>0 and p2>0 are preset parameters.

[0029] Preferably, the suspension system is a magnetorheological semi-active suspension system, and the model of the magnetorheological damper in the magnetorheological semi-active suspension system is:

[0030]

[0031] In the formula, k0 is the stiffness coefficient of the damper, c0 is the damping coefficient of the damper, x is the displacement of the damper, and F MR is the damping force of the damper, F d is the Coulomb force generated by the external magnetic field, f0 is the damping force caused by the deviation between measurement and sealing, and A affects the shape of the hysteresis loop.

[0032] The beneficial effects of the present invention are as follows: In the present invention, a backstepping controller based on a radial basis neural network is designed to solve the problem of unknown sprung mass of the suspension system. Then, a first-order low-pass filter is added on the basis of the backstepping controller to solve the vibration interference problem of the suspension system. Finally, an input saturation function module is added on the basis of the first-order low-pass filter to solve the problem that the system input control force exceeds the limit value, and the system stability of the input saturation function is ensured by the auxiliary system. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a principle block diagram of a backstepping control system of a suspension system based on a radial basis function neural network of the present invention;

[0034] Figure 2 It is the simulation diagram of input excitation of Class A road surface;

[0035] Figure 3 This is the B-level road input excitation simulation diagram;

[0036] Figure 4 This is the C-level road input excitation simulation diagram;

[0037] Figure 5 It is the vertical displacement curve of sprung mass with different controls under the conditions of Class A road and vehicle speed of 35km / h;

[0038] Figure 6 It is the vertical velocity curve of sprung mass with different controls under the condition of Class A road and vehicle speed of 35km / h;

[0039] Figure 7 It is the vertical displacement curve of sprung mass with different controls under the condition of Class B road and vehicle speed of 35km / h;

[0040] Figure 8 It is the vertical velocity curve of sprung masses with different controls under the condition of Class B road and vehicle speed of 35km / h; DETAILED DESCRIPTION

[0041] The present invention will now be further described in conjunction with the accompanying drawings. These drawings are simplified schematic diagrams that only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0042] like Figure 1 As shown, a backstepping control system of a suspension system based on a radial basis neural network includes a backstepping controller, and defines a state variable: x1 = z s , x3=z us , The state space expression of the semi-active suspension system of a quarter vehicle is:

[0043]

[0044]

[0045] Among them, m s is the sprung mass, m us is the unsprung mass, z s is the displacement of the sprung mass, z us is the displacement of the unsprung mass; z t is the disturbance of the road surface; k is the suspension stiffness coefficient, k t is the tire stiffness coefficient, c is the suspension damping coefficient, c t is the tire damping coefficient; u is the input damping force of the suspension system;

[0046] The output of the backstepping controller is used as the input of a first-order low-pass filter, the output of the first-order low-pass filter is used as the input of an input saturation function module, and the output of the input saturation function module is used as the input damping force of the suspension system;

[0047] The design method of the backstepping controller is as follows:

[0048] 1) Design tracking error e1: e1 = x1-x 1r , x 1r is a continuously differentiable reference trajectory;

[0049] 2) Introduce virtual controller α: t1 is a positive constant greater than zero;

[0050] 3) Filter signal of first-order low-pass filter γ1 is a constant greater than zero;

[0051] 4) Define filtering error

[0052] 5) Design auxiliary signal tracking error

[0053] 6) Use radial basis function neural network to approximate the sprung mass m s Function And an adaptive method is used to estimate the weights in the neural network:

[0054]

[0055] in, is the estimated value of H1(X1), γ is the approximate error value, X1 is the input of the radial basis neural network; vector w1 is the ideal weight value of the neural network, is the estimated value of w1, and the basis function vector φ1(X1) of the radial basis neural network selects the Gaussian basis function;

[0056] 7) Design the actual output u1 of the backstepping controller:

[0057]

[0058] F z =F k +F c ;

[0059] t2 is a preset parameter, t2>0.

[0060] Specifically, in an optional implementation manner, the backstepping controller is an adaptive backstepping controller, and the adaptive law of the backstepping controller is: Λ1 is a positive definite symmetric matrix, r1 is a preset parameter, and r1>0.

[0061] Specifically, in an optional implementation, the input damping force u of the suspension system satisfies the following saturation function by inputting a saturation function module:

[0062] u max It is the maximum active control force of the suspension system.

[0063] Specifically, in an optional implementation manner, an auxiliary system is introduced into the input saturation function module:

[0064]

[0065] η is the function of the auxiliary system, is the derivative of the auxiliary system function,

[0066] Where p1>0 and p2>0 are preset parameters. The auxiliary system is added to the input saturation function in order to carry out the stability analysis of the Lyapunov function in the next step.

[0067] Specifically, in order to simplify the calculation, let F z =F k +F c , the state space expression of the suspension system can be changed to:

[0068] Specifically, in the radial basis neural network,

[0069] The basis function vector of the neural network φ1(X1)=[φ 11 ,φ 12 , ..., φ 1l ] T ,

[0070] c ij is the center of the Gaussian function, b i is the width of the Gaussian function;

[0071] The estimated error of H1(X1) is:

[0072]

[0073] In the formula, ε is The approximate error of

[0074] The ideal weight value of the neural network is: w1 = [w 11 ,w 12 ,...,w 1l ] T ;

[0075] make is the vector of weight errors, we can get:

[0076] This formula is used for stability analysis of Lyapunov functions.

[0077] Specifically, in an optional implementation, the suspension system is a magnetorheological semi-active suspension system, and the model of the magnetorheological damper in the magnetorheological semi-active suspension system is:

[0078]

[0079] In the formula, k0 is the stiffness coefficient of the damper, c0 is the damping coefficient of the damper, x is the displacement of the damper, and F MR is the damping force of the damper, F d is the Coulomb force generated by the external magnetic field, f0 is the damping force caused by the deviation between measurement and sealing, and A affects the shape of the hysteresis loop.

[0080] The magnetorheological damper was subjected to tensile test using sinusoidal displacement excitation, with an amplitude of 35 mm and a frequency of 0.25 Hz; the current I was 0 A, 0.2 A, 0.4 A, 0.6 A, 0.8 A, and 1 A respectively. The parameter identification data are shown in Table 1:

[0081] Table 1 Parameter identification of magnetorheological damper

[0082] Current / parameters <![CDATA[k0]]> <![CDATA[c0]]> <![CDATA[F d ]]> <![CDATA[f0]]> A 0A 1.3973 27.9193 -10.3911 -6.2874 0.0343 0.2A 1.0448 11.0423 223.2827 -15.0510 363.7912 0.4A 0.9842 16.8923 317.0771 -24.0928 247.1038 0.6A 0.7839 22.6469 358.9706 -107.4778 191.0610 0.8A 0.6387 37.5727 418.8581 49.0271 136.0640 1A 0.6534 56.6566 454.4479 -128.5164 108.1995

[0083] Take the average value of the parameter f0, f0 = -21.5538, and use linear functions to fit the remaining parameters. The fitting functions of the four parameters are:

[0084]

[0085] In the present invention, a backstepping controller based on a radial basis neural network is designed to solve the problem of unknown sprung mass of the suspension system. Then, a first-order low-pass filter is added on the basis of the backstepping controller to solve the vibration interference problem of the suspension system. Finally, an input saturation function module is added on the basis of the first-order low-pass filter to solve the problem that the system input control force exceeds the limit value, and the system stability of the input saturation function is ensured by the auxiliary system.

[0086] On the basis of the backstepping controller, a radial basis neural network, a first-order low-pass filter and an input saturation function module are introduced to improve the ride smoothness of the vehicle. The backstepping control system of the present invention is applied to the suspension system of the vehicle, which can greatly improve the problem that the nonlinear factors of the suspension system under certain load excitation and frequency domain range will have a huge impact on the control effect. Based on the stability analysis of the Lyapunov function, it is proved that the backstepping control system of the present invention is uniformly ultimately bounded and stable.

[0087] The stability analysis of the Lyapunov function of the backstepping control system of the present invention is as follows:

[0088] The time derivative of e1 is:

[0089] The first Lyapunov function:

[0090] The time derivative of e2 is:

[0091] The second Lyapunov function: Where Λ1 is a positive definite symmetric matrix.

[0092] Derivative of V2:

[0093]

[0094] make There exists a constant p3 such that |M(·)|<p3,

[0095] Available,

[0096]

[0097]

[0098] A2>A1>0,

[0099] State variable |x1|≤|e1|+|x 1r |,

[0100]

[0101] This means that all signals in the semi-active suspension system are bounded. The backstepping control system of the suspension system of the present invention is uniformly ultimately bounded and stable, and can ensure that the signals of the semi-active suspension system are within the required boundaries.

[0102] In order to verify the control performance of the backstepping control system of the present invention, passive control (without backstepping control system, only the magnetorheological damper outputs damping force to the suspension system), backstepping control (conventional backstepping controller), and backstepping control system based on radial basis function neural network (the present invention) are simulated and verified. The selection of suspension parameters is shown in Table 2. In the radial basis function neural network, the neural network parameter c ij and b i The values ​​are [-2 -1.5 -1 -0.5 0 0.5 1 1.5 2] and 3.0, and the initial weights of the radial basis neural network are set to 0. Two different working conditions are selected for the simulation, and the simulation time is selected to be 10 seconds.

[0103] Table 2 Parameters of 1 / 4 vehicle suspension system

[0104]

[0105] The specific information of the simulation conditions is as follows:

[0106] Select filtered white noise as the road interference signal, and the time domain mathematical model of filtered white noise is:

[0107]

[0108] in, is a constant coefficient, v is the vehicle speed, selected as 35km / h, and w(t) is Gaussian white noise.

[0109] The power spectral density of Gaussian white noise is:

[0110] G q (f) = 4π 2 G q (n0)n0 2 v;

[0111] Among them, G q (n0) is the road roughness coefficient. The power spectrum density of A-, B-, and C-grade roads is 6.1411×10-5m2 / s, 2.4564×10-4m2 / s, and 9.8257×10-4m2 / s, respectively. The input excitation simulation of A-, B-, and C-grade roads is as follows: Figure 2 , Figure 3 , Figure 4 shown.

[0112] like Figure 5 , Figure 7 As shown, in the vertical displacement of the sprung mass, the backstepping control system of the present invention is used, and the oscillation amplitude of the displacement is small, which is significantly improved compared with the passive control; Figure 6 , Figure 8As shown, in the vertical speed of the sprung mass, the backstepping control system of the present invention has a smaller oscillation amplitude of the speed, which is significantly improved compared with the passive control, thus indicating that the suspension system travels more smoothly.

[0113] Table 3 shows the RMS values ​​of acceleration under different road excitations and different controls;

[0114] Table 4 shows the peak acceleration under different road excitations and different controls;

[0115] Table 3 RMS value of acceleration

[0116]

[0117] Table 4 Peak acceleration

[0118]

[0119] By analyzing Table 3 and Table 4, the performance of the suspension system can be determined. As shown in Table 3 and Table 4, the back-stepping control system of the present invention is used to act on the suspension system of the vehicle, and the root mean square value of the acceleration of the suspension system is increased by 52% on Class A roads, 46% on Class B roads, and 38% on Class C roads. At the same time, the back-stepping control system of the present invention is used to act on the suspension system of the vehicle, which greatly reduces the peak acceleration, that is, the peak acceleration of Class A roads is increased by 51%, the peak acceleration of Class B roads is increased by 47%, and the peak acceleration of Class C roads is increased by 33%. The simulation results show that the back-stepping control system of the present invention is used to act on the suspension system of the vehicle, which improves the ride smoothness.

[0120] The above-mentioned implementation modes are only for illustrating the technical concept and features of the present invention, and their purpose is to enable people familiar with this technology to understand the contents of the present invention and implement them. They cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made according to the spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. A backstepping control system for a suspension system based on a radial basis neural network, including a backstepping controller, defining a state variable: x1 = z s , x3=z us , The state space expression of the semi-active suspension system of a quarter vehicle is: in, m s is the sprung mass, m us is the unsprung mass, z s is the displacement of the sprung mass, z us is the displacement of the unsprung mass; z t is the disturbance of the road surface; k is the suspension stiffness coefficient, k t is the tire stiffness coefficient, c is the suspension damping coefficient, c t is the tire damping coefficient; u is the input damping force of the suspension system; It is characterized in that: the output of the backstepping controller is used as the input of the first-order low-pass filter, the output of the first-order low-pass filter is used as the input of the input saturation function module, and the output of the input saturation function module is used as the input damping force of the suspension system; The design method of the backstepping controller is as follows: 1) Design tracking error e1: e1 = x1-x 1r , x 1r is a continuously differentiable reference trajectory; 2) Introduce virtual controller α: t1 is a positive constant greater than zero; 3) Filter signal of first-order low-pass filter γ1 is a constant greater than zero; 4) Define filtering error 5) Design auxiliary signal tracking error 6) Use radial basis function neural network to approximate the sprung mass m s Function And an adaptive method is used to estimate the weights in the neural network: in, is the estimated value of H1(X1), ε is the approximate error value, X1 is the input of the radial basis neural network; vector w1 is the ideal weight value of the neural network, is the estimated value of w1, and the basis function vector φ1(X1) of the radial basis neural network selects the Gaussian basis function; 7) Design the actual output u1 of the backstepping controller: F z =F k +F c ; t2 is a preset parameter, t2>0.

2. The suspension backstepping control system based on radial basis function neural network according to claim 1, characterized in that: The backstepping controller is an adaptive backstepping controller, and the adaptive law of the backstepping controller is: Λ1 is a positive definite symmetric matrix, r1 is a preset parameter, and r1>

0.

3. The suspension backstepping control system based on radial basis function neural network according to claim 1, characterized in that: By inputting the saturation function module, the input damping force u of the suspension system satisfies the following saturation function: u max It is the maximum active control force of the suspension system.

4. The suspension backstepping control system based on radial basis function neural network according to claim 3, characterized in that: The input saturation function module introduces an auxiliary system: η is the function of the auxiliary system, is the derivative of the function of the auxiliary system. Wherein, p1>0 and p2>0 are preset parameters.

5. The suspension backstepping control system based on radial basis function neural network according to claim 1, characterized in that: The suspension system is a magnetorheological semi-active suspension system, and the model of the magnetorheological damper in the magnetorheological semi-active suspension system is: In the formula, k0 is the stiffness coefficient of the damper, c0 is the damping coefficient of the damper, x is the displacement of the damper, and F MR is the damping force of the damper, F d is the Coulomb force generated by the external magnetic field, f0 is the damping force caused by the deviation between measurement and sealing, and A affects the shape of the hysteresis loop.