Distributed driving vehicle preset performance fault-tolerant control method and system based on adaptive neural network observer

Abstract

The invention provides a distributed driving vehicle preset performance fault-tolerant control method and system based on an adaptive neural network observer. The method comprises the following steps: constructing a vehicle dynamics model containing motor faults and unknown disturbance; in order to realize real-time estimation of faults and disturbance and break through the limitation on disturbance derivative boundaries, a self-adaptive neural network observer is designed; aiming at the problem that the control direction is uncertain, an upper-layer controller is designed based on a preset performance control theory, a Nussbaum function and an obstacle Lyapunov function are introduced, it is ensured that system dynamic errors meet preset performance constraints all the time, and accurate tracking control over the transverse force, the longitudinal force and the yawing moment of a vehicle is achieved; in order to improve the system coordination capability under the multi-motor fault working condition, an optimal torque distribution model fusing the motor torque constraint and the tire force limit is constructed, collaborative optimization control of the driving and steering system is realized through a quadratic programming algorithm, and the stability and safety of the whole vehicle in the fault state are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of distributed drive vehicle control, and in particular to a method and system for fault-tolerant control of preset performance of a distributed drive vehicle based on an adaptive neural network observer. Background Art

[0002] Compared to centralized drive systems, distributed drive vehicles break free from the constraints of mechanical transmission system layout, effectively improving drive efficiency and reducing energy consumption. Furthermore, distributed drive independently controls the driving torque at each wheel, resulting in enhanced low-speed maneuverability and high-speed stability. This drive method not only improves the vehicle's maneuverability on harsh roads but also enables compound steering through the synergy of differential steering technology and the power steering motor, providing redundancy for the steering system.

[0003] However, the complex system design and large number of components of distributed drive vehicles will bring a higher risk of potential failures, including actuator failure and sensor failure. When the steering motor or drive motor fails, the steering assist will be weakened or additional yaw torque will be generated due to the imbalance of driving force on both sides, which will in turn affect the vehicle's yaw stability. If effective fault-tolerant control measures are not taken, the vehicle's tracking performance will be seriously affected, resulting in deviation from the expected trajectory, which seriously threatens driving safety. Active fault-tolerant control can effectively compensate for the impact of faults, maintain stable system operation, and significantly improve vehicle safety and robustness through real-time fault diagnosis and dynamic control strategy adjustment. However, the output capacity degradation caused by motor failure has time-varying characteristics, and there is a strong dynamic coupling between the distributed drive system and the active steering system, which significantly increases the risk of vehicle instability under fault conditions, which places higher requirements on fault-tolerant control strategies.

[0004] At present, there are still some problems in the fault-tolerant control of distributed drive vehicles, mainly as follows:

[0005] 1. Relying on accurate vehicle dynamics models or preset thresholds makes it difficult to effectively handle system parameter uncertainties and unmodeled dynamics, and difficult to adapt to complex and changing fault types and disturbances in real time.

[0006] 2. It is difficult to handle the complex coupling problem of coordinated failure of multiple motors in distributed drive vehicles. It does not consider the stability when the compensation control direction is unknown (for example, instability may occur when the control gain changes due to partial failure of the motor), and cannot adaptively adjust the control gain.

[0007] 3. The inherent chattering phenomenon of the commonly used sliding mode control will directly affect the control effect.

[0008] 4. Lack of dynamic performance constraints and insufficient optimization of multi-physics constraints. Summary of the Invention

[0009] (1) Technical issues to be resolved

[0010] Based on this, the present invention provides a method and system for fault-tolerant control of preset performance of distributed drive vehicles based on an adaptive neural network observer, which effectively improves the robustness and safety of distributed drive vehicles under complex faults and disturbances.

[0011] (2) Technical solution

[0012] To achieve the above objectives, the present invention provides a method for fault-tolerant control of preset performance of a distributed drive vehicle based on an adaptive neural network observer, comprising:

[0013] S1: Establish a vehicle dynamics model that includes faults and disturbances; specifically,

[0014] S101: Establish the three-degree-of-freedom vehicle dynamics equation;

[0015] S102: Establishing a tire model of the vehicle;

[0016] S103: Establishing a wheel dynamics model of the vehicle;

[0017] S104: Establishing a vehicle fault model;

[0018] S105: Obtaining a vehicle dynamics model including faults and disturbances;

[0019] S2: Design an adaptive neural network observer; specifically including:

[0020] S201: Real-time estimation of faults and disturbances using a radial basis function neural network observer;

[0021] S202: Verify the stability of the radial basis function neural network observer;

[0022] S3: Design a preset performance controller as the upper-level controller; specifically, it includes:

[0023] S301: defining a tracking error and defining a preset performance function to set an error boundary;

[0024] S302: Constructing barrier Lyapunov functions;

[0025] S303: Introducing Nussbaum function to deal with the unknown direction of control gain;

[0026] S304: Designing a preset performance controller based on the barrier Lyapunov function;

[0027] S304: Perform stability analysis;

[0028] S4: Establishing a lower controller to realize torque distribution; specifically including:

[0029] S401: Establishing vehicle tire adhesion constraints and motor output constraints;

[0030] S402: Designing an optimization objective function to perform torque distribution.

[0031] On the other hand, the application provides a distributed drive vehicle preset performance fault-tolerant control system based on an adaptive neural network observer, comprising a vehicle information acquisition module, a calculation module and an execution module; the vehicle information acquisition module comprises a vehicle state sensor, a GPS and an inertial navigation system, for collecting vehicle motion parameters and position information in real time and transmitting the vehicle motion parameters and position information to the calculation module; the calculation module comprises a high-performance processor, for running a neural network observer, a preset performance controller and a torque distribution algorithm according to the vehicle motion parameters and position information, and transmitting the obtained torque distribution result to the execution module; the execution module comprises four independently driven wheel hub motors and an electric steering system, for realizing torque and steering control of the vehicle according to the torque distribution result of the calculation module.

[0032] (III) Beneficial effects

[0033] From the above technical solution, the application provides a distributed drive vehicle preset performance fault-tolerant control method and system based on an adaptive neural network observer, which has the beneficial effects that:

[0034] 1. A vehicle dynamics model containing motor fault modes and unknown disturbances is constructed, and a hierarchical control architecture is used to process the strong coupling of four-wheel drive and steering.

[0035] 2. An adaptive neural network observer based on RBF is designed to realize high-precision real-time estimation of unknown faults and disturbances, breaking through the limitation of the boundedness of the derivative of the disturbance in traditional methods. The observer significantly improves the accuracy and real-time performance of fault detection by combining the joint observation algorithm of motor fault modes and unknown disturbances with the weight adaptive update law with finite time convergence characteristics.

[0036] 3. Nussbaum function and preset performance control are combined to solve the stability problem when the control direction is uncertain due to partial failure of the motor; the obstacle Lyapunov function is used to strictly constrain the tracking error within the preset safety boundary, avoiding the risk of vehicle trajectory deviation.

[0037] 4. Based on a multi-objective optimization algorithm, a torque distribution strategy that integrates tire adhesion constraints and motor output capabilities is used to achieve optimal torque distribution in fault conditions, significantly improving the vehicle stability and energy utilization efficiency in fault conditions.

[0038] 5. A three-layer collaborative architecture of observation-control-distribution has been constructed. Through the seamless switching logic of fault detection and fault-tolerant control, a complete fault-tolerant control solution has been formed, ensuring the reliability and stability of the system in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The features and advantages of the present invention will be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the present invention in any way. In the accompanying drawings:

[0040] Figure 1 This is a schematic diagram of the principle of the fault-tolerant control of the preset performance of a distributed drive vehicle based on an adaptive neural network observer according to the present invention;

[0041] Figure 2 A schematic diagram of a distributed drive vehicle model of the present invention;

[0042] Figure 3 Comparison charts of vehicle tracking performance under the single lane change condition of the present invention; (a) shows a comparison chart of longitudinal velocity; (b) shows a comparison chart of lateral velocity; (c) shows a comparison chart of yaw rate; (d) shows a comparison chart of longitudinal velocity error; (e) shows a comparison chart of lateral velocity error; (f) shows a comparison chart of yaw rate error;

[0043] Figure 4 A comparison diagram of unknown disturbance and fault estimation values ​​between the adaptive neural network observer and the sliding mode observer of the present invention;

[0044] Figure 5 This is a comparison chart of the trajectory tracking results of the present invention. DETAILED DESCRIPTION

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0046] Example 1:

[0047] like Figure 1 As shown, the present invention provides a method for fault-tolerant control of preset performance of a distributed drive vehicle based on an adaptive neural network observer, comprising:

[0048] S1: Establish a vehicle dynamics model including fault models and unknown disturbances;

[0049] S101: Establish the three-degree-of-freedom vehicle dynamics equation;

[0050] Four-wheel independent drive vehicle (distributed drive vehicle) model Figure 1 As shown, ignoring pitch and roll motion, only considering the front wheel steering. According to Newton's second law, the three-degree-of-freedom dynamic equation of the vehicle is constructed as follows:

[0051]

[0052] Among them are:

[0053] F X =(F xfl +F xfr )cosδ+F xrl +F xrr -(F yfl +F yfr )sinδ;

[0054] F Y =(F xfl +F xfr )sinδ+(F yfl +F yfr )cosδ+F yrl +F yrr ;

[0055]

[0056] In the above formula, v lon 、v lat 、ω veh Respectively represent the longitudinal velocity, lateral velocity, and yaw rate of the vehicle; Respectively represent the longitudinal acceleration, lateral acceleration, and yaw angular acceleration of the vehicle; F X 、F Y 、M Z They represent the longitudinal force, lateral force, and yaw moment of the vehicle respectively; F xfl 、F xfr 、F xrl 、F xrr Respectively represent the longitudinal forces acting on the left front wheel, right front wheel, left rear wheel, and right rear wheel; l f 、l r Respectively represent the distance from the front and rear axles to the center of mass; l s Indicates half of the wheelbase; I Z and M represent the vehicle’s moment of inertia and vehicle mass respectively; δ represents the front wheel turning angle; C arepresents the aerodynamic drag coefficient; d1(t), d2(t), and d3(t) are external disturbances that represent load changes, parameter perturbations, and unmodeled terms in the vehicle system, such as crosswind and tire rolling resistance. The external disturbances d1(t), d2(t), and d3(t) are assumed to be bounded.

[0057] S102: Establishing a tire model of the vehicle;

[0058] When a vehicle is driving normally, lateral acceleration generally does not exceed 0.4g (g represents the acceleration due to gravity). At this point, the tire's cornering characteristics can be considered to be in the linear domain, meaning that the slip angle and cornering force are linearly related. Furthermore, assuming the slip angles of the coaxial left and right tires are equal, we have:

[0059]

[0060] In the above formula: F yf 、F yr Represents the lateral force of the front and rear axles of the vehicle respectively; C f 、C r Represents the cornering stiffness of the front and rear tires respectively; α f , α r Respectively represent the side slip angles of the front and rear tires; expressed as follows:

[0061]

[0062] According to formulas (2) and (3), we have

[0063] F y =F yc +F y (δ) (4)

[0064] In the above formula, F y represents the wheel lateral force matrix, F y =[F yfl ,F yfr ,F yrl ,F xrr ] T ;

[0065]

[0066] F y (δ)=[C f δ,C f δ,0,0] T .

[0067] S103: Establishing a wheel dynamics model of the vehicle;

[0068] The vehicle's four-wheel independent drive system consists of four motors, and the torque of the i-th motor is T iIt can be expressed as:

[0069] T i =μ i u i (5)

[0070] In the above formula, the subscripts i=1, 2, 3, and 4 represent the left front wheel, right front wheel, left rear wheel, and right rear wheel respectively; T i represents the i-th motor torque (for example, T1 represents the motor torque of the left front wheel); u i represents the torque control signal of the i-th motor (for example, u1 represents the motor torque control signal of the left front wheel); μ i represents the control gain of the i-th motor.

[0071] The wheel dynamics model can be described as:

[0072]

[0073] In the above formula, R f Indicates the rolling radius of the wheel; F xi Represents the longitudinal forces of the four wheels, (e.g. F x1 represents the longitudinal force of the left front wheel); I W represents the wheel moment of inertia; ω i represents the rotational angular velocity of wheel i (for example, ω1 represents the rotational angular velocity of the left front wheel).

[0074] S104: Establishing a vehicle fault model;

[0075] Vehicle motor faults generally include multiplicative faults and additive faults, which can be uniformly modeled as follows:

[0076] u i =λ i u di +Δu i i∈{1,2,3,4,5} (7)

[0077] In the above formula, the subscripts i=1, 2, 3, and 4 represent the left front wheel, right front wheel, left rear wheel, and right rear wheel respectively; u1, u2, u3, and u4 represent the motor torque control signals of the left front wheel, right front wheel, left rear wheel, and right rear wheel respectively, which are the actual outputs of the i-th motor; u5 represents the front wheel angle; u di , i=1,2,3,4 represents the expected output of the i-th motor; u d5 represents the expected value of the front wheel turning angle; Δu i represents the disturbance caused by the fault; λ i represents the failure loss factor, and λ i ∈[0,1]; the motor output has upper and lower bounds, i.e. u i ∈[u imin,u imax ],u imin is the minimum output value of motor i, u imax is the maximum output of motor i.

[0078] Define the output of the lower controller of the system as u=[u1,u2,u3,u4,u5] T , λ=diag{λ1, λ2, λ3, λ4, λ5}.

[0079] The above fault model can describe the following motor failure modes:

[0080] (1) Multiplicative fault - partial failure mode: indicates abnormal response to the control signal, that is, 0 < λ i <1,Δu i =0.

[0081] (2) Multiplicative fault - complete failure mode: Indicates that the motor fails completely and no longer outputs torque, i.e., λ i =0,Δu i =0.

[0082] (3) Additive fault mode: It means that the performance of the motor deviates from the normal working state during operation, that is, λ i =0,Δu i ≠0.

[0083] The control signal of the motor torque when there is no fault is the expected output of the motor, λ i =1,Δu i = 0. In order to analyze the impact of motor output on the vehicle, the wheel longitudinal force matrix F is defined x =[F xfl ,F xfr ,F xrl ,F xrr ] T According to formulas (5)-(7), we can get

[0084]

[0085] S105: Acquire a vehicle dynamics model including the fault model and the unknown disturbance.

[0086] In order to introduce the unknown disturbance ξ1(t) based on the fault model and derive the vehicle dynamics model, it is defined as follows:

[0087] x=[v lon ,v lat ,ω veh ] T , F1(x)=[f1(x),f2(x),f3(x)] T ,

[0088]

[0089]

[0090]

[0091] Substituting formulas (2)-(8) into formula (1), we obtain the three-degree-of-freedom vehicle dynamics model including faults and unknown disturbances:

[0092]

[0093] In the above formula, U=B2(δ)Ku,

[0094]

[0095]

[0096]

[0097] According to formula (6), we can get

[0098]

[0099] Define ω=[ω1,ω2,ω3,ω4] T =[ω fl ,ω fr ,ω rl ,ω rr ] T ,

[0100]

[0101] Since the longitudinal and lateral accelerations and yaw angular velocity are bounded in the vehicle system, and the motor output u i is also bounded, so we can see that ξ1(t) is bounded.

[0102] S2: Design an adaptive neural network observer;

[0103] S201: Real-time estimation of faults and disturbances using a radial basis function neural network observer;

[0104] RBF has the ability to approximate unknown nonlinear functions and is often used to compensate for modeling errors and external disturbances, thereby improving the control performance of the system. Therefore, the present invention designs a radial basis function neural network (RBF neural network) observer to estimate unknown faults. and disturbances (In this article, the parameter with ∧ indicates the estimated value of the parameter, and the parameter with · indicates the derivative of the parameter). The expression is:

[0105]

[0106]

[0107] In the above formula, X1 and X2 represent the input of the neural network; Represents the neural network weight matrix; h(x) represents the output of hidden layer neurons, and the specific form is:

[0108]

[0109] Among them, c k represents the center point of the kth basis function of the hidden layer; b k Indicates the width of the kth basis function in the hidden layer. Considering the modeling error of the RBF neural network, the actual unknown fault g(t) and the disturbance ξ(t) can be modeled as:

[0110]

[0111]

[0112] In the above formula, ε1 and ε2 represent the estimation errors of unknown faults and disturbances respectively, assuming that their upper bounds satisfy ||ε1||≤ε M1 ,||ε2||≤ε M2 .

[0113] The RBF neural network estimation method used in this invention does not require boundedness assumptions for derivatives of unknown faults and disturbances. It is highly adaptable and suitable for highly uncertain working conditions, helping to improve system estimation accuracy and control robustness. Based on the estimation results, a neural network observer is further designed to achieve joint estimation and dynamic compensation of the system state. Let the observer joint state be ζ ​​= [x; ω]. According to formulas (9) and (10), the form of the designed RBF neural network observer is:

[0114]

[0115] In the above formula, in, represents the estimated values ​​of λ and ξ(t); L f represents the observer gain; represents the observation error; σ{·}=diag{·}; the inputs of the neural network are m represents the adjustable parameter of the observer.

[0116] In order to ensure the stable approximation performance of the network weights under uncertain perturbations, the update rate of the network weights is designed to be:

[0117]

[0118]

[0119] In the above formula, P, Q, and R are all positive definite diagonal matrices, and the observation error e ζ and the weight estimation error e of the neural network W1 , e W2 is bounded, where

[0120] S202: Verify the stability of the radial basis function neural network observer;

[0121] To verify the stability of the observer, the Lyapunov function is constructed as follows:

[0122]

[0123] From formula (16), we know that F3(ζ) satisfies the Lipschitz condition and there exists K1>0, such that:

[0124]

[0125] Then the Lyapunov derivative satisfies:

[0126]

[0127] Substituting the weight adaptive law, i.e., formula (17)-(18), we get:

[0128]

[0129] in,

[0130]

[0131] In the above formula, is the relaxed cross term.

[0132] For b∈R, c∈R, p>0, q>0, d>0, the following inequality holds:

[0133]

[0134] Let b = 1, have

[0135]

[0136] In the above formula, Substitute formula (23) (25) into formula (22), and let Then there is

[0137]

[0138] In the above formula,

[0139] in Represent the minimum eigenvalue of matrices Q and R respectively.

[0140] Formula (26) satisfies the finite time boundedness condition, then the system observation error e ζ and the neural network weight error e W1 、e W2 It will converge to the neighborhood of the origin in a finite time, and its upper bound of convergence time is Where γ>0; θ∈(0,1); x0 represents the starting point of the system, meaning that convergence is related to the starting position. This completes the proof of observer stability and finite-time convergence.

[0141] S3: Design a preset performance controller as the upper controller;

[0142] S301: defining a tracking error and defining a preset performance function to set an error boundary;

[0143] To improve the tracking performance and stability of the vehicle under disturbances and faults, a preset performance function is first used to dynamically convert the error into an adjustable constraint. Then, a barrier Lyapunov function is introduced to enhance the regulation to suppress the disturbance when the error approaches the boundary. Finally, the Nussbaum function is used to compensate for the control gain change, achieving dynamic correction and system stability under fault conditions.

[0144] In order to set the error bound, the tracking error is defined according to the vehicle model, that is, formula (9): The error dynamics equation is obtained

[0145]

[0146] In the above formula, e=[e1,e2,e3] T , in Represents the system state x=[v lon ,v lat ,ω veh ] T reference value.

[0147] Set the error bounds by defining a preset performance function:

[0148]

[0149] In the above formula, p0, p ∞denote constants defining the boundary of the performance function, a denotes a decay rate used to adjust the performance function; T0>0 is a parameter to be designed.

[0150] Considering the error dynamics equation, i.e., equation (27), in order to make the path tracking error e i (t) satisfy the constraint of the preset performance function, the controller is designed to satisfy the following condition:

[0151]

[0152] Define z i = e i / p i , i = 1, 2, 3, wherein e i denotes the i-th error, p i denotes the boundary of the i-th error. Then for all t ≥ 0, if there is a constraint |z i | < 1, the predetermined error boundary of the tracking error can be maintained. Meanwhile, the derivative of z i can be expressed as

[0153]

[0154] To solve the control problem of z i after transformation, define the actual control input as U = ΓU d , wherein U d is the expected control torque, and Γ is a torque loss factor caused by motor failure. According to equation (27), the system state equation can be obtained as follows:

[0155]

[0156] In the above equation, z = [z1, z2, z3] T ,

[0157] S302: Construct a barrier Lyapunov function;

[0158] In order to ensure that z is always within the constraint |z i (t)| < 1, i = 1, 2, 3, a barrier Lyapunov function is constructed as follows

[0159]

[0160] In the above equation, k0 is a positive number. Its derivative is

[0161]

[0162] In the above equation,

[0163] S303: Introducing Nussbaum function to deal with the unknown direction of control gain;

[0164] Due to the uncertainty of the drive gain caused by motor failure in the lateral dynamics of the system, and the possible change of the gain sign, conventional control is difficult to ensure stability. To deal with the problem of unknown gain direction, the Nussbaum function H(χ) is introduced, which must satisfy the following characteristics

[0165]

[0166]

[0167] These characteristics make the Nussbaum function widely used in dealing with nonlinear systems with unknown control direction or uncertain sign. A new type of H(χ) is designed as:

[0168]

[0169] S304: Designing a preset performance controller based on the barrier Lyapunov function;

[0170] Based on this, the control law can be designed as:

[0171]

[0172] In the above formula, κ = diag{κ1,κ2,κ3} is the positive gain matrix; e = [e1,e2,e3] T , p=[p1,p2,p3] T ; and H(χ)=diag{H(χ1),H(χ2),H(χ3)} is a Nussbaum-type function designed by formula (36); χ=[χ1,χ2,χ3] T is generated by the following adaptive law:

[0173]

[0174] S304: Perform stability analysis;

[0175] Substituting formula (37) into formula (33), the derivative of the Lyapunov function is:

[0176]

[0177] in,

[0178] z T B1ΓU d =z T B1U s -z T (-B1ΓH(χ)+B1)Us =z T B1U s +Ω(χ i ) (40)

[0179] In the above formula,

[0180] Substituting formulas (37) and (40) into formula (39), we obtain:

[0181]

[0182] In the above formula, κ min =min{κ1,κ2,κ3}.

[0183] To further simplify, according to the properties of the barrier Lyapunov function, the following inequality holds:

[0184]

[0185] Simplifying formula (41) we can get:

[0186]

[0187] In the above formula, ψ=2κ min .

[0188] Let function V:[0,∞)→R and t0∈(0,t), in the case where And under the premise of α>0, we can get:

[0189]

[0190] Consider the Lyapunov function satisfying And α>0, according to the integral inequality (44), we can get:

[0191]

[0192] To further analyze the boundedness of V2, we need to estimate the integral with Nussbaum-type gain terms. Consider V(t) and χ i (t)(i=1,2,…,N) is defined in the interval [0,t f ] and satisfies V(t)≥0 and χ i (0) = 0. In addition, assume that H(χ) is a Nussbaum type function. If the following inequality holds:

[0193]

[0194] Among them, c0 is a bounded constant, c1>0, gi (τ)∈[g - ,g + ] are bounded functions of the same sign. Then, V(t), χ i (t), In [0,t f ] is bounded.

[0195] The above analysis shows that V2, χ i (t), In [0,t f ] is bounded. When z i (0)∈Z, z i In [0,t f ] still remains in Z, that is, the error satisfies the preset performance function. Let v be The upper bound of , then when When V2 reaches stability, consider

[0196] V2≤V2(0)e -αt +ν (47)

[0197] make have:

[0198]

[0199] get That is, the system state always remains within the domain of the barrier function. Combining the exponential decay characteristics of the system's Lyapunov function V2 and the boundedness of the disturbance term, and based on the theory of semiglobal finite-time stability, it can be deduced that the system's state will converge to the safe set defined by the barrier Lyapunov function within a finite time. In summary, the proposed control strategy ensures good boundedness and fault-tolerant tracking performance in the presence of external disturbances and control gain uncertainty.

[0200] S4: Establish the lower-level controller to realize torque distribution;

[0201] S401: Establishing vehicle tire adhesion constraints and motor output constraints;

[0202] Considering the tire adhesion ellipse constraint, we can get

[0203]

[0204] In the above formula, F xi Indicates the longitudinal force of each tire; F yi Indicates the lateral force of each tire; F zi represents the vertical force of each tire; μ represents the road adhesion coefficient. According to formulas (5) and (6), considering the impact of drive motor failure on the vehicle longitudinal force, it can be expressed as:

[0205] λ i u imin / R f ≤F xi ≤λ i u imax / R f ,i=1,2,3,4 (50)

[0206] S402:design optimization objective function, torque distribution;

[0207] In order to ensure that the vehicle has good robustness and tracking performance when the motor fails, the upper controller output U d is distributed to each drive motor (drive motor of the left front wheel, right front wheel, left rear wheel, right rear wheel) and steering motor, considering the torque distribution model:

[0208] U d =B2(δ)Ku (51)

[0209] Wherein, U d =[F X ,F Y ,M Z ] T =[U1,U2,U3] T , Denotes the horizontal, vertical force and yaw moment; Control variable u=[u1,u2,u3,u4,u5] T is the driving torque of the four wheels (i.e. motor torque) and the front wheel steering angle.

[0210] In order to realize the optimal distribution of four-wheel torque and front wheel steering angle, considering the constraint of ground adhesion condition, the objective function is designed as:

[0211] J=β1||B2(δ)Ku-U d ||2+β2||u||2 (52)

[0212] In the formula: β1 is the weight coefficient matrix of longitudinal force, lateral force and yaw moment, and β2 is the weight coefficient matrix of four-wheel torque and front wheel steering angle.

[0213] Then, the torque distribution model of the drive motor and the steering motor can be expressed as

[0214]

[0215] For the optimization problem described in formula (53), the invention uses quadratic programming to solve the optimal control amount of the drive motor and the steering motor as the output of the lower controller.

[0216] To verify the fault-tolerant control effect of the application, a joint simulation platform is established based on CarSim / Simulink, and the vehicle parameters used in the simulation are shown in Table 1. Among them, the controller parameters are set as follows: L f = diag{65, 50, 20, 50, 50, 50, 50}, Q = 1.3, R = 4, P = diag{100, 80, 10}, k1 = 0.3, k2 = 0.1, κ = 50; in the preset performance function, p0 = 0.1, p ∞ = 0.01, T0 = 0.5, a = 3; the neural network parameters are: b1 = 2, b2 = 4,

[0217] Table 1 Vehicle parameters

[0218] parameter symbol Numerical Vehicle quality M 1416kg Vehicle wheelbase <![CDATA[2l s ]]> 1.739m Distance from center of mass to front axle <![CDATA[l f ]]> 1.016m Distance from center of mass to rear axle <![CDATA[l r ]]> 1.562m Air resistance coefficient <![CDATA[C a ]]> 0.5 Vehicle moment of inertia <![CDATA[I Z ]]> <![CDATA[1523kg·m 2 ]]> Tire rolling radius <![CDATA[R f ]]> 0.311m Front wheel cornering stiffness <![CDATA[C f ]]> 25796N / rad Rear wheel cornering stiffness <![CDATA[C r ]]> 25796N / rad

[0219] To verify the effectiveness of the fault-tolerant control method proposed in this paper, a controller without fault tolerance and a fast terminal sliding mode controller are selected for comparison, wherein the fast terminal sliding mode controller adopts the method in the prior art (GUO B, CHEN Y. Robust Adaptive Fault-Tolerant Control of Four-Wheel Independently Actuated Electric Vehicles [J]. IEEE Transactions on Industrial Informatics, 2020, 16(5): 2882-2894.).

[0220] The target vehicle performs single lane tracking with an initial longitudinal speed of 40 km / h and a reference acceleration At t = 6s, the left front wheel is partially failed λ1 = 0.6; at t = 10s, the left front wheel is completely failed λ1 = 0. The simulation results are shown in Figure 3 Figure 3 (a)-(c) compare the tracking performance of different control strategies. The controller without fault tolerance causes the trajectory to deviate continuously due to the inability to compensate for faults; the fast terminal sliding mode control can asymptotically converge, but there is a transient error; while the method proposed in this application still maintains high-precision tracking after the fault occurs, and its error curve Figure 3 (d)-(f) show that the convergence characteristics of the method of the application are superior to the fast terminal sliding mode control, and the maximum longitudinal speed error, lateral speed error, and yaw rate error caused by the fault are reduced by 45.3%, 22.5%, and 45.7%, respectively. Figure 4 Further verify that the neural network observer can accurately estimate unknown disturbances and faults in real time, and provide reliable input for the controller to compensate; Figure 5 ​The trajectory tracking results are presented. Experimental results show that the fault-tolerant control method of this application can still maintain the vehicle's motion performance under the coupled fault and disturbance scenario, verifying its robustness and safety.

[0221] Example 2:

[0222] The present invention provides a distributed drive vehicle preset performance fault-tolerant control system based on an adaptive neural network observer, comprising: a vehicle information acquisition module, a calculation module and an execution module; the vehicle information acquisition module includes a vehicle state sensor, GPS, inertial navigation, etc., which is used to collect vehicle motion parameters and position information in real time, and transmit the vehicle motion parameters and position information to the calculation module; the calculation module includes a high-performance processor, which is used to run the neural network observer, the preset performance controller and the torque distribution algorithm according to the vehicle motion parameters and position information, and transmit the obtained torque distribution results to the execution module; the execution module includes four independently driven wheel hub motors and an electric steering system, which is used to realize torque and steering control of the vehicle according to the torque distribution results of the calculation module.

[0223] The system realizes dynamic monitoring of vehicle status, fault diagnosis and fault-tolerant control, ensuring the stability and safety of the system under motor failure or external disturbance.

[0224] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A method for fault-tolerant control of preset performance of distributed drive vehicles based on an adaptive neural network observer, characterized in that: include: S1: Establish a vehicle dynamics model including faults and disturbances; S2: Design an adaptive neural network observer; Specifically include: S201: Real-time estimation of faults and disturbances using a radial basis function neural network observer; S202: Verify the stability of the radial basis function neural network observer; S3: Design a preset performance controller as the upper-level controller; specifically, it includes: S301: defining a tracking error and defining a preset performance function to set an error boundary; S302: Constructing barrier Lyapunov functions; S303: Introducing Nussbaum function to handle the unknown direction of control gain; S304: Designing a preset performance controller based on the barrier Lyapunov function; S305: Perform stability analysis; S4: Establish a lower-level controller to achieve torque distribution.

2. The method according to claim 1, characterized in that S1 specifically includes: S101: Establish the three-degree-of-freedom vehicle dynamics equation; Ignoring pitch and roll motions and considering only the front wheel steering, the three-degree-of-freedom dynamic equations of the vehicle are constructed as follows: Among them are: F X =(F xfl +F xfr )cosδ+F xrl +F xrr -(F yfl +F yfr )sinδ; F Y =(F xfl +F xfr )sinδ+(F yfl +F yfr )cosδ+F yrl +F yrr ; In the above formula, v lon 、v lat 、ω veh Respectively represent the longitudinal velocity, lateral velocity, and yaw rate of the vehicle; Respectively represent the longitudinal acceleration, lateral acceleration, and yaw angular acceleration of the vehicle; F X 、F Y 、M Z They represent the longitudinal force, lateral force, and yaw moment of the vehicle respectively; F xfl 、F xfr 、F xrl 、F xrr Respectively represent the longitudinal forces acting on the left front wheel, right front wheel, left rear wheel, and right rear wheel; l f 、l r Respectively represent the distance from the front and rear axles to the center of mass; l s Indicates half of the wheelbase; I Z and M represent the vehicle’s moment of inertia and vehicle mass respectively; δ represents the front wheel turning angle; C a represents the aerodynamic drag coefficient; d1(t), d2(t), and d3(t) are all external disturbances that can represent load changes, parameter disturbances, and unmodeled items in the vehicle system; it is assumed that the external disturbances d1(t), d2(t), and d3(t) are bounded; S102: Establishing a tire model of the vehicle; Assuming that the side slip angles of the left and right tires on the same axis are equal, the tire model of the vehicle is: F y =F yc +F y (δ) In the above formula, F y represents the wheel lateral force matrix, F y =[F yfl ,F yfr ,F yrl ,F xrr ] T ; F y (δ)=[C f δ,C f δ,0,0] T ; C f 、C r Represents the cornering stiffness of the front and rear tires respectively; α f , α r Represents the side slip angles of the front and rear tires respectively; S103: Establishing a wheel dynamics model of the vehicle; The wheel dynamics model is described as: In the above formula, the subscripts i=1, 2, 3, and 4 represent the left front wheel, right front wheel, left rear wheel, and right rear wheel respectively; T i Represents the torque of the i-th motor; R f Indicates the rolling radius of the wheel; F xi Represents the longitudinal forces of the four wheels respectively; I W represents the wheel moment of inertia; ω i represents the angular velocity of wheel i; S104: Establishing a vehicle fault model; The vehicle fault model is as follows: you i =λ i you di +Du i i∈{1,2,3,4,5} In the above formula, the subscripts i=1, 2, 3, and 4 represent the left front wheel, right front wheel, left rear wheel, and right rear wheel respectively; u1, u2, u3, and u4 represent the motor torque control signals of the left front wheel, right front wheel, left rear wheel, and right rear wheel respectively, which are the actual outputs of the i-th motor; u5 represents the front wheel angle; u di , i=1,2,3,4 represents the expected output of the i-th motor; u d5 represents the expected value of the front wheel turning angle; Δu i represents the disturbance caused by the fault; λ i represents the failure loss factor, and λ i ∈[0,1]; the motor output has upper and lower bounds, i.e. u i ∈[u imin ,u imax ],u imin is the minimum output value of motor i, u imax is the maximum value of motor i output; The above fault model can describe the following motor failure modes: (1) Multiplicative fault - partial failure mode: indicates abnormal response to the control signal, that is, 0 < λ i <1,Δu i =0; (2) Multiplicative fault - complete failure mode: Indicates that the motor fails completely and no longer outputs torque, i.e., λ i =0,Δu i =0; (3) Additive fault mode: It means that the performance of the motor deviates from the normal working state during operation, that is, λ i =0,Δu i ≠0; The control signal of the motor torque when there is no fault is the expected output of the motor, λ i =1,Δu i =0; S105: Obtaining a vehicle dynamics model including faults and disturbances; Based on the fault model, the unknown disturbance ξ1(t) is introduced and defined as follows: x=[v lon ,v lat ,ω veh ] T ,F1(x)=[f1(x),f2(x),f3(x)] T , The three-degree-of-freedom vehicle dynamics model including faults and unknown disturbances is obtained as follows: In the above formula, U = B2(δ)Ku, u = [u1, u2, u3, u4, u5] T , d(t) = [d1(t), d2(t), d3(t)] T , 3. The method according to claim 2, characterized in that S201 specifically includes: Designing a radial basis function neural network observer to estimate unknown faults and disturbances Its expression is: In the above formula, X1 and X2 represent the input of the neural network; represents the neural network weight matrix; h(x) represents the hidden layer neuron output; considering the modeling error of the RBF neural network, the actual unknown fault g(t) and disturbance ξ(t) can be modeled as: In the above formula, ε1 and ε2 represent the estimation errors of unknown faults and disturbances respectively; let the joint state of the observer be ζ ​​= [x; ω], and the form of the designed RBF neural network observer is: In the above formula, in, Represents the estimated value of λ, ξ(t); λ=diag{λ1, λ2, λ3, λ4, λ5}; L f represents the observer gain; represents the observation error; σ{·}=diag{·}; the inputs of the neural network are ω=[ω1,ω2,ω3,ω4] T =[ω fl ,ω fr ,ω rl ,ω rr ] T ; m represents the adjustable parameter of the observer; The update rate of the designed network weights is: In the above formula, P, Q, and R are all positive definite diagonal matrices, and the observation error e ζ and the weight estimation error e of the neural network W1 , e W2 is bounded, where 4. The method according to claim 3, characterized in that S301 specifically includes: Defining Tracking Error The error dynamics equation is as follows: In the above formula, e=[e1,e2,e3] T , in Represents the system state x=[v lon ,v lat ,ω veh ] T Reference value of The error bounds are set by defining a preset performance function, which is as follows: In the above formula, p0, p ∞ represents the constant defining the boundary of the performance function, a represents the decay rate used to adjust the performance function; T0>0 is the parameter to be designed; In order to make the path tracking error e i (t) To meet the constraints of the preset performance function, the designed controller meets the following conditions: Define z i =e i / p i ,i=1,2,3, where e i represents the i-th error, p i represents the bound of the i-th error, then for all t≥0, if there is a constraint |z i |<1, the predetermined error limit of the tracking error can be maintained; Define the actual control input as U = ΓU d , where U d is the desired control torque, Γ is the torque loss factor caused by motor failure; the system state equation is as follows: In the above formula, z=[z1,z2,z3] T , 5. The method according to claim 4, characterized in that S302 specifically includes: To ensure that z is always within the constraint |z i (t)|<1, i=1,2,3, construct the following obstacle Lyapunov function: In the above formula, k0 is a positive number; the derivative of V2 is In the above formula, 6. The method according to claim 5, characterized in that The Nussbaum function H(χ) in S303 is:

7. The method according to claim 6, characterized in that S304 specifically includes: The control law of the preset performance controller is designed as In the above formula, κ = diag{κ1,κ2,κ3} is the positive gain matrix; e = [e1,e2,e3] T , p=[p1,p2,p3] T ; and H(χ)=diag{H(χ1),H(χ2),H(χ3)}; χ=[χ1,χ2,χ3] T is generated by the following adaptive law:

8. The method according to claim 7, characterized in that S4 specifically includes: S401: Establishing vehicle tire adhesion constraints and motor output constraints; The tire attachment ellipse constraints are as follows: In the above formula, F xi Indicates the longitudinal force of each tire; F yi Indicates the lateral force of each tire; F zi represents the vertical force of each tire; μ represents the road adhesion coefficient; The motor output constraint is as follows: i u imin / R f ≤F xi ≤λ i u imax / R f ,i=1,2,3,4; S402: Designing an optimization objective function and performing torque distribution; The design objective function is: J = β1||B2(δ)Ku-U d ||2+β2||u||2; Where: β1 is the weight coefficient matrix of longitudinal force, lateral force and yaw moment; β2 is the weight coefficient matrix of four-wheel torque and front wheel angle; U d =[F X ,F Y ,M Z ] T =[U1,U2,U3] T , represents the lateral and longitudinal resultant forces and yaw moment; control variable u=[u1,u2,u3,u4,u5] T The driving torque of the four wheels and the front wheel angle; The torque distribution model of the drive motor and the steering motor is expressed as 9. The method according to claim 8, characterized in that The torque distribution model of the drive motor and the steering motor is solved by quadratic programming to obtain the optimal control variables of the drive motor and the steering motor as the output of the lower controller.

10. A distributed drive vehicle preset performance fault-tolerant control system based on an adaptive neural network observer, characterized in that: When executing the method according to any one of claims 1 to 9, the system comprises: a vehicle information acquisition module, a calculation module, and an execution module; The vehicle information acquisition module includes a vehicle status sensor, GPS, and inertial navigation, which is used to collect vehicle motion parameters and position information in real time and transmit the vehicle motion parameters and position information to the calculation module; The computing module includes a high-performance processor for running a neural network observer, a preset performance controller, and a torque distribution algorithm based on the vehicle motion parameters and position information, and transmitting the obtained torque distribution results to the execution module; The execution module includes four independently driven wheel hub motors and an electric steering system, which is used to achieve torque and steering control of the vehicle based on the torque distribution results of the calculation module.

Images