Intelligent fault-tolerant control method for electro-hydraulic active suspension considering hydraulic cylinder internal leakage fault

Abstract

The application discloses an intelligent fault-tolerant control method for electro-hydraulic active suspension considering internal leakage fault of a hydraulic cylinder, and comprises the following steps: (1) for the vehicle electro-hydraulic active suspension, considering parameter uncertainty, nonlinear friction and internal leakage of the hydraulic cylinder, a nonlinear dynamic state space model of the electro-hydraulic active suspension is established; (2) a RBF neural network observer is constructed for a generalized fault function, and the omnipotent approximation characteristic thereof is used to reconfigure the fault dynamic on line; a weight value self-adaptive updating law is designed; (3) a preset performance control strategy is introduced, a preset performance function is defined, and an error dynamic envelope line is set; (4) the system is decomposed based on a backstepping framework, and a virtual control law is designed respectively; a dynamic surface controller is introduced, and a first-order low-pass filter is used to obtain the differential of the virtual control quantity; (5) the fault compensation signal of the observer is combined with the output of the dynamic surface controller to synthesize a final voltage control law. The application solves the problems of nonlinear leakage model uncertainty, parameter time-varying and external disturbance existing in the electro-hydraulic suspension system.

Description

Technical Field

[0001] This invention relates to the fields of vehicle engineering and advanced control technology, and in particular to an intelligent fault-tolerant control method for electro-hydraulic active suspension that considers hydraulic cylinder leakage faults. Specifically, it is an adaptive fault-tolerant control method for electro-hydraulic active suspension that combines radial basis function neural networks, preset performance control, and dynamic surface control technology. Background Technology

[0002] With the rapid development of the automotive industry, people have placed increasingly higher demands on vehicle ride comfort and handling stability. Electro-hydraulic active suspension systems, with their advantages of high power density, fast response speed, and large output force, can actively adjust suspension output in real time according to road excitation and driving conditions, and have become a key actuator for improving vehicle chassis performance. However, electro-hydraulic active suspension systems operate under harsh environments of reciprocating high-frequency motion and high temperature and pressure, making the hydraulic actuators (hydraulic cylinders) highly susceptible to internal leakage failures. Internal leakage is usually caused by seal wear, aging, or oil contamination, and its leakage characteristics often include a mixed mode of laminar and turbulent flow, exhibiting strong nonlinearity and uncertainty. Once internal leakage occurs, it leads to a decrease in the output force of the hydraulic cylinder, causing the actual control force to deviate from the expected value, thereby causing a deterioration in vehicle posture control performance, and in severe cases, even leading to suspension system instability, endangering driving safety.

[0003] To address the actuator failure problem in suspension systems, the industry has conducted relevant research and proposed some fault-tolerant control strategies. For example, Chinese patent application CN202210802834.9 discloses a "vehicle seat active suspension system and control method with actuator failure," which models and compensates for potential failures such as performance loss in suspension actuators, thus improving the system's reliability to some extent. However, when applying existing fault-tolerant control methods (including the aforementioned patented technology) to actual high-order, highly nonlinear vehicle electro-hydraulic active suspension systems, the following significant limitations still exist:

[0004] First, it is difficult to accurately describe and compensate for generalized, strongly nonlinear faults. Existing actuator fault models are often oversimplified, typically assuming that the fault conforms to a linear parameterized model or a simple proportional decay. However, the internal leakage of actual electro-hydraulic suspensions not only includes nonlinear mixed leakage that varies with differential pressure, but also unmodeled Coulomb friction and viscous friction. When the actual fault dynamics deviate from the preset simplified mathematical model, the compensation accuracy and robustness of traditional adaptive control will be greatly reduced, easily leading to control failure or even system instability.

[0005] Second, the inability to quantify and constrain transient overshoot and convergence speed can easily lead to mechanical limit block impacts and structural damage. Due to the limitations of the vehicle's mechanical structure, the suspension has strict physical constraints on its dynamic travel. Existing fault-tolerant control strategies mostly focus on ensuring the system's eventual steady-state convergence, failing to quantify and strictly constrain the transient dynamic boundaries of suspension displacement during the adjustment process from a mathematical perspective. Under sudden faults or severe road impacts, excessive transient overshoot can easily cause the suspension to impact the mechanical limit block, resulting in structural damage and deterioration of ride comfort.

[0006] Third, the high computational complexity and amplification of noise make it difficult to meet the real-time requirements of automotive ECUs. While traditional backstepping methods are often used when dealing with high-order, strictly feedback nonlinear systems like electro-hydraulic active suspensions, they require repeated analytical differentiation of the virtual control law, which contains complex nonlinear terms. As the system order increases, this leads to a severe "differential term inflation" problem, amplifying high-frequency measurement noise and causing an exponential increase in computational complexity, making it difficult to achieve millisecond-level real-time calculations in automotive electronic control units (ECUs) with limited computing power. Therefore, there is an urgent need for an intelligent fault-tolerant control strategy that does not require precise mathematical models, can accommodate transient safety physical constraints, and is computationally efficient, in order to fundamentally solve the challenge of high-precision and high-safety control of electro-hydraulic active suspensions under complex leakage fault conditions. Summary of the Invention

[0007] This invention provides an intelligent fault-tolerant control method for electro-hydraulic active suspension that considers hydraulic cylinder leakage faults. It aims to address the complex hydraulic cylinder leakage and friction characteristics of electro-hydraulic active suspensions, which traditional methods struggle to accurately describe and compensate for generalized strong nonlinear faults, easily leading to control failure or even system instability. Due to the physical travel limitations of the suspension, existing methods cannot quantify and constrain transient overshoot and convergence speed, easily causing mechanical limit impacts and damaging the structure. For high-order systems, the traditional backstepping method repeatedly calculates derivatives, causing differential term expansion, resulting in high computational complexity and amplified noise, making it difficult to meet the real-time requirements of onboard ECUs.

[0008] To address the aforementioned technical problems, this invention proposes an intelligent fault-tolerant control method for electro-hydraulic active suspension that considers internal leakage faults in hydraulic cylinders, comprising the following steps:

[0009] (1) For the vehicle electro-hydraulic active suspension, considering parameter uncertainty, nonlinear friction and leakage in the hydraulic cylinder, a nonlinear dynamic state space model of the electro-hydraulic active suspension is established; all leakage and unmodeled dynamics are uniformly represented as generalized unknown nonlinear fault functions, and road excitation is regarded as external disturbance.

[0010] (2) Construct an RBF neural network observer for the generalized fault function and use its universal approximation characteristics to reconstruct the fault dynamics online; design a weight adaptive update law based on Lyapunov theory to ensure that the weights are bounded and output a fault compensation signal to offset the pressure loss caused by internal leakage and friction.

[0011] (3) Introduce a preset performance control strategy, define a preset performance function to set the dynamic envelope of the error; use error transformation and logarithmic barrier Lyapunov function to ensure that the system error is strictly within the preset range in both transient and steady state stages, and prevent violation of physical constraints.

[0012] (4) Based on the backstepping framework, the system is decomposed and virtual control laws are designed respectively; a dynamic surface controller is introduced, and a first-order low-pass filter is used to obtain the differential of the virtual control quantity, avoiding the "differential explosion" problem caused by repeated differentiation in the traditional backstepping method and reducing the computational complexity.

[0013] (5) Combine the fault compensation signal of the observer with the output of the dynamic surface controller to synthesize the final voltage control law; control the pressure difference by adjusting the opening of the servo valve to actively compensate for the force attenuation caused by internal leakage, realize fault-tolerant control under fault conditions, and ensure that the suspension performance meets the preset indicators.

[0014] This invention integrates deep learning approximation capabilities, pre-defined performance theories, and dynamic surface control technology to achieve intelligent fault-tolerant control of electro-hydraulic active suspension. It addresses the complex hydraulic cylinder leakage and friction characteristics of electro-hydraulic active suspension, which traditional methods struggle to accurately describe and compensate for generalized strong nonlinear faults, easily leading to control failure or even system instability. It also solves the problem that, limited by the suspension's physical travel, existing methods cannot quantify and constrain transient overshoot and convergence speed, easily causing mechanical limit impacts and structural damage. Furthermore, for high-order systems, traditional backstepping methods repeatedly derive, causing differential term expansion, high computational complexity, and amplified noise, making it difficult to meet the real-time requirements of onboard ECUs.

[0015] As a further improvement to this technical document:

[0016] The nonlinear dynamic state-space model of the electro-hydraulic active suspension in step (1) is specifically represented as follows:

[0017] (1)

[0018] in, Let be the system state vector. Represents the vertical displacement of the vehicle body. Represents the vertical speed of the vehicle body. Represents the load pressure of the hydraulic cylinder; These represent the vehicle's vertical velocity, vertical acceleration, and the rate of change of hydraulic cylinder load pressure, respectively. For the sprung mass; and These are the suspension stiffness and damping coefficient, respectively. Input is the vertical displacement of the road surface. Input the vertical velocity of the road surface; The effective area of ​​the hydraulic cylinder piston; The effective bulk modulus of hydraulic oil; This refers to the total volume of the hydraulic cylinder; The total leakage coefficient; For servo valve flow gain; For control voltage input; Let be a generalized unknown nonlinear fault function that includes nonlinear internal leakage and friction, and assume that ,in It is the physical upper limit of the degree of leakage in the hydraulic cylinder, which cannot be measured by conventional means but can be approximated by algorithms.

[0019] In state-space modeling, state vectors are selected. These represent the vehicle's vertical displacement, vertical velocity, and hydraulic cylinder load pressure, respectively; the nonlinear dynamic state-space model of the electro-hydraulic active suspension describes the servo valve output force and spring force. and damping force Changes in vehicle body acceleration under combined effects; Hydraulic pressure-flow equation: describes the servo valve control voltage. Controlled flow gain Pressure dynamics caused by changes in hydraulic cylinder volume and leakage.

[0020] In particular, this invention abandons the traditional linear leakage assumption and considers laminar and turbulent leakage within the hydraulic cylinder (which varies with pressure difference) as differentiating factors. Variation), unmodeled Coulomb friction and viscous friction, and parameter perturbations (such as oil elastic modulus) The time-varying nature of the fault is uniformly packaged and represented as a generalized unknown nonlinear fault function. This function, as the total disturbance term in the pressure subsystem, no longer depends on a specific analytical expression, laying the foundation for subsequent "model-free" intelligent compensation.

[0021] As a further improvement to this technical document:

[0022] The RBF neural network observer and the adaptive weight update law in step (2) specifically include:

[0023] (201) Establish a neural network approximation model:

[0024] Generalized unknown nonlinear fault function Approaching as:

[0025] (2)

[0026] in, For the ideal weight vector, This represents the Gaussian radial basis function vector of the hidden layer. For network approximation error and .

[0027] Fault Observer Output Represented as:

[0028] (3)

[0029] in, This represents the weight estimation vector for the neural network. This represents the number of hidden layer nodes. Let be a Gaussian radial basis function vector, the first... basis functions Defined as:

[0030] (4)

[0031] in, For the first The center vector of each node This is the width parameter of the node. The Euclidean norm is represented by exp; the natural exponential function is represented by exp; utilizing the universal approximation property of the RBF neural network, the system state vector is used as the basis for the approximation. As the input layer, the Gaussian function of the hidden layer Perform nonlinear mapping; Each basis function is in Gaussian form. This allows for the capture of system dynamics within a local space.

[0032] (202) Online adaptive update law for weights:

[0033] The design incorporates an online adaptive update law for weights to eliminate tracking errors in the pressure loop. The specific form is as follows:

[0034] (5)

[0035] in, The learning rate gain matrix is ​​a positive definite symmetric matrix, which determines the convergence speed of the weights; This represents the Gaussian radial basis function vector of the hidden layer. The dynamic surface error variable of the pressure ring; To robustly adjust parameters, preventing weight drift and ensuring an online adaptive update law for weights. Bounded; the update law contains a key... - Correction items ( The function of modification is to ensure that the weights converge to the optimal value while preventing the weights from drifting to infinity due to approximation errors or external disturbances, thus ensuring the boundedness and safety of the neural network in closed-loop control.

[0036] As a further improvement to this technical document:

[0037] The construction process of the preset performance function and the barrier Lyapunov function in step (3) is as follows:

[0038] (301) Define the preset performance function :

[0039] The exponential decay function is selected as the performance envelope:

[0040] (6)

[0041] in, As the initial performance boundary, it must satisfy ,in The displacement tracking error at the initial moment; This represents the maximum permissible error in steady state, and Convergence rate factor; system tracking error Must always satisfy ,in This is the overshoot adjustment parameter;

[0042] (302) Construct the error transformation function :

[0043] Introducing an error transformation function will limit the error. Mapping to unconstrained variables :

[0044] (7)

[0045] Its inverse transformation is:

[0046] (8)

[0047] in, : Represents the unconstrained error variable after transformation. Introducing this variable maps the error, which was originally limited to physical boundaries, to the entire real number domain. This transforms a constrained control problem into an unconstrained control problem. : Represents the actual limited tracking error of the system, that is, the deviation between the actual displacement of the vehicle body and the expected displacement; : Represents the preset performance function, which represents the boundary of the dynamic envelope that decays exponentially with time, i.e., the maximum positive boundary of the error tolerance; : Indicates the overshoot adjustment parameter, typically set to a value of It and Together, they determine the dynamic lower bound of the error, and the actual system error. Strictly restricted to Within the range; : represents the natural constant, with a value of 2.71828, and serves as the base of the exponential function; : Represents a time variable.

[0048] (303) Constructing a logarithmic barrier Lyapunov function :

[0049] For the displacement subsystem, a logarithmic barrier Lyapunov function is selected:

[0050] (9)

[0051] in, The transformed error variable The preset virtual safety boundary constant, the function property guarantees that when This results in an infinitely large penalty gain in the controller design to constrain the error boundary; the mathematical properties of this function guarantee that when the conversion error... Approaching the virtual boundary When, the function value The value tends towards infinity, thus forcing the controller to generate a huge penalty gain to "push" the error back to the safe region, theoretically ensuring that the system state will never violate the preset dynamic envelope at any time.

[0052] As a further improvement to this technical document:

[0053] The virtual control law in step (4) includes the virtual control law for the displacement loop. And speed loop virtual control law The specific design process is as follows:

[0054] (401) Displacement loop virtual control law design:

[0055] Define displacement tracking error ,in The desired displacement; based on the logarithmic barrier Lyapunov function in step (3). Design of virtual control law for displacement :

[0056] (10)

[0057] in, : Represents the virtual control law of the displacement loop; : indicates the displacement loop control gain, and This parameter is used to adjust the convergence speed of displacement tracking error; the larger the value, the faster the error is eliminated. : Indicates intermediate substitution variables generated during the differentiation process; : Indicates the overshoot adjustment parameter; : Represents the actual displacement tracking error; where, The control gain for the displacement loop; in the design of the virtual control law, the virtual control law is designed. This control law is used to stabilize displacement errors. It includes a specific gain term resulting from a preset performance conversion to ensure that displacement tracking meets the envelope requirements. A virtual control law is designed. The desired hydraulic cylinder load pressure is used to track the desired speed. The displacement loop virtual control law is used to achieve error constraints. Based on the specific logarithmic barrier Lyapunov function and error transformation function constructed in the previous step, it is derived through mathematical differentiation and scaling. It contains specific nonlinear penalty terms to achieve suspension anti-collision.

[0058] (402) Velocity loop virtual control law design:

[0059] Define speed tracking error ,in for Output after low-pass filter; design of velocity loop virtual control law. , where is the desired hydraulic cylinder pressure:

[0060] (11)

[0061] in, : Represents the virtual control law of the speed loop; : Indicates the sprung mass of a vehicle; : Indicates the effective area of ​​the hydraulic cylinder piston. : indicates the speed loop control gain, and Used to adjust speed tracking error The convergence speed; : Indicates the speed tracking error between the actual vertical speed of the vehicle body and the desired speed; : Represents the desired velocity signal after filtering; : Represents the time constant of the first low-pass filter in dynamic surface control; : Indicates the suspension stiffness coefficient; : Indicates the suspension damping coefficient; : Indicates the actual vertical displacement of the vehicle body; : Indicates the actual vertical speed of the vehicle body; : Indicates the vertical displacement input of the road surface; : Indicates the vertical velocity input of the road surface; , which is the speed loop control gain.

[0062] (403) Introduction of the first-stage low-pass filter and signal processing

[0063] In the traditional backstepping method, in order to design the control law for the next step, it is necessary to calculate... .because It contains complex logarithmic functions and performance functions Directly taking its derivative will produce an extremely lengthy mathematical expression and amplify measurement noise.

[0064] Introducing the first first-order low-pass filter to obtain The filtered value and its derivative:

[0065] (12)

[0066] in This represents the filter's time constant. The filter output... As the "desired velocity after filtering", and its derivative It can be obtained directly through algebraic operations, completely avoiding the need for analytical differentiation of nonlinear functions.

[0067] (404) Introduction of the second-stage low-pass filter

[0068] Similarly, in order to obtain expected pressure Differential signal For subsequent pressure loop design, a second first-order low-pass filter is introduced:

[0069] (13)

[0070] Physical parameter tuning: This is the filter time constant. The pressure loop is the inner loop, and its response speed should be faster than the outer loop (velocity loop). (Setting) This ensures that the inner loop can respond quickly to the commands of the outer loop, following the cascade control design principle of the inner loop being fast and the outer loop being slow.

[0071] Output signal: This is the filtered desired load pressure, which is the final tracking target for pressure fault-tolerant control in subsequent step (5). The pressure tracking error is defined as... .

[0072] The speed loop virtual control law combines the dynamic surface controller introduced in this invention, uses the filtered signal instead of analytical differentiation, and incorporates the feedforward compensation term of the suspension physical model for derivation.

[0073] As a further improvement to this technical document:

[0074] The first-order low-pass filter in step (4) is as follows:

[0075] For virtual control laws and Two first-order low-pass filters are introduced:

[0076] (14)

[0077] in, These are the time constants of the filter; The filtered desired velocity signal, This is the filtered desired pressure signal;

[0078] Define the dynamic surface error variable as:

[0079] (15)

[0080] In the controller design, the introduction of an extremely complex logarithmic barrier Lyapunov function for collision prevention results in an exceptionally large virtual control law. Directly differentiating it using the traditional backstepping method would trigger a severe "differential explosion," causing the ECU's computing power to collapse. To avoid this collapse, the output of the filter is directly used... and Replace virtual control law and The analytical derivative is used to eliminate the computational bloat caused by higher-order nonlinear terms, thus enabling the overall fault-tolerant control architecture to be realized.

[0081] As a further improvement to this technical document:

[0082] The final voltage control law in step (5) The calculation expression is:

[0083] (16)

[0084] in: This is the control gain coefficient of the pressure loop (or force loop); To obtain the virtual control quantity differential signal using a dynamic surface controller; For the output of the RBF neural network targeted Fault compensation items; This is a robustness term used to overcome reconstruction errors in neural networks. To mitigate external residual interference and ensure the stability of the closed-loop system, its design is as follows:

[0085] (17)

[0086] (18)

[0087] in, : Indicates the final servo valve control voltage input; : Represents the control gain coefficient of the pressure loop, and is a positive number. It determines the convergence speed of the actual pressure tracking the desired pressure. : Indicates pressure tracking error; : This indicates the robust compensation control term. This term is introduced to overcome the inherent reconstruction approximation error of the neural network and the residual disturbances that are not measurable from the outside, and to prevent these errors from causing the system to diverge. : represents the robust gain constant, whose value must be greater than the upper bound of the neural network approximation error and residual disturbance to theoretically guarantee the Lyapunov stability of the closed-loop system; where, For robustness gain, The smoothing factor for the hyperbolic tangent function is... It is a symbolic function;

[0088] After passing through displacement loops, velocity loops, and finally pressure loops, the final voltage control law is derived using Lyapunov stability theory; in particular, the fault compensation term output by the RBF neural network fault observer is used. And the virtual control quantity differential signal obtained using the low-pass filter of the dynamic surface controller. A specific solution was proposed to address the physical fault of internal leakage in the suspension, ultimately achieving intelligent fault-tolerant control.

[0089] As a further improvement to this technical document:

[0090] The aforementioned intelligent fault-tolerant control method for electro-hydraulic active suspension considering leakage faults in hydraulic cylinders further includes step (6), which is a system stability analysis that requires constructing a Lyapunov candidate function for the entire system. :

[0091] (19)

[0092] in, For weight estimation error, For filter boundary layer error;

[0093] By adjusting the control gain Filter time constant and robust parameters , making ,in Since the value is a positive constant, it proves that all signals in the closed-loop system are uniformly eventually bounded (UUB), and the tracking error converges to a compact set near the origin.

[0094] Compared with the prior art, the present invention exhibits the following significant advantages in the field of electro-hydraulic active suspension control:

[0095] First, addressing the challenge of modeling the highly complex and difficult-to-model internal leakage and friction characteristics of hydraulic actuators, this invention abandons the traditional parameter adaptation method that relies on linearized models and innovatively introduces a radial basis function neural network observer. Utilizing its powerful universal approximation properties, this method can perceive and reconstruct the dynamics of generalized unknown faults, including laminar and turbulent mixed leakage and nonlinear friction, online in real time, achieving "model-free" intelligent compensation for arbitrary forms of nonlinear faults. This means that designers do not need to possess the precise mathematical structure or prior knowledge of the fault; they can achieve high-precision fault cancellation solely based on system input and output data, greatly enhancing the system's robustness under extreme operating conditions.

[0096] Secondly, this invention constructs a strict transient safety constraint mechanism by introducing a preset performance function and a logarithmic barrier Lyapunov function (BLF). This mechanism dynamically sets an insurmountable "soft boundary" for the suspension system's errors. Regardless of how severe the road excitation or how sudden the malfunction, the control algorithm can forcibly limit the vehicle's displacement and attitude errors within the preset dynamic envelope. This characteristic not only theoretically guarantees the steady-state accuracy of the closed-loop system but also effectively eliminates the risk of suspension mechanical limit impact caused by excessive overshoot at the physical level, significantly improving vehicle driving safety while ensuring ride comfort.

[0097] Furthermore, addressing the common problem of differential term expansion in backstepping design of high-order nonlinear systems, this invention creatively incorporates Dynamic Surface Control (DSC) technology. By introducing a first-order low-pass filter to obtain the differential signal of the virtual control quantity, it successfully avoids the repetitive analytical differentiation process involving neural network weights and complex nonlinear functions in traditional methods. This improvement reduces the computational complexity of the control algorithm by an order of magnitude, making it easily adaptable to vehicle electronic control units (ECUs) with limited computing power, achieving real-time operation with millisecond-level cycles. Simultaneously, it effectively filters out high-frequency measurement noise interference to the servo valve, extending the lifespan of the actuators, and achieving a perfect balance between the depth of control theory and the value of engineering applications.

[0098] In summary, the composite control architecture constructed in this invention utilizes model information (through feedback linearization), combines a data-driven approach (through neural networks), and introduces robust control terms. Simulation and theoretical analysis show that this method enables the closed-loop system signal to achieve uniform eventual boundedness (UUB). Even under severe internal leakage in the hydraulic cylinder (e.g., causing an output force reduction of more than 30%), this invention can still maintain vibration suppression effects comparable to those under healthy conditions, demonstrating excellent fault tolerance and adaptability to different operating conditions. Attached Figure Description

[0099] Figure 1 It is a flowchart of the overall control method.

[0100] Figure 2 This is a physical model of a quarter-vehicle electro-hydraulic active suspension (with key parameters labeled).

[0101] Figure 3 This is a block diagram of the RBF neural network observer structure.

[0102] Figure 4 This is a schematic diagram of the preset performance function envelope constraint (showing the error boundary).

[0103] Figure 5 It is a block diagram of the overall control system structure, including DSC filter and BLF. Detailed Implementation

[0104] The specific embodiments of the present invention will now be described with reference to the accompanying drawings to enable those skilled in the art to better understand the invention. It should be noted that in the description of the present invention, detailed descriptions of known functions and designs that might obscure the main content of the invention will be omitted here.

[0105] The intelligent fault-tolerant control method for electro-hydraulic active suspension considering internal leakage faults in hydraulic cylinders in this embodiment includes the following five main steps:

[0106] Step (1): For the vehicle's electro-hydraulic active suspension, considering parameter uncertainty, nonlinear friction and leakage in the hydraulic cylinder, establish a nonlinear dynamic state-space model of the electro-hydraulic active suspension; characterize all leakage and unmodeled dynamics as a generalized unknown nonlinear fault function, and regard road excitation as an external disturbance.

[0107] The nonlinear dynamic state-space model of the electro-hydraulic active suspension in step (1) is specifically represented as follows:

[0108] (1)

[0109] in, Let be the system state vector. Represents the vertical displacement of the vehicle body. Represents the vertical speed of the vehicle body. Represents the load pressure of the hydraulic cylinder; These represent the vehicle's vertical velocity, vertical acceleration, and the rate of change of hydraulic cylinder load pressure, respectively. For the sprung mass; and These are the suspension stiffness and damping coefficient, respectively. Input is the vertical displacement of the road surface. Input the vertical velocity of the road surface; The effective area of ​​the hydraulic cylinder piston; The effective bulk modulus of hydraulic oil; This refers to the total volume of the hydraulic cylinder; The total leakage coefficient; For servo valve flow gain; For control voltage input; Let be a generalized unknown nonlinear fault function that includes nonlinear internal leakage and friction, and assume that ,in It is the physical upper limit of the degree of leakage in the hydraulic cylinder, which cannot be measured by conventional means but can be approximated by algorithms.

[0110] In state-space modeling, state vectors are selected. These represent the vehicle's vertical displacement, vertical velocity, and hydraulic cylinder load pressure, respectively; the nonlinear dynamic state-space model of the electro-hydraulic active suspension describes the servo valve output force and spring force. and damping force Changes in vehicle body acceleration under combined effects; Hydraulic pressure-flow equation: describes the servo valve control voltage. Controlled flow gain Pressure dynamics caused by changes in hydraulic cylinder volume and leakage.

[0111] In particular, this invention abandons the traditional linear leakage assumption and considers laminar and turbulent leakage within the hydraulic cylinder (which varies with pressure difference) as differentiating factors. Variation), unmodeled Coulomb friction and viscous friction, and parameter perturbations (such as oil elastic modulus) The time-varying nature of the fault is uniformly packaged and represented as a generalized unknown nonlinear fault function. This function, as the total disturbance term in the pressure subsystem, no longer depends on a specific analytical expression, laying the foundation for subsequent "model-free" intelligent compensation.

[0112] To address the challenge of accurately describing internal leakage in hydraulic cylinders, this embodiment introduces a generalized unknown nonlinear fault function. This function is not a simple constant or linear term, but rather encompasses the following complex physical fault characteristics:

[0113] Nonlinear internal leakage: includes laminar leakage that varies linearly with pressure difference. and turbulent leakage as a function of the square root of the pressure difference When the seal is severely worn, the turbulent component increases significantly.

[0114] Nonlinear friction: Coulomb friction and viscous friction during the piston movement of a hydraulic cylinder.

[0115] Time-varying uncertainty of parameters: bulk modulus due to oil temperature rise Decrease, and leakage coefficient The drift.

[0116] therefore, Defined as the sum of all the unmodeled dynamics and faults mentioned above, satisfying ,in The upper bound is unknown. Subsequent steps in this invention will no longer rely on... Instead of providing a specific analytical expression, we approximate it online.

[0117] Step (2): Construct an RBF neural network observer for the generalized fault function and reconstruct the fault dynamics online using its universal approximation characteristics; design an adaptive update law for weights based on Lyapunov theory to ensure that the weights are bounded and output a fault compensation signal to offset the pressure loss caused by internal leakage and friction.

[0118] For the generalized faults defined in step (1) This embodiment utilizes the universal approximation property of radial basis function (RBF) neural networks for real-time estimation. The specific implementation process is as follows:

[0119] (2.1) RBF Neural Network Topology and Input Design

[0120] Generalized unknown nonlinear fault function Approaching as:

[0121] (2)

[0122] in, For the ideal weight vector, This represents the Gaussian radial basis function vector of the hidden layer. For network approximation error and ;

[0123] Fault Observer Output Represented as:

[0124] (3)

[0125] (4)

[0126] in, As the center vector, This is the node width parameter.

[0127] (2.2) Design and physical significance of the adaptive weight update law

[0128] In order to estimate the weights To achieve rapid convergence and prevent weight drift caused by external disturbances, a design with... - Adaptive update law for the correction term (same as Equation 3):

[0129] (5)

[0130] The physical meanings and parameter tuning strategies for each item are as follows:

[0131] Error driving term Based on the gradient descent principle, utilizing the pressure loop error... Adjust weights in real time. At that time, the algorithm automatically adjusts the weights to reduce pressure tracking deviation.

[0132] Robust damping term This is a key improvement of the present invention. When approximation errors exist... In such cases, traditional adaptive laws may cause weights to accumulate to infinity. Introducing... This is equivalent to adding a "soft damper," which forces the weights to adjust when the system reaches steady state. Keep within bounded areas.

[0133] Parameter tuning: In this embodiment, robust correction parameters are selected. Select the learning rate matrix This setting is designed to ensure that, after a fault occurs (such as a sudden breakage of the seal), the observer can complete fault reconstruction within 0.2 seconds.

[0134] (2.3) Observer convergence analysis

[0135] Define Lyapunov candidate functions ,in This represents the weight estimation error. By differentiating and substituting into the update law, and using Young's inequality for scaling, it can be proved that... It is uniformly eventually bounded (UUB). Regardless of the initial weights, the observer's output is... Ultimately, they will all converge to the actual fault. Within a very small neighborhood, it provides a precise fault-tolerant compensation signal for the subsequent step (5).

[0136] Step (3): Introduce a preset performance control strategy, define a preset performance function to set the error dynamic envelope; use error transformation and logarithmic barrier Lyapunov function to ensure that the system error is strictly within the preset range in both transient and steady state stages, and prevent violation of physical constraints.

[0137] (301) Define the preset performance function To quantify the dynamic constraint range of the suspension, we first define the displacement tracking error. In this embodiment, a performance function that converges over time is designed. As the dynamic boundary of the error, an exponentially decaying function is selected as the performance envelope, and its function expression is defined as:

[0138] (6)

[0139] in, As the initial performance boundary, it must satisfy ,in The displacement tracking error at the initial moment; This represents the maximum permissible error in steady state, and Convergence rate factor; system tracking error Must always satisfy ,in This is the overshoot adjustment parameter;

[0140] (302) Construct the error transformation function Because the original error e_1(t) is constrained by inequalities, directly designing a controller is quite difficult. This embodiment introduces an error transformation technique, mapping the "constrained error space" homeomorphically to an "unconstrained virtual error space," and introducing an error transformation function to convert the constrained error... Mapping to unconstrained variables The error transformation function ε(t) is defined as follows:

[0141] (7)

[0142] Its inverse transformation is:

[0143] (8)

[0144] in, : Represents the unconstrained error variable after transformation. Introducing this variable maps the error, which was originally limited to physical boundaries, to the entire real number domain. This transforms a constrained control problem into an unconstrained control problem. : Represents the actual limited tracking error of the system, that is, the deviation between the actual displacement of the vehicle body and the expected displacement; : Represents the preset performance function, which represents the boundary of the dynamic envelope that decays exponentially with time, i.e., the maximum positive boundary of the error tolerance; : Indicates the overshoot adjustment parameter, typically set to a value of It and Together, they determine the dynamic lower bound of the error, and the actual system error. Strictly restricted to Within the range; : represents the natural constant, with a value of 2.71828, and serves as the base of the exponential function; : Represents the time variable. This transformation will constrain the error to a bounded state. Mapped to unconstrained variables in the whole space .when Approaching the boundary hour, Therefore, as long as control Bounded, thus strictly guaranteed Do not cross the boundary.

[0145] (303) Constructing a logarithmic barrier Lyapunov function :

[0146] To address the physical travel limitations of electro-hydraulic active suspension systems and the high standards required for ride comfort, this embodiment departs from traditional unrestrained control methods and innovatively introduces a preset performance control strategy. The core objective of this step is to establish an insurmountable "dynamic safety channel" for the vertical displacement of the vehicle body, ensuring that the system will not experience mechanical impact under any extreme conditions.

[0147] (9)

[0148] in In this embodiment, a virtual safety limit for conversion error can be taken as... Alternatively, it can be set to a larger constant to simplify to a quadratic form, but to preserve the barrier properties, it is usually set to... A fixed safety threshold is defined. Physical protection mechanism: This function possesses unique gradient characteristics. When the system state is normal, i.e., the error is small, Approximates a traditional quadratic function The controller exhibits typical linear regulation characteristics; however, errors can occur due to severe road impacts or malfunctions. Increase and try to get closer When, the function value Its derivative, i.e., the resulting control gain, increases dramatically towards infinity. This is equivalent to establishing a virtual potential energy wall of "infinite height" at the limit of the suspension travel. Once the vehicle approaches this wall, the algorithm automatically generates a huge reverse restoring force, "pushing" the vehicle back to the safe area. This mechanism no longer relies on the hard collision of mechanical limit blocks, but achieves "soft limiting" through software algorithms, completely eliminating the risk of collision and greatly improving the safety margin of the system.

[0149] Step (4): Decompose the system based on the backstepping framework and design virtual control laws respectively; introduce a dynamic surface controller and use a first-order low-pass filter to obtain the differential of the virtual control quantity, avoid the "differential explosion" problem caused by repeated differentiation in the traditional backstepping method, and reduce the computational complexity.

[0150] To address the high-order, nonlinear, and strictly feedback-oriented characteristics of electro-hydraulic active suspension systems, this embodiment employs the backstepping method as the basic control framework. However, traditional backstepping methods require repeated analytical differentiation of the virtual control law when dealing with high-order systems. This leads to an exponential expansion of the control law expression with increasing order, making it difficult for the algorithm to run in real-time within the vehicle ECU. Therefore, this embodiment creatively introduces dynamic surface control technology, replacing complex analytical differentiation calculations with a first-order low-pass filter. This step specifically includes the following four sub-steps:

[0151] (401) Displacement Loop (First Subsystem) Control Design

[0152] First, regarding the vehicle displacement subsystem The goal is to design a virtual speed control variable. This causes displacement tracking error The preset performance constraints described in step (3) are satisfied. The logarithmic barrier Lyapunov function constructed based on step (3) is... By taking its derivative and combining it with the error transformation relationship, the first virtual control law is designed. That is, the desired vertical speed of the vehicle body:

[0153] (10)

[0154] Gain term: In this embodiment, This is the displacement loop feedback gain. A larger value means a stronger ability to correct displacement deviations, but an excessively large value may lead to overshoot. Nonlinear damping term: When the error Approaching the safety boundary When the denominator approaches zero, this term increases rapidly, generating a strong counterforce that pulls the system back to the safe zone. Intermediate variable: This is the Jacobian term generated by differentiating the error transformation. Dynamic compensation term: includes... The term is used to compensate for dynamic errors caused by performance envelope contraction, ensuring that during the rapid contraction phase of the envelope (i.e., The system can still track stably.

[0155] (402) Design of virtual control law for velocity loop (second subsystem)

[0156] Next, regarding the vehicle speed subsystem The goal is to design virtual pressure control quantities. This makes the actual speed Tracking the expected speed .

[0157] Define speed tracking error Constructing quadratic Lyapunov functions Design the second virtual control law (i.e., the desired hydraulic cylinder load pressure):

[0158] (11)

[0159] This control law consists of three parts, each with a clear physical meaning:

[0160] Error Feedback Item .in (This embodiment takes) () is the speed loop gain, which is responsible for eliminating speed tracking errors.

[0161] Inertia compensation term Introducing feedforward information about the desired acceleration improves the system's ability to respond to rapid changes.

[0162] Model Auxiliary Items Actively utilizing the physical model information of the suspension (spring force and damping force) for feedforward cancellation not only reduces the burden on feedback control but also achieves "model-assisted control," thereby improving control accuracy.

[0163] (403) Introduction of the first-stage low-pass filter and signal processing

[0164] In the traditional backstepping method, in order to design the control law for the next step, it is necessary to calculate... .because It contains complex logarithmic functions and performance functions Directly taking its derivative will produce an extremely lengthy mathematical expression and amplify measurement noise.

[0165] This embodiment introduces a first-order low-pass filter to obtain... The filtered value and its derivative:

[0166] (12)

[0167] in This represents the filter's time constant. The filter output... As the "desired velocity after filtering", and its derivative It can be obtained directly through algebraic operations, completely avoiding the need for analytical differentiation of nonlinear functions.

[0168] (404) Introduction of the second-stage low-pass filter

[0169] Similarly, in order to obtain expected pressure Differential signal For subsequent pressure loop design, a second first-order low-pass filter is introduced:

[0170] (13)

[0171] Physical parameter tuning: This is the filter time constant. The pressure loop is the inner loop, and its response speed should be faster than the outer loop (velocity loop). (Setting) This ensures that the inner loop can respond quickly to the commands of the outer loop, following the cascade control design principle of the inner loop being fast and the outer loop being slow.

[0172] Output signal: This is the filtered desired load pressure, which is the final tracking target for pressure fault-tolerant control in subsequent step (5). The pressure tracking error is defined as... .

[0173] Through the above four steps, this embodiment successfully decouples the complex nonlinear high-order system into three first-order subsystems, and solves the computational complexity problem through filtering technology, laying a solid algorithmic foundation for the final fault-tolerant control.

[0174] Step (5): Combine the observer fault compensation signal and the output of the dynamic surface controller to synthesize the final voltage control law; control the differential pressure by adjusting the opening of the servo valve to actively compensate for the force attenuation caused by internal leakage, realize fault-tolerant control under fault conditions, and ensure that the suspension performance meets the preset indicators.

[0175] For virtual control laws and Two first-order low-pass filters are introduced:

[0176] (14)

[0177] in, These are the time constants of the filter; The filtered desired velocity signal, This is the filtered desired pressure signal;

[0178] Define the dynamic surface error variable as:

[0179] (15)

[0180] First, define the third Lyapunov function for the pressure loop (third subsystem). .right Differentiating and substituting into the dynamic equation of hydraulic pressure (Equation 1), we obtain the error dynamics:

[0181] (16)

[0182] In order to Design the final servo valve control voltage :

[0183] (17)

[0184] in: , is the pressure loop gain; The fault compensation term output by the RBF neural network directly cancels out the unknown faults in the equation. ; This is a robustness term used to overcome reconstruction errors in neural networks. This mitigates boundary layer errors generated by filters, enhancing system robustness. Specifically, it can be designed in one of the following two forms: the ideal form commonly used in theoretical analysis is:

[0185] (18)

[0186] To avoid high-frequency chattering in practical engineering, this embodiment preferably uses a hyperbolic tangent function for continuous smoothing approximation, that is:

[0187] (19)

[0188] in, For robust gain, its value must satisfy the following condition: (That is, greater than the upper bound of the approximation error of the neural network), so as to theoretically guarantee the Lyapunov stability of the closed-loop system; The smoothing factor of the hyperbolic tangent function determines the degree of smoothness near the dead zone.

[0189] Step (6): System stability analysis. Based on Lyapunov stability theory, construct candidate functions for the entire system:

[0190] (20)

[0191] in For weight error, This represents the boundary layer error of the filter.

[0192] Through the Differentiate and substitute into the above control law And the adaptive law, using the scaling of Young's inequality, can be proved as follows:

[0193] (twenty one)

[0194] in This is a positive constant. This indicates that all signals in the closed-loop system (tracking error, weight estimation, control voltage) are uniformly eventually bounded (UUB), and the tracking error converges to a minimal compact set near the origin. The convergence radius can be adjusted by parameters. and Adjustments will be made.

[0195] The specific execution logic of the ECU:

[0196] The control program runs in the ECU in a 1ms cycle (1kHz):

[0197] Data acquisition: Reading signals from each sensor ,calculate And road surface input estimates.

[0198] Performance calculation: based on the current time Calculate the preset performance boundary and conversion error .

[0199] Network forward propagation: Calculating the Gaussian function vector And read the weights from the previous time step. Calculate fault estimation .

[0200] Instruction calculation: Calculate the virtual control law sequentially. Update filter state .

[0201] Output control: Calculate the voltage according to the final formula. It is then output to the servo valve drive circuit via a D / A converter.

[0202] Weight update: Update the weights according to the adaptive law. It is stored in memory for use in the next cycle.

[0203] The above description is merely a preferred embodiment of the present invention and does not limit the scope of the present invention. All equivalent structural changes made based on the description and drawings of the present invention are included within the scope of the present invention.

Claims

1. An intelligent fault-tolerant control method for electro-hydraulic active suspension considering internal leakage faults in hydraulic cylinders, characterized in that, Includes the following steps: (1) For the vehicle electro-hydraulic active suspension, considering parameter uncertainty, nonlinear friction and leakage in the hydraulic cylinder, a nonlinear dynamic state-space model of the electro-hydraulic active suspension is established; all leakage and unmodeled dynamics are uniformly represented as a generalized unknown nonlinear fault function, and road excitation is regarded as an external disturbance; (2) An RBF neural network observer is constructed for the generalized fault function, and its universal approximation characteristic is used to reconstruct the fault dynamics online; an adaptive update law for weights is designed based on Lyapunov theory to ensure that the weights are bounded and to output a fault compensation signal to offset the pressure loss caused by internal leakage and friction; (3) Introduce a preset performance control strategy, define a preset performance function to set the dynamic envelope of the error; use error transformation and logarithmic barrier Lyapunov function to ensure that the system error is strictly within the preset range in both transient and steady state stages, to prevent violation of physical constraints; (4) The system is decomposed based on the backstepping framework, and virtual control laws are designed separately. A dynamic surface controller is introduced, and a first-order low-pass filter is used to obtain the differential of the virtual control quantity, avoiding the "differential explosion" problem caused by repeated differentiation in the traditional backstepping method and reducing computational complexity. (5) Combine the fault compensation signal of the observer with the output of the dynamic surface controller to synthesize the final voltage control law; control the pressure difference by adjusting the opening of the servo valve to actively compensate for the force attenuation caused by internal leakage, realize fault-tolerant control under fault conditions, and ensure that the suspension performance meets the preset indicators.

2. The intelligent fault-tolerant control method for electro-hydraulic active suspension considering leakage faults in hydraulic cylinders according to claim 1, characterized in that, The nonlinear dynamic state-space model of the electro-hydraulic active suspension in step (1) is specifically represented as follows: (1) in, Let be the system state vector. Represents the vertical displacement of the vehicle body. Represents the vertical speed of the vehicle body. Represents the load pressure of the hydraulic cylinder; These represent the vehicle's vertical velocity, vertical acceleration, and the rate of change of hydraulic cylinder load pressure, respectively. For the sprung mass; and These are the suspension stiffness and damping coefficient, respectively. Input is the vertical displacement of the road surface. Input the vertical velocity of the road surface; The effective area of ​​the hydraulic cylinder piston; The effective bulk modulus of hydraulic oil; This refers to the total volume of the hydraulic cylinder; The total leakage coefficient; For servo valve flow gain; For control voltage input; Let be a generalized unknown nonlinear fault function that includes nonlinear internal leakage and friction, and assume that ,in It is the physical upper limit of the degree of leakage in the hydraulic cylinder, which cannot be measured by conventional means but can be approximated by algorithms.

3. The intelligent fault-tolerant control method for electro-hydraulic active suspension considering leakage faults in hydraulic cylinders according to claim 1, characterized in that, The RBF neural network observer and the adaptive weight update law in step (2) specifically include: (201) Establish a neural network approximation model: Generalized unknown nonlinear fault function Approaching as: (2) in, For the ideal weight vector, This represents the Gaussian radial basis function vector of the hidden layer. For network approximation error and ; Fault Observer Output Represented as: (3) in, This represents the weight estimation vector for the neural network. This represents the number of hidden layer nodes. Let be a Gaussian radial basis function vector, the first... basis functions Defined as: (4) in, For the first The center vector of each node This is the width parameter of the node. The Euclidean norm is represented by exp; the natural exponential function is represented by exp. (202) Online adaptive update law for weights: The design incorporates an online adaptive update law for weights to eliminate tracking errors in the pressure loop. The specific form is as follows: (5) in, The learning rate gain matrix is ​​a positive definite symmetric matrix, which determines the convergence speed of the weights; This represents the Gaussian radial basis function vector of the hidden layer. The dynamic surface error variable of the pressure ring; To robustly adjust parameters, preventing weight drift and ensuring an online adaptive update law for weights. Bounded.

4. The intelligent fault-tolerant control method for electro-hydraulic active suspension considering leakage faults in hydraulic cylinders according to claim 1, characterized in that, The construction process of the preset performance function and the barrier Lyapunov function in step (3) is as follows: (301) Define the preset performance function : The exponential decay function is selected as the performance envelope: (6) in, As the initial performance boundary, it must satisfy ,in The displacement tracking error at the initial moment; This represents the maximum permissible error in steady state, and Convergence rate factor; system tracking error Must always satisfy ,in This is the overshoot adjustment parameter; (302) Construct the error transformation function : Introducing an error transformation function will limit the error. Mapping to unconstrained variables : (7) Its inverse transformation is: (8) : Represents the unconstrained error variable after transformation. Introducing this variable maps the error, which was originally limited to physical boundaries, to the entire real number domain. This transforms a constrained control problem into an unconstrained control problem. : Represents the actual limited tracking error of the system, that is, the deviation between the actual displacement of the vehicle body and the expected displacement; : Represents the preset performance function, which represents the boundary of the dynamic envelope that decays exponentially with time, i.e., the maximum positive boundary of the error tolerance; : Indicates the overshoot adjustment parameter, typically set to a value of It and Together, they determine the dynamic lower bound of the error, and the actual system error. Strictly restricted to Within the range; : represents the natural constant, with a value of 2.71828, and serves as the base of the exponential function; : Represents a time variable; (303) Constructing a logarithmic barrier Lyapunov function : For the displacement subsystem, a logarithmic barrier Lyapunov function is selected: (9) in, The transformed error variable The preset virtual safety boundary constant, the function property guarantees that when This results in an infinite penalty gain in the controller design to constrain the error boundary.

5. The intelligent fault-tolerant control method for electro-hydraulic active suspension considering leakage faults within the hydraulic cylinder as described in claim 1, characterized in that, The virtual control law in step (4) includes the virtual control law for the displacement loop. And speed loop virtual control law The specific design process is as follows: (401) Displacement loop virtual control law design: Define displacement tracking error ,in The desired displacement; based on the logarithmic barrier Lyapunov function in step (3). Design of virtual control law for displacement : (10) : Represents the virtual control law of the displacement loop; : indicates the displacement loop control gain, and This parameter is used to adjust the convergence speed of displacement tracking error; the larger the value, the faster the error is eliminated. : Indicates intermediate substitution variables generated during the differentiation process; : Indicates the overshoot adjustment parameter; : Represents the actual displacement tracking error; where, The displacement loop control gain is used to achieve error constraints. Based on the specific logarithmic barrier Lyapunov function and error transformation function constructed in the previous step, it is derived through mathematical differentiation and scaling. It contains specific nonlinear penalty terms to achieve suspension collision avoidance. (402) Velocity loop virtual control law design: Define speed tracking error ,in for Output after low-pass filter; design of velocity loop virtual control law. , where is the desired hydraulic cylinder pressure: (11) : Represents the virtual control law of the speed loop; : Indicates the sprung mass of a vehicle; : Indicates the effective area of ​​the hydraulic cylinder piston; : indicates the speed loop control gain, and Used to adjust speed tracking error The convergence speed; : Indicates the speed tracking error between the actual vertical speed of the vehicle body and the desired speed; : Represents the desired velocity signal after filtering; : Represents the time constant of the first low-pass filter in dynamic surface control; : Indicates the suspension stiffness coefficient; : Indicates the suspension damping coefficient; : Indicates the actual vertical displacement of the vehicle body; : Indicates the actual vertical speed of the vehicle body; : Indicates the vertical displacement input of the road surface; : Indicates the vertical velocity input of the road surface; , which is the speed loop control gain; the speed loop virtual control law is derived by combining a dynamic surface controller, using the filtered signal instead of analytical differentiation, and incorporating the feedforward compensation term of the suspension physical model. (403) Introduction of the first-stage low-pass filter and signal processing In the traditional backstepping method, in order to design the control law for the next step, it is necessary to calculate... ;because It contains complex logarithmic functions and performance functions Directly taking its derivative will produce an extremely lengthy mathematical expression and amplify measurement noise; Introducing the first first-order low-pass filter to obtain The filtered value and its derivative: (12) in The filter time constant; the filter output. As the "desired velocity after filtering", its derivative It can be obtained directly through algebraic operations, completely avoiding the analytical differentiation of nonlinear functions; (404) Introduction of the second-stage low-pass filter Similarly, in order to obtain expected pressure Differential signal For subsequent pressure loop design, a second first-order low-pass filter is introduced: (13) Physical parameter tuning: The filter time constant; the pressure loop is the inner loop, and its response speed should be faster than the outer loop (velocity loop); set This ensures that the inner loop can respond quickly to the commands of the outer loop, following the cascade control design principle of the inner loop being fast and the outer loop being slow; Output signal: This is the filtered desired load pressure, which is the final tracking target of the pressure fault-tolerant control in the subsequent step (5); the pressure tracking error is defined as... .

6. The intelligent fault-tolerant control method for electro-hydraulic active suspension considering internal leakage faults in hydraulic cylinders according to claim 5, characterized in that, The first-order low-pass filter in step (4) is as follows: For virtual control laws and Two first-order low-pass filters are introduced: (14) in, These are the time constants of the filter; The filtered desired velocity signal, This is the filtered desired pressure signal; Define the dynamic surface error variable as: (15) In the controller design, an extremely complex logarithmic barrier Lyapunov function was introduced to prevent collisions, resulting in an exceptionally large virtual control law. Directly differentiating it using the traditional backstepping method would trigger a severe "differential explosion," causing the ECU's computing power to collapse. To avoid this collapse, the output of the filter is directly used... and Replace virtual control law and The analytical derivative is used to eliminate the computational bloat caused by higher-order nonlinear terms, thus enabling the overall fault-tolerant control architecture to be realized.

7. The intelligent fault-tolerant control method for electro-hydraulic active suspension considering internal leakage faults in hydraulic cylinders according to claim 1, characterized in that, The final voltage control law in step (5) The calculation expression is: (16) in: This is the control gain coefficient of the pressure loop (or force loop); To obtain the virtual control quantity differential signal using a dynamic surface controller; For the output of the RBF neural network targeted Fault compensation items; This is a robustness term used to overcome reconstruction errors in neural networks. To mitigate external residual interference and ensure the stability of the closed-loop system, its design is as follows: (17) (18) : Indicates the final servo valve control voltage input; : Represents the control gain coefficient of the pressure loop, and is a positive number. It determines the convergence speed of the actual pressure tracking the desired pressure. : Indicates pressure tracking error; : This indicates the robust compensation control term. This term is introduced to overcome the inherent reconstruction approximation error of the neural network and the residual disturbances that are not measurable from the outside, and to prevent these errors from causing the system to diverge. : represents the robust gain constant, whose value must be greater than the upper bound of the neural network approximation error and residual disturbance to theoretically guarantee the Lyapunov stability of the closed-loop system; where, For robustness gain, The smoothing factor for the hyperbolic tangent function is... The function is a sign function; it passes through displacement loop, velocity loop, and finally pressure loop. Using Lyapunov stability theory, the final voltage control law is derived. This final voltage control law includes a fault compensation term output by the RBF neural network fault observer. And the virtual control quantity differential signal obtained using the low-pass filter of the dynamic surface controller. A solution is proposed for the specific physical fault of internal leakage in the suspension: intelligent fault-tolerant control.

8. The intelligent fault-tolerant control method for electro-hydraulic active suspension considering leakage faults within hydraulic cylinders according to any one of claims 1-7, characterized in that, It also includes step (6), which is a system stability analysis that requires constructing a Lyapunov candidate function for the entire system. : (19) in, For weight estimation error, For filter boundary layer error; By adjusting the control gain Filter time constant and robust parameters , making ,in Since it is a positive constant, it proves that all signals in the closed-loop system are uniformly bounded and that the tracking error converges to a compact set near the origin.

Images