Data-based fault-tolerant control method for on-orbit satellite

Abstract

The invention belongs to the technical field of satellite fault-tolerant control, and discloses a data-based in-orbit satellite fault-tolerant control method, which comprises the steps of data-driven control framework construction, double-dynamic linear processing, real-time fault detection and fuzzy neural network fault-tolerant compensation. According to the scheme, construction of a complex dynamic model is avoided; the control bottleneck of nonlinear scenes such as a strong-coupling multiple-input multiple-output system and large-angle attitude adjustment is broken through; a fault model does not need to be preset, and the quick response requirement of sudden faults in a space environment is met; a fuzzy membership function is utilized to quantify a progressive fault, fault estimation is realized through neural network weight adaptive updating, an estimated value is directly incorporated into control law correction, the fault influence is offset, and the limitation of traditional fixed rule compensation is avoided.

Description

Technical Field

[0001] The present invention belongs to the technical field of satellite fault-tolerant control, and in particular relates to a data-based fault-tolerant control method for an on-orbit satellite. Background Art

[0002] The in-orbit satellite utilizes a multi-sensor design and dual-backup actuators, ensuring excellent stability. This configuration effectively mitigates sensor failures and data loss caused by space radiation, enabling stable operation in the complex space environment. Furthermore, its data processing architecture provides efficient computing support, enabling rapid mobilization of redundant resources in the event of a sudden failure to ensure the normal operation of essential functions. This is crucial for fields such as communications, remote sensing, and deep space exploration, making it an indispensable core piece of equipment in these fields. For existing satellites in orbit, fault-tolerant control algorithms are key to achieving precise control and stable operation. By precisely adjusting actuator outputs such as momentum wheel speed and thrust, and reconstructing faulty sensor data, satellites can ensure efficient mission execution under various operating conditions. However, in the unstructured environment of space, faults are sudden and complex. Faults such as equipment anomalies caused by solar storms and component aging caused by long-term operation place extremely high demands on the adaptability and robustness of fault-tolerant control algorithms. Existing fault-tolerant control methods, such as model-based fault reconstruction and adaptive observer-based methods, often rely on complex satellite dynamics models and fault mechanism modeling. This not only increases the difficulty of algorithm development and implementation, but also limits their adaptability to the volatile space environment. When faced with unknown fault types, these methods struggle to respond and adjust quickly, potentially leading to satellite attitude instability and mission interruption. Therefore, designing fault-tolerant control algorithms that do not require complex modeling, exhibit fast response speed, and exhibit strong robustness is of great significance for improving satellite operational reliability in space.

[0003] The following methods are currently common: 1. Fault diagnosis and finite-time fault-tolerant control method based on adaptive observer CN 107861383 A: Construct a satellite kinematic and dynamic model under the influence of actuator failure and external interference; design an observer, establish a residual evaluation function through the deviation between the observer output and the actual system output, and determine that a fault has occurred when the evaluation function exceeds the threshold, ensuring robustness to interference and sensitivity to faults; after the fault occurs, use an adaptive observer to estimate the fault size and change online to achieve accurate identification of the fault; 2. The relatively common one is the sliding mode fault-tolerant control method, such as the discrete integral sliding mode fault-tolerant control method for satellite control systems CN103699131A: Using a system model without faults and uncertainties as a reference model, calculate the residual and sliding surface variables to achieve basic control of the system; obtain the actuator health factor through fault diagnosis and parameter identification, provide a basis for subsequent control allocation, and clarify the health status of the actuator; adopt a direct allocation method based on the torque reachable set to obtain the control torque of each actuator to improve allocation efficiency; 3. The more cutting-edge technology uses knowledge-driven methods to build models to achieve adaptive control of satellite attitude (1000-1328(2023)06-0829-12). The knowledge-driven approach establishes a satellite ACS model including dynamics, uses nonlinear Kalman filtering to estimate states and predict faults, or builds an if-then rule base based on expert knowledge to infer faults and then adjust actuators to achieve fault tolerance.

[0004] Deficiencies of existing technology: 1. Existing satellite attitude control methods are mostly based on precise system models. However, after entering orbit, changes in mass distribution due to fuel consumption and structural parameter drift caused by thermal deformation can cause the nominal model to gradually deviate from the actual system (i.e., "model mismatch"), directly impairing control accuracy. When a satellite experiences sensor or actuator failure, system characteristics can further deviate from the nominal model.

[0005] 2. While current satellite fault detection mechanisms employ model-based approaches to identify faults, the design of detection thresholds and residual analysis relies on precise satellite mathematical models, making them difficult to adapt to on-orbit parameter drift. For satellite systems subject to strong spatial interference and multi-component coupling, model construction and modification can become extremely complex. Parameter tuning of traditional fault detection algorithms typically requires extensive ground-based simulations and on-orbit debugging, making them incapable of handling the rapid system changes caused by sudden faults.

[0006] 3. While current adaptive control methods attempt to address parameter variations, they still rely on pre-defined system nonlinear structures (e.g., known coupling terms), making them difficult to handle inexplicably nonlinear scenarios such as large-angle satellite attitude adjustments and thruster thrust decay. Convergence analysis for some methods relies on global topology information or global system parameters. Single-satellite control often involves MIMO scenarios. Existing technologies for handling strongly coupled MIMO systems often require complex decoupling algorithms designed to address these coupling relationships. These algorithms also struggle to maintain decoupling accuracy when model parameters drift, resulting in reduced control effectiveness.

[0007] 4. Traditional fault-tolerant control methods have obvious limitations: These methods usually require a precise dynamic model of the satellite as a basis, and design compensation strategies through preset fault models. However, when the satellite is in orbit, changes in rotational inertia caused by fuel consumption and structural parameter drift caused by thermal deformation will cause deviations between the nominal model and the actual system. Model mismatch directly leads to a decrease in fault detection accuracy and even false alarms or missed alarms. At the same time, its fault handling relies heavily on fixed rules, making it difficult to deal with ambiguous faults. Summary of the Invention

[0008] To overcome the above technical problems, the present invention provides a data-based fault-tolerant control method for on-orbit satellites. This method includes dynamic linearization of on-orbit satellite attitude control data, construction of an ideal controller, and design of a data-driven fault-tolerant control algorithm for satellite sensor failures.

[0009] The present invention adopts the following technical solutions: The core of the program is divided into four steps: Data-driven control framework construction: A precise satellite dynamics model is not required. Control relationships are established solely through input and output data from control commands (such as reaction wheel torque and thruster force) and satellite status (such as attitude angle and angular velocity). Control parameters are updated using real-time operational data to adapt to model parameter drift caused by factors such as fuel consumption and thermal deformation.

[0010] Dual dynamic linearization processing: The nonlinear dynamic characteristics of the satellite (such as the coupling effect during large-angle attitude adjustment) and the ideal controller are converted into equivalent linear data models respectively. The input-output relationship is quantified through pseudo-partial derivatives to solve the control problems in strong nonlinear scenarios.

[0011] Real-time fault detection: Design an output observer to compare the deviation between the actual sensor measurement value and the estimated value in real time. Identify the fault type (such as sensor bias, noise enhancement) and characteristic parameters (such as offset and noise intensity) through the deviation amplitude and change rate, and achieve rapid fault location.

[0012] Fuzzy neural network fault-tolerant compensation: Fuzzy neural networks are trained based on historical fault data. Progressive faults (such as slow gyroscope drift) are quantified through fuzzy membership functions. The neural network dynamically optimizes the compensation strategy and outputs corrected control instructions to offset the impact of faults on control performance.

[0013] Compared with the prior art, the present invention has the following beneficial effects: The fault-tolerant control method designed by the present invention, when applied to the attitude control system of an on-orbit satellite, only utilizes input and output data during operation. The proposed control method does not require any precise model information, including unknown inertial variables, thus avoiding the construction of a complex dynamic model. In the data-driven framework, the pseudo-Jacobian matrix ( 、 ) Convert the satellite's nonlinear attitude dynamics and ideal controller into linear data models, respectively. Combined with a parameter reset mechanism, ensure that the model tracks time-varying characteristics, breaking through the control bottleneck of nonlinear scenarios such as strongly coupled multi-input and multi-output systems and large-angle attitude adjustments; Design the fault decision logic based on the output observer by estimating the error and the dynamic threshold ( ) to quickly identify sensor faults without presetting fault models, adapting to the rapid response requirements of sudden faults in space environments; Fuzzy membership functions are used to quantify progressive faults (such as gyroscope drift), and fault estimation is achieved through adaptive updating of neural network weights (including learning rate and momentum coefficient adjustment). The estimated value is directly incorporated into the control law correction to offset the impact of the fault and avoid the limitations of traditional fixed rule compensation. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is the fault-tolerant control block diagram of an on-orbit satellite. DETAILED DESCRIPTION

[0015] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings. The same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. Unless otherwise specified, the raw materials and equipment used are commercially available or commonly used in the art. The methods in the embodiments, unless otherwise specified, are conventional methods in the art. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.

[0016] A data-based fault-tolerant control method for on-orbit satellites is used for on-orbit satellite system control. The core is divided into four steps: 1. Data-driven control framework construction: This approach eliminates the need for a precise satellite dynamics model. Instead, the control relationship is established using input and output data from control commands (such as reaction wheel torque and thruster force) and satellite status (such as attitude angle and angular velocity). Control parameters are updated using real-time operational data to adapt to model parameter drift caused by factors such as fuel consumption and thermal deformation.

[0017] 2. Dual dynamic linearization processing: The nonlinear dynamic characteristics of the satellite (such as the coupling effect during large-angle attitude adjustment) and the ideal controller are converted into equivalent linear data models. The input-output relationship is quantified through pseudo-partial derivatives to solve the control problems in strong nonlinear scenarios.

[0018] 3. Real-time fault detection: Design an output observer to compare the deviation between the actual sensor measurement value and the estimated value in real time. Identify the fault type (such as sensor bias, noise enhancement) and characteristic parameters (such as offset and noise intensity) through the deviation amplitude and change rate to achieve rapid fault location.

[0019] 4. Fuzzy neural network fault-tolerant compensation: Fuzzy neural networks are trained based on historical fault data. Progressive faults (such as slow gyroscope drift) are quantified through fuzzy membership functions. The neural network dynamically optimizes the compensation strategy and outputs corrected control instructions to offset the impact of faults on control performance.

[0020] The technical solutions are as follows: 1. Linearization of nonlinear attitude data in on-orbit satellites Step 1.1: Obtain the control torque in the satellite control system during operation and actual attitude speed , using nonlinear autoregressive moving average model to describe the system dynamic characteristics (1) in, , is about the control torque in the satellite data and actual attitude speed A nonlinear function of the data, right The partial derivative of is continuous, and are the angular velocity vectors at the current moment and the next moment respectively. , corresponding to the angular velocity of the satellite's X, Y, and Z axes. is the control torque at the current moment (bounded signal). , corresponding to the moments of the X, Y, and Z axes. is the time interval, and Unknown parameters.

[0021] Step 1.2: Use the tight-form dynamic linearization method of model-free adaptive control to transform the satellite nonlinear attitude control system model into a dynamic linear model.

[0022] (2) in, . It is the pseudo-Jacobian matrix between the input and output differential signals and changes dynamically. , and for any moment , , , , , .in, .

[0023] Step 1.3: In the actual satellite attitude control system, it is impossible to directly obtain the , so the estimation method is designed based on the criterion function: (3) in, for The estimated value of can be used to derive the estimated dynamic model , To prevent the denominator from being 0.

[0024] In order to enhance the ability of the estimation algorithm to track time-varying parameters, a reset algorithm is introduced: when ,or ,but ; when ,or ,but .

[0025] 2. Fault Detection Mechanism Design Step 2.1: Design a fault detection observer (4) in, yes The estimated value of Indicates feedback gain. It is the control output The estimation error.

[0026] Step 2.2: Design the fault detection threshold (5) in, , .

[0027] Step 2.3: Based on the above steps, the system determines whether to perform fault-tolerant control: When , no fault-tolerant control is performed; when When a fault occurs, the system defaults to fault-tolerant control.

[0028] 3. Ideal controller design Step 3.1: According to the actual attitude angular velocity of the satellite and target angular velocity To define the error in angular velocity: (6) in, = , For satellite operation Required angular velocity. .

[0029] Step 3.2: Obtain the control torque in the satellite control system during operation and the error in angular velocity during operation , using nonlinear autoregressive moving average model to describe the system dynamic characteristics (7) in, , is about the control torque in the satellite data and angular velocity error A nonlinear function of the data, right The partial derivative of is continuous, and are the angular velocity errors at the current moment and the next moment respectively. , corresponding to the angular velocity error of the satellite X, Y, and Z axes. is the control torque at the current moment (bounded signal). , corresponding to the moments of the X, Y, and Z axes. is the time interval, and Unknown parameters.

[0030] Step 3.3: Use the tight-form dynamic linearization method of model-free adaptive control to transform the satellite nonlinear attitude control system model into a dynamic linear model.

[0031] (8) in, . It is the pseudo-Jacobian matrix between the input and output differential signals and changes dynamically. , and for any moment , , , , , .in, Among them, there is a non-causal term in formula (8) , which means that Equation (8) cannot be used directly in practical engineering. However, as an ideal controller, the output of the ideal controller (8) will generate an ideal control input, achieving perfect tracking one step ahead, that is, At this point, the actual controller (8) becomes (9) Step 3.4: In the actual satellite attitude control system, it is impossible to directly obtain the , so the estimation method is designed based on the criterion function: (10) in, for The estimated value of can be used to derive the estimated dynamic model , To prevent the denominator from being 0.

[0032] In order to enhance the ability of the estimation algorithm to track time-varying parameters, a reset algorithm is introduced: , or .

[0033] 4. Fault-tolerant control design Step 4.1: Based on the detection mechanism, the system is judged to have suffered a fault, and fault-tolerant control is mainly performed for sensor failure. Due to the combination of the reasoning ability of fuzzy logic and the self-learning ability of neural networks, fuzzy neural networks are used to change the adaptive law.

[0034] (11) in, is the estimated value for unknown faults, Represents fuzzy rules, membership function center (Determines the core interval of the rule) and the width of the membership function (determines the coverage of the rule) are important parameters in the Gaussian membership function. One of the independent variables of the function Represents the input variable.

[0035] Step 4.2: Use the current estimation error to correct the historical weights, weights The update follows the following recurrence relationship: (12) in is the learning rate (controls the update step size), is the momentum coefficient (accelerates convergence).

[0036] Step 4.3: Continuously learn through neural networks, weights Gradually optimize to Converge to the real fault. Fuzzy neural network estimates the fault After the fault occurs, the control strategy is used to offset its impact on the system. The designed fault-tolerant control law directly incorporates the fault estimate into the input adjustment.

[0037] The final overall controller is expressed as: (13) (14)

[0038] Although the embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations may be made to the above embodiments without departing from the principles and purpose of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A data-based fault-tolerant control method for an on-orbit satellite, characterized in that: include: Data-driven control framework architecture: Build control relationships through input and output data of control commands and satellite status; Update control parameters through real-time collected operating data to adapt to model parameter drift caused by fuel consumption and thermal deformation; Dual dynamic linearization processing: The nonlinear dynamic characteristics of the satellite and the ideal controller are converted into equivalent linear data models, and the input-output relationship is quantified through pseudo-partial derivatives to solve the control problem in strong nonlinear scenarios. Real-time fault detection: Design an output observer to compare the deviation between the actual sensor measurement value and the estimated value in real time, identify the fault type and characteristic parameters through the deviation amplitude and change rate, and realize rapid fault location; Fuzzy neural network fault tolerance compensation: A fuzzy neural network is trained based on historical fault data, and progressive faults are quantified through fuzzy membership functions. The neural network dynamically optimizes the compensation strategy and outputs corrected control instructions to offset the impact of faults on control performance.

2. The data-based fault-tolerant control method for an on-orbit satellite according to claim 1, characterized in that: Linearization of nonlinear attitude data in an on-orbit satellite: Step 1: Obtain the control torque in the satellite control system during operation and actual attitude speed ,The nonlinear autoregressive moving average model is used to describe the system dynamic characteristics; (1) in, , is about the control torque in the satellite data and actual attitude speed A nonlinear function of the data, right The partial derivative of is continuous, and are the angular velocity vectors at the current moment and the next moment respectively; , corresponding to the angular velocity of the satellite's X, Y, and Z axes; is the control torque at the current moment (bounded signal); , corresponding to the moments of the X, Y, and Z axes; is the time interval, and Unknown parameters; Step 2: Using the tight-form dynamic linearization method of model-free adaptive control, the satellite nonlinear attitude control system model is converted into a dynamic linear model; (2) in, ; It is the pseudo-Jacobian matrix between the input and output differential signals, which changes dynamically; , and for any moment , , , , , ;in, ; Step 3: In the actual satellite attitude control system, it is impossible to obtain directly without considering the system model , so the estimation method is designed based on the criterion function: (3) in, for The estimated value of can be used to derive the estimated dynamic model , To prevent the denominator from being a constant of 0; In order to enhance the ability of the estimation algorithm to track time-varying parameters, a reset algorithm is introduced: when ,or ,but ; when ,or ,but .

3. The method for fault-tolerant control of an on-orbit satellite based on data according to claim 1, characterized in that: Fault detection mechanism design: Step 1: Design a fault detection observer: (4) in, yes The estimated value of represents feedback gain; It is the control output The estimation error of Step 2: Design the fault detection threshold: (5) in, , ; Step 3: Based on the above steps, the system determines whether to perform fault-tolerant control: When , no fault-tolerant control is performed; when When a fault occurs, the system defaults to fault-tolerant control.

4. The method for fault-tolerant control of an on-orbit satellite based on data according to claim 1, characterized in that: Ideal controller design: Step 1: According to the actual attitude angular velocity of the satellite and target angular velocity To define the error in angular velocity: (6) in, = , The angular velocity required for the satellite to operate; ; Step 2: Obtain the control torque in the satellite control system during operation and the error in angular velocity during operation , the nonlinear autoregressive moving average model is used to describe the dynamic characteristics of the system: (7) in, , is about the control torque in the satellite data and angular velocity error A nonlinear function of the data, right The partial derivative of is continuous, and are the angular velocity errors at the current moment and the next moment respectively; , corresponding to the angular velocity error of the satellite X, Y, and Z axes; is the control torque at the current moment (bounded signal); , corresponding to the moments of the X, Y, and Z axes; is the time interval, and Unknown parameters; Step 3: Use the tight-form dynamic linearization method of model-free adaptive control to transform the satellite nonlinear attitude control system model into a dynamic linear model: (8) in, ; It is the pseudo-Jacobian matrix between the input and output differential signals, which changes dynamically; , and for any moment , , , , , ; in, ; Among them, there is a non-causal term in formula (8) , which means that Equation (8) cannot be used directly in practical engineering; however, as an ideal controller, the output of the ideal controller (8) will generate an ideal control input to achieve perfect tracking one step ahead, that is, ; At this point, the actual controller (8) becomes: (9) Step 4: In the actual satellite attitude control system, it is impossible to obtain directly without considering the system model , so the estimation method is designed based on the criterion function: (10) in, for The estimated value of can be used to derive the estimated dynamic model , To prevent the denominator from being a constant of 0; In order to enhance the ability of the estimation algorithm to track time-varying parameters, a reset algorithm is introduced: , or .

5. The data-based fault-tolerant control method for an on-orbit satellite according to claim 1, characterized in that: Fault-tolerant control design: Step 1: Determine if the system has experienced a fault based on the detection mechanism, and perform fault-tolerant control mainly for sensor failures. Fuzzy neural networks are used to change the adaptive law due to the combination of fuzzy logic reasoning and neural network self-learning capabilities. (11) in, is the estimated value for unknown faults, Represents fuzzy rules, membership function center (Determines the core interval of the rule) and the width of the membership function (determines the coverage of the rule) are important parameters in the Gaussian membership function; one of the independent variables of the function represents the input variable; Step 2: Use the current estimation error to correct the historical weights, weights The update follows the following recurrence relationship: (12) in is the learning rate (controls the update step size), is the momentum coefficient (accelerated convergence); Step 3: Continuously learn through neural networks, weights Gradually optimize to Converge to the real fault; fuzzy neural network estimates the fault After that, the influence of the fault on the system needs to be offset by the control strategy; the designed fault-tolerant control law directly incorporates the fault estimate into the input adjustment; The final overall controller is expressed as: (13) (14)。

Images