Model and data driven fault diagnosis methods for deep space exploration spacecrafts

Abstract

This invention discloses a model- and data-driven fault diagnosis method for deep space exploration spacecraft, belonging to the field of deep space exploration technology. Addressing the shortcomings of traditional methods in model adaptability, data utilization, and fault location, the method constructs a four-layer fusion architecture comprising the entity layer, model layer, data layer, and application layer of the deep space exploration spacecraft system, achieving dynamic integration and interaction between the mechanistic model and multi-source telemetry data. Through a mechanistic-data hybrid model, real-time system state prediction, fault injection simulation, multi-source data comparative analysis, and health status assessment are achieved. A corresponding software prototype has been developed, possessing functions such as fault mode library management, telemetry-simulation data-driven comparison, 3D visualization, and simulation verification of fault handling strategies. This method can effectively assist in the rapid and accurate location and handling decisions of on-orbit faults, improving the safety and reliability of deep space exploration missions.

Description

Technical Field

[0001] This invention belongs to the field of deep space exploration technology, specifically involving a model- and data-driven method for diagnosing faults in deep space exploration spacecraft. Background Technology

[0002] As deep space exploration missions evolve towards greater distances, longer durations, and higher complexity, the reliable operation and rapid fault handling of spacecraft systems have become key challenges to mission success. Traditional spacecraft fault diagnosis methods primarily rely on the experience and rule-based judgments of ground experts based on telemetry data, which suffers from low diagnostic timeliness, ambiguous fault location, and poor adaptability to unknown faults. Especially in the deep space environment, the large communication delays and difficulties in real-time intervention place higher demands on autonomous fault diagnosis and prediction capabilities.

[0003] In recent years, digital twin technology has provided new insights into spacecraft health management, enabling virtual-real integrated state monitoring and simulation analysis by constructing high-fidelity physical models. However, existing methods still have shortcomings in the deep integration of models and data: on the one hand, mechanistic models often fail to fully cover complex on-orbit conditions and degradation behaviors, leading to model prediction biases; on the other hand, the value of massive telemetry data has not been fully explored, and the fusion analysis and real-time driving capabilities of multi-source heterogeneous data are weak. Furthermore, the fault diagnosis process still relies heavily on fixed thresholds and preset rules, and the accuracy and interpretability of diagnosing complex situations such as cascading faults and intermittent faults need to be improved.

[0004] Currently, data-driven artificial intelligence methods are being gradually applied in fault diagnosis, such as using machine learning and deep learning for anomaly detection and pattern recognition. However, purely data-driven methods rely on a large number of labeled samples, which limits their generalization ability in scenarios where deep space missions have few fault samples and variable operating conditions, and they also lack physical interpretability. Therefore, how to effectively combine the physical interpretability of mechanistic models with the adaptive capabilities of data-driven methods to form a fusion-driven fault diagnosis system has become an important research direction for improving the autonomous health management capabilities of deep space exploration spacecraft. Summary of the Invention

[0005] To address the aforementioned technical issues, this invention provides a model- and data-driven method for diagnosing deep space exploration spacecraft faults. By constructing a mechanism- and data-driven hybrid architecture, it enables real-time state prediction, precise fault location, and simulation verification of handling strategies, providing reliable technical support for deep space exploration missions.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] Model- and data-driven methods for fault diagnosis of deep space exploration spacecraft include:

[0008] Step 1: Construct a four-layer fault diagnosis architecture comprising the detector system entity layer, model layer, data layer, and application layer;

[0009] Step 2: Deploy sensors in the physical layer of the detector system to collect data on the physical operational status of the spacecraft;

[0010] Step 3: In the model layer, construct a normal behavior mechanism model for predicting operating status and a fault behavior mechanism model for simulating fault data;

[0011] Step 4: In the data layer, receive and manage the on-orbit telemetry data collected from the physical layer of the detector system, and map and fuse the on-orbit telemetry data with the simulation data generated by the model layer;

[0012] Step 5: At the application layer, fault diagnosis is performed using the fused data, including: real-time status prediction and health assessment of the detector system based on the normal behavior mechanism model and the fused data; when an anomaly is detected or fault analysis is required, the fault mode is injected into the fault behavior mechanism model to perform fault simulation and realize fault location and diagnosis.

[0013] Furthermore, step 3 includes: constructing a unified multidisciplinary model library based on the Modelica modeling language, encompassing energy systems, thermal control systems, propulsion systems, and control systems; constructing component-level, single-machine-level, subsystem-level, and system-level models in a bottom-up manner based on the physical topology; and iteratively fitting and calibrating the model parameters using the least squares method based on ground test data or historical on-orbit data.

[0014] Furthermore, step 4 specifically includes: receiving and parsing the telemetry data stream and extracting effective physical quantity values; establishing a structured mapping relationship between telemetry parameters and model variables through an XML-based mapping configuration file; converting the telemetry data into parameters recognizable by the model based on the mapping relationship, and injecting them into the simulation model to drive the simulation operation.

[0015] Furthermore, in step 5, the real-time state prediction and health assessment includes: estimating the unmeasurable state of the system using an observer and filtering algorithm; and performing real-time comparison and health scoring of simulation results and telemetry data based on thresholds and association rules in the knowledge base.

[0016] Furthermore, in step 5, fault location and diagnosis includes: calculating the similarity between the telemetry curve and the simulation curve using a dynamic time warping algorithm; and matching abnormal features to determine the fault type and confidence level by combining the fault mode library and the interpretation knowledge base.

[0017] Furthermore, step 5 also includes fault handling verification: extracting candidate handling strategies from the contingency plan library and converting them into control parameters that the model can recognize; performing strategy deduction simulation in the fault model to evaluate safety, task impact and energy balance indicators; using the analytic hierarchy process to comprehensively score the candidate strategies and output the optimal handling solution.

[0018] Furthermore, step 5 also includes a closed-loop verification mechanism: after the actual command is executed, the telemetry data and simulation prediction results are continuously compared; if the deviation exceeds the threshold, the strategy backtracking and model parameter correction are triggered to optimize the handling effect.

[0019] Furthermore, the fault behavior mechanism model is constructed using an object-oriented inheritance and redeclaration mechanism, supporting the unified extension and management of single-point fault models, system fault models, and fault display models.

[0020] In a second aspect, the present invention provides an electronic device comprising: one or more processors; and a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned model and data-driven deep space exploration spacecraft fault diagnosis method.

[0021] Thirdly, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enable the processor to implement the aforementioned model- and data-driven deep space exploration spacecraft fault diagnosis method.

[0022] The beneficial effects of this invention are as follows:

[0023] Improve the accuracy and timeliness of fault diagnosis: Through the deep integration of mechanism models and telemetry data, dynamic tracking and real-time simulation comparison of spacecraft status are realized, which can detect anomalies earlier and more accurately, significantly shorten the fault location time, and is especially suitable for long-latency communication constraints in deep space environment.

[0024] Enhanced adaptability to complex and unknown faults: By combining a fault mode library with data-driven analysis, it can not only diagnose known faults but also identify unknown or cascading faults through anomaly detection and correlation analysis. The converged architecture supports iterative analysis of "hypothesis-simulation-verification," improving the diagnostic capability for coupled faults in multiple systems.

[0025] Enables pre-verification and optimization of fault handling: Supports the injection of faults into digital twin models and performs ultra-real-time simulations, enabling pre-simulation and evaluation of various handling strategies, thereby selecting safe and effective handling solutions, reducing on-orbit handling risks, and improving the scientific nature and reliability of mission decisions.

[0026] Enhancing the system's engineering practicality and scalability: The proposed four-layer architecture and modular design facilitate integration with existing ground-based telemetry and control systems. It supports online model updates, knowledge base expansion, and rapid embedding of new fault modes, providing a customizable and scalable fault diagnosis solution for various deep space exploration missions. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the model and data-driven fault diagnosis method for deep space exploration spacecraft of the present invention.

[0028] Figure 2 Logic flowchart for injecting faults before operation;

[0029] Figure 3 This is a schematic diagram illustrating the working principle of fault diagnosis.

[0030] Figure 4 This is a comparison chart of simulation and telemetry curves. Detailed Implementation

[0031] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0032] This invention provides a model- and data-driven method for diagnosing faults in deep space exploration spacecraft, specifically including:

[0033] Step 1: Construct a four-layer fault diagnosis architecture comprising the detector system entity layer, model layer, data layer, and application layer;

[0034] Step 2: Deploy sensors in the physical layer of the detector system to collect data on the physical operational status of the spacecraft;

[0035] Step 3: In the model layer, construct a normal behavior mechanism model for predicting operating status and a fault behavior mechanism model for simulating fault data;

[0036] Step 4: In the data layer, receive and manage the on-orbit telemetry data collected from the physical layer of the detector system, and map and fuse the on-orbit telemetry data with the simulation data generated by the model layer;

[0037] Step 5: At the application layer, fault diagnosis is performed using the fused data, specifically including:

[0038] Based on the normal behavior mechanism model and fused data, the detector system is subjected to real-time status prediction and health assessment to identify system anomalies and potential risks.

[0039] When an anomaly is detected or fault analysis is required, the fault mode is injected into the fault behavior mechanism model to perform fault simulation and generate fault simulation data. The on-orbit telemetry data and the fault simulation data are compared and analyzed to locate and diagnose the fault. The handling strategies generated for the diagnosed faults are simulated, verified, and optimized in the model.

[0040] Specifically, in step 1, such as Figure 1 As shown, this invention constructs a four-layer fault diagnosis architecture comprising a detector system entity layer, a model layer, a data layer, and an application layer. This architecture achieves deep integration of digital models with physical entities, normal behavior models and fault behavior mechanism models, and virtual simulation data with real physical data through interaction and collaboration between each layer.

[0041] In step 2, the detector system entity layer mainly includes status monitoring and the detector system entity itself. During system operation, the detector system senses the external environment and makes adaptive adjustments through interfaces such as sensors and data transmission protocols. Various operating status data of the detector system, such as analog quantity data (e.g., current, voltage, temperature) and I / O data (e.g., switch signals) collected by sensors, are transmitted to the ground station through telemetry channels.

[0042] In step 3, the model layer mainly refers to the mechanism-data fusion model. The normal behavior model is built based on the physical principles of the detector system and is used to predict the operating state. The fault behavior model is an expression of the fault behavior of the detector system. It generates fault data under various fault modes through simulation, providing comprehensive data support for condition monitoring and diagnosis. The implementation of the model layer is based on the Modelica modeling language, which constructs a unified multidisciplinary model library including energy systems, thermal control systems, propulsion systems, and control systems. Specifically, it includes: model building, model calibration and verification, and fault model expansion and integration.

[0043] The model is built using a unified multi-domain modeling approach based on physical mechanisms, employing Modelica as the modeling language. Following the physical topology of the actual detector system, it progressively constructs component-level, unit-level, subsystem-level, and system-level models in a bottom-up manner. Taking the energy system as an example, specific models include: an energy environment model (simulating illumination, shadows, shading, and solar incidence angle); a solar cell array model (based on the physical characteristics of the cells, considering temperature, radiation attenuation, and shading losses); a battery pack model (based on a second-order equivalent circuit model, dynamically calculating SOC and cycle life); a shunt regulator model (simulating S3R power regulation and shunt dissipation); a charge / discharge regulator model (implementing constant current / constant voltage charging and boost discharge control for the battery); a bus filter model (for bus voltage stabilization and ripple suppression); and a load model (simulating steady-state and transient power consumption). These models are connected through clearly defined physical interfaces (such as electrical ports, thermal ports, and signal interfaces) to ensure the correct transmission of energy, signals, and data. Ultimately, they are integrated into a complete energy system simulation model, capable of accurately simulating the dynamic behavior of the system under normal operating conditions.

[0044] Model calibration and validation are crucial steps in ensuring model accuracy and confidence. Model calibration involves feeding ground-based experimental data or historical on-orbit data back into the simulation model and using parameter estimation methods such as least squares to iteratively fit and optimize key physical parameters (such as battery internal resistance, capacitance, and loss coefficient) until the model output matches the measured data within the allowable error range. Model validation further utilizes Latin hypercube sampling to systematically evaluate the impact of model input uncertainties on the output results, generating a parameter sensitivity matrix to identify key parameters that significantly affect system performance, providing a basis for model correction and uncertainty quantification. This process iterates repeatedly until a calibrated model capable of accurately reproducing the system's true physical behavior is obtained, serving as a reliable foundation for state prediction and health assessment.

[0045] Fault model extension and integration builds fault behavior mechanism models for fault simulation analysis based on calibrated normal behavior models. An object-oriented inheritance and redeclaration mechanism is adopted to establish a unified fault modeling framework. This framework defines abstract base classes for faults (including single-point fault model base classes, system fault model base classes, and fault display model base classes), encapsulating common attributes and behaviors such as fault injection points, fault detection parameters, fault type identifiers, and status visualization. Based on this framework, through an interactive interface, users can select target devices or subsystems, specify fault modes (such as partial solar panel shading, battery capacity degradation, valve jamming, etc.), and configure relevant fault parameters (such as shading rate, attenuation coefficient, jamming location). The system can automatically generate corresponding fault instance models. This fault model inherits the interface and most behaviors of the normal model, while introducing fault characteristics by redeclaring key components or modifying local parameters. This achieves seamless integration and flexible switching between the fault model and the normal model, providing a standardized and reusable model foundation for subsequent fault injection, simulation solution, and impact analysis.

[0046] In step 4, the data layer serves as a bridge between the detector system entity layer and the twin model layer, encompassing the physical data of the detector system entity, simulation data (composed of normal and fault simulation data), digital twin application data, and the fused and derived data between them. In practice, this is achieved through three sub-steps: telemetry data reception, telemetry data management, and data mapping.

[0047] The telemetry data receiving stage employs a network transmission protocol to receive external telemetry data streams from the probe system in real time. The system unpacks and parses the raw data packets according to a predefined telemetry data encapsulation format, extracting valid physical quantity values. To cope with the large-scale, high-concurrency data reception requirements of deep space exploration missions, the system adopts a multi-threaded processing architecture to handle data reception, real-time parsing, and storage management tasks separately, ensuring continuous, stable access and low-latency processing of the telemetry data stream. For example, for energy systems, it can simultaneously receive multiple parameters such as battery discharge capacity, bus voltage, and shunt status.

[0048] In the telemetry data management phase, the system provides parameterized and visualized data management functions. Users can perform fuzzy searches or precise filters on massive amounts of parameters through the telemetry parameter browser, supporting the display of data trends and real-time status in various formats such as curves, tables, and panels. The system has an embedded data quality control module that can automatically filter outliers and remove outliers according to preset rules, and cache the cleaned data in real time, providing a high-quality data source for subsequent analysis. The energy system's management panel can centrally display the telemetry parameters and their variation curves of key units such as battery banks, solar panels, and power regulation units.

[0049] Data mapping is crucial for dynamically connecting telemetry data with simulation models. The system employs an XML-based tree-structured configuration file to establish a structured mapping relationship from telemetry parameters to model variables. The configuration file defines parameter names, data types, unit conversion factors, and data validity rules. Upon system startup, the XML parser automatically loads the mapping table into memory. When telemetry data arrives, the preprocessing module quickly matches the mapping entries based on the parameter ID, completing data format conversion and physical dimension unification, such as converting raw hexadecimal values ​​to actual voltage values ​​(unit: V), and injecting them into the corresponding simulation model's input port in real time. This mechanism ensures that telemetry data can accurately and automatically drive the simulation model, supporting real-time synchronization and fusion analysis of virtual and real data.

[0050] In step 5, the application layer provides fault diagnosis and analysis applications for the detector system's operation. Based on a virtual-real interaction and collaboration mechanism that integrates mechanisms and data, the digital twin model can diagnose and handle abnormal fault states, providing an intuitive analytical tool for detector system maintenance. This is accomplished collaboratively through a state prediction and health assessment module, a fault simulation and analysis module, a fault location and diagnosis module, and a fault handling verification module.

[0051] The state prediction and health assessment module continuously predicts and assesses the health of the detector system based on a normal behavior mechanism model and real-time fused data. This module utilizes prior knowledge from the mechanism model and real-time telemetry data, employing built-in observers and filtering algorithms (such as Kalman filtering) to estimate and reconstruct key unmeasurable states of the system, achieving rolling predictions of future operating states. The health assessment function continuously compares real-time simulation results with corresponding telemetry parameters, and combines preset thresholds, association rules, and normal operating condition models in the knowledge base to quantitatively score and analyze the health status of the overall system and individual units. Once parameter deviations consistently exceed tolerances or the health score falls below a preset threshold, the module triggers an anomaly alarm, providing a clear initiation signal and initial risk location for subsequent fault analysis.

[0052] The fault simulation and analysis module starts upon receiving an abnormal alarm or proactive analysis command, performing model-based fault reproduction and impact analysis. This module first selects one or more potential fault modes from the fault mode library that match the current abnormal symptoms. Then, through... Figure 2The fault injection process, as shown, dynamically injects these fault modes and their parameters into the fault behavior mechanism model, generating a corresponding "hypothetical fault scenario model." The module supports both real-time and ultra-real-time simulation modes: real-time simulation accurately reproduces the temporal process of fault occurrence and propagation; ultra-real-time simulation is used to quickly extrapolate the long-term development trend and impact boundaries of the fault. During the simulation, the change curves of key parameters are recorded in real time and preliminarily compared with the original anomaly telemetry data, providing a simulation data foundation for fault diagnosis.

[0053] The fault location and diagnosis module is the core of fault diagnosis, and its working principle is as follows: Figure 3 As shown, this module comprehensively utilizes real-time telemetry data and the fault simulation data generated in the previous step to perform multi-dimensional and multi-level comparative analysis. Analysis methods include, but are not limited to: threshold interpretation, trend comparison, dynamic time warping (DTW) algorithm to calculate curve similarity, and using machine learning models for fault mode recognition. The module's embedded interpretation knowledge base provides a criterion system ranging from simple thresholds to complex association rules. By matching the abnormal features of telemetry data with the features of simulation data under different fault hypotheses, the system can calculate the confidence level of each hypothesis. When the matching degree between the simulation data and telemetry data of a certain fault hypothesis exceeds a preset threshold (e.g., 90%), the hypothesis is determined to be a possible cause of the fault, achieving precise localization from "abnormal phenomenon" to "root cause of fault".

[0054] The fault handling verification module is activated after a fault is diagnosed and confirmed, and is responsible for simulating, verifying, and optimizing the generated handling strategies. First, based on the fault diagnosis results, the system retrieves one or more candidate handling strategies from the contingency plan library or dynamically generates them based on expert rules (such as switching backup equipment, adjusting operating modes, or executing specific command sequences). These strategies are converted into parameterized commands that the model can recognize. Subsequently, each candidate strategy is simulated in a simulation model with the fault injected, evaluating the recovery of key system parameters (such as voltage, temperature, and attitude), the impact on mission continuity, and potential secondary risks after strategy execution. The system uses evaluation methods such as the Analytic Hierarchy Process (AHP) to comprehensively score various indicators and ultimately recommend the optimal handling solution. This module also establishes a closed-loop verification mechanism: after the actual command is executed, the telemetry data is continuously compared with the prediction results of the handling simulation. If the deviation is too large, strategy backtracking and model correction are triggered, forming a complete closed loop of "diagnosis-handling-verification-optimization," which greatly improves the safety and effectiveness of on-orbit fault handling.

[0055] Example

[0056] To verify the effectiveness of this method, a specific embodiment of the energy system of a deep space probe was used. The four-layer fusion drive fault diagnosis method described above was applied to diagnose and simulate the handling of a partial shading fault of the solar panel that occurred in an actual mission.

[0057] First, at the model level, a high-fidelity energy system mechanism model was constructed based on the aforementioned methods. This model integrates key components such as an energy environment model, a solar cell array model, and a battery pack model. The model parameters were calibrated and verified using ground-based test data to ensure their predictive accuracy under normal operating conditions. Before the simulated fault occurred, the system was in normal orbital operation, and the energy environment model was input with a periodic solar irradiance trapezoidal wave (91-minute period, 54-minute period in the illuminated area), with the solar incidence angle varying according to a preset curve.

[0058] At a certain point during the mission, an anomaly was detected in the telemetry data collected by the sensors in the physical layer of the detector system: the bus voltage experienced an unexpected fluctuation and drop, and the output power of the solar array was approximately 15% lower than the predicted value. The status prediction and health assessment module detected this deviation in real time, and the health score decreased, triggering a level two alarm, indicating a potential fault in the energy system.

[0059] At this point, the fault simulation and analysis module is activated. Based on the anomaly characteristics, the operators select "partial solar panel shading" as the primary assumed fault mode from the fault mode library. In the interactive interface, the fault parameters are set so that the shading rate (ShieldRate) jumps from 0 to 30% on a specific matrix, simulating the condition of partial solar panel shading. This fault mode is then analyzed... Figure 2 The pre-operational fault injection logic shown is dynamically loaded into the normal energy system digital twin model.

[0060] After the fault is injected, the system initiates ultra-real-time fault simulation to rapidly predict the fault's development. The simulation results are as follows: Figure 4 As shown, the simulated output power and bus voltage curves of the solar array are overlaid on the real-time telemetry curves using a curve comparison view. It can be observed that the simulated curves (usually represented by dashed lines or different colors) exhibit a downward trend and fluctuation pattern that closely matches the telemetry curves (solid lines) after the fault injection point.

[0061] The fault location and diagnosis module used the Dynamic Time Warping (DTW) algorithm to calculate the similarity between the two curves. The result showed a similarity of 92%, far exceeding the 90% matching threshold. Simultaneously, association rules in the knowledge base were triggered, confirming that the causal relationship between the output power decrease and the bus voltage fluctuation was consistent with the "partial shading" fault characteristics. Therefore, the system diagnosis conclusion was: "Approximately 30% partial shading occurred in the Xth solar array," achieving rapid and accurate fault location.

[0062] In response to the diagnosed "partial solar panel shading" fault, the fault handling verification module was activated. The system first read three candidate handling strategies from the contingency plan library: Strategy A (adjust detector attitude and optimize illumination angle), Strategy B (adjust shunt regulator operating point and redistribute power), and Strategy C (activate backup power and reduce non-critical loads).

[0063] Each strategy was converted into specific model control parameters. Subsequently, each strategy was independently simulated and derived in a twin model incorporating a "30% occlusion" fault. The derivation process rigorously simulated the timing of command upload, execution, and system response. Simulation results were output as recovery curves and comprehensive indices for key parameters (bus voltage stability, battery charge / discharge depth, and load coverage ratio).

[0064] The simulation results of the three strategies were evaluated using the Analytic Hierarchy Process (AHP). Evaluation dimensions included: safety (avoiding secondary failures), mission impact (ensuring the operational time of scientific payloads), and energy balance (maintaining a safe state of charge for the battery). Simulation and evaluation results show:

[0065] Strategy A has the best effect on restoring the stability of the bus voltage, but it has a significant impact on the attitude system and poses a risk of mission interruption.

[0066] Strategy B can be implemented quickly and can be completed within the energy system, but it has limited impact on overall power output.

[0067] Strategy C can most reliably guarantee bus voltage, but it will sacrifice some scientific observation opportunities.

[0068] After comprehensive evaluation, the system recommends Strategy B as the preferred execution plan, with Strategy C as a backup plan. Operations personnel, referring to this conclusion and considering real-time task priorities, ultimately selected Strategy B. The system generates the corresponding remote control command sequence, which, after confirmation, is uploaded to the on-orbit probe.

[0069] After the command was executed, the ground continuously received telemetry data. The recovery curves of the actual bus voltage and power parameters were compared with the simulation prediction curves of Strategy B. The actual data curves fell within the envelope of the simulation predictions, and the key parameters recovered to above the safe threshold within the expected time, verifying the effectiveness of the handling strategy. This complete process, from anomaly detection and fault diagnosis to handling verification, successfully demonstrated the rapid response and decision support capabilities of this method for typical energy system faults in actual deep space exploration missions, significantly improving the intelligence level of on-orbit fault management and mission reliability.

[0070] In a second aspect, the present invention provides an electronic device comprising: one or more processors; and a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned model and data-driven deep space exploration spacecraft fault diagnosis method.

[0071] Thirdly, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enable the processor to implement the aforementioned model- and data-driven deep space exploration spacecraft fault diagnosis method.

[0072] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A model- and data-driven method for fault diagnosis of deep space exploration spacecraft, characterized in that, include: Step 1: Construct a four-layer fault diagnosis architecture comprising the detector system entity layer, model layer, data layer, and application layer; Step 2: Deploy sensors in the physical layer of the detector system to collect data on the physical operational status of the spacecraft; Step 3: In the model layer, construct a normal behavior mechanism model for predicting the operating state and a fault behavior mechanism model for simulating fault data; Step 4: In the data layer, receive and manage the on-orbit telemetry data collected from the physical layer of the detector system, and map and fuse the on-orbit telemetry data with the simulation data generated by the model layer; Step 5: At the application layer, fault diagnosis is performed using the fused data, including: real-time status prediction and health assessment of the detector system based on the normal behavior mechanism model and the fused data. When an anomaly is detected or fault analysis is required, the fault mode is injected into the fault behavior mechanism model to perform fault simulation and achieve fault location and diagnosis.

2. The model- and data-driven fault diagnosis method for deep space exploration spacecraft according to claim 1, characterized in that, Step 3 includes: constructing a unified multidisciplinary model library based on the Modelica modeling language, which includes energy systems, thermal control systems, propulsion systems, and control systems; constructing component-level, single-machine-level, subsystem-level, and system-level models in a bottom-up manner based on the physical topology; and iteratively fitting and calibrating the model parameters using ground test data or historical on-orbit data.

3. The model- and data-driven fault diagnosis method for deep space exploration spacecraft according to claim 1, characterized in that, Step 4 specifically includes: receiving and parsing the telemetry data stream and extracting effective physical quantity values; establishing a structured mapping relationship between telemetry parameters and model variables through an XML-based mapping configuration file; converting the telemetry data into parameters that the model can recognize based on the mapping relationship, and injecting them into the simulation model to drive the simulation operation.

4. The model- and data-driven fault diagnosis method for deep space exploration spacecraft according to claim 1, characterized in that, In step 5, the real-time state prediction and health assessment includes: estimating the unmeasurable state of the system using an observer and filtering algorithm; and performing real-time comparison and health scoring of simulation results and telemetry data based on thresholds and association rules in the knowledge base.

5. The model- and data-driven fault diagnosis method for deep space exploration spacecraft according to claim 1, characterized in that, In step 5, fault location and diagnosis include: calculating the similarity between telemetry curves and simulation curves using a dynamic time warping algorithm; and matching abnormal features to determine the fault type and confidence level by combining the fault mode library and the interpretation knowledge base.

6. The model- and data-driven fault diagnosis method for deep space exploration spacecraft according to claim 1, characterized in that, Step 5 further includes fault handling verification: extracting candidate handling strategies from the contingency plan library and converting them into control parameters that the model can recognize; performing strategy deduction simulation in the fault model to evaluate safety, task impact and energy balance indicators; using the analytic hierarchy process to comprehensively score the candidate strategies and output the optimal handling solution.

7. The model- and data-driven fault diagnosis method for deep space exploration spacecraft according to claim 1, characterized in that, Step 5 also includes a closed-loop verification mechanism: after the actual command is executed, the telemetry data and simulation prediction results are continuously compared; if the deviation exceeds the threshold, the strategy backtracking and model parameter correction are triggered to optimize the handling effect.

8. The model- and data-driven fault diagnosis method for deep space exploration spacecraft according to claim 1, characterized in that, The fault behavior mechanism model is constructed using an object-oriented inheritance and redeclaration mechanism, supporting unified expansion and management of single-point fault models, system fault models, and fault display models.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When one or more programs are executed by the one or more processors, the one or more processors implement the model and data-driven deep space exploration spacecraft fault diagnosis method according to any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, It stores executable instructions that, when executed by a processor, enable the processor to implement the model- and data-driven deep space exploration spacecraft fault diagnosis method as described in any one of claims 1-8.

Images