A star cluster fault diagnosis system based on cloud edge-end cooperation
Patent Information
- Application Number
- CN202610919255.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-24
- Publication Date
- 2026-10-30
AI Technical Summary
现有的技术还停留在单一层面进行故障诊断，缺乏一个统一的云边端协同框架，无法在资源约束条件下实现星群故障诊断的实时性、准确性与全生命周期管理能力的统一
（1）本发明提出了一种基于“云-边-端”三层协同的诊断架构与功能解耦分配机制，将从星部署为“端”诊断单元、主星部署为“边”诊断单元、地面系统部署为“云”诊断单元，并定义了三层之间的双向数据流与诊断指令接口。解决了现有技术中各层级诊断功能割裂、信息孤岛严重、无法形成协同闭环的问题，实现了诊断对象、诊断信息与诊断功能的全覆盖与协同优化
Smart Images
Figure CN123012384A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a cluster fault diagnosis system based on cloud-edge-device collaboration, belonging to the field of cluster fault diagnosis technology. Background Technology
[0002] For master-slave constellations, fault diagnosis is mainly divided into master satellite fault diagnosis, slave satellite fault diagnosis, and ground fault diagnosis, each with different characteristics due to differences in information acquisition and computing resources. Slave satellites have limited computing and storage capabilities, therefore slave satellite fault diagnosis can only be based on single-satellite measurement data, focusing on relatively simple diagnoses of critical faults. The master satellite can interact with each slave satellite, so master satellite fault diagnosis can be based on multi-satellite data, achieving fault diagnosis through multi-information fusion. For slave satellites with relatively simple configurations, when components gradually fail and become undiagnosable, master satellite fault diagnosis becomes crucial, utilizing inter-satellite information for fault diagnosis. For ground fault diagnosis units, because the computers used are not limited by computing power and storage space, they employ richer information, more advanced methods, and more powerful functions. Current technologies still remain at a single level for fault diagnosis, lacking a unified cloud-edge-device collaborative framework, and cannot achieve unified real-time, accuracy, and full lifecycle management capabilities for constellation fault diagnosis under resource constraints. Summary of the Invention
[0003] The technical problem solved by this invention is to overcome the shortcomings of the prior art and propose a cloud-edge-device collaborative satellite cluster fault diagnosis system. Based on the satellite cluster structure, a cloud-edge-device collaborative satellite cluster fault diagnosis method is formed, which integrates front-end computing from the satellite, edge computing from the main satellite, and ground cloud computing, so as to achieve rapid and accurate diagnosis of satellite cluster faults.
[0004] The technical solution of this invention is: A constellation fault diagnosis system based on cloud-edge-device collaboration includes a "device" fault diagnosis unit, an "edge" fault diagnosis unit, and a "cloud" fault diagnosis unit. The "end" fault diagnosis unit is deployed on each slave satellite in the constellation. Each slave satellite performs preliminary fault detection and location based on on-orbit operation data, and sends the on-orbit operation data and the located fault data of the slave satellite to the "side" fault diagnosis unit through inter-satellite links. The "edge" fault diagnosis unit is deployed on the main star in the constellation. It uses inter-satellite multi-source information fusion to improve and correct the results of the "end" fault diagnosis unit. The improved and corrected results and the on-orbit operation data of each star are sent to the "cloud" fault diagnosis unit through the satellite-to-ground link. The "cloud" fault diagnosis unit is deployed on the ground system to perform fault prediction, health status analysis and assessment, and remaining lifetime prediction for each star in the constellation.
[0005] Furthermore, the "end" fault diagnosis unit includes a data preprocessing module, a feature extraction module, and a fault diagnosis module; The data preprocessing module cleans, augments, and resamples the on-orbit operating data collected by the satellite's sensors, and outputs standardized time-series data. The feature extraction module extracts data related to fault determination from the time series data and forms a feature vector; The fault diagnosis module monitors whether any component on the satellite has malfunctioned based on feature vectors. If a malfunction is detected, the module locates the faulty component and outputs the located data, including on-orbit operation data, compressed feature vectors, faulty component number, fault type, fault occurrence time, and diagnostic confidence level.
[0006] Furthermore, the "edge" fault diagnosis unit includes a multi-source data receiving and alignment module, a coordinate system transformation and information mapping module, and a multi-information fusion diagnosis module; The multi-source data receiving and alignment module receives data output from the fault diagnosis unit of each slave satellite, and simultaneously collects the satellite's on-orbit operation data and inter-satellite relative position and attitude data, and performs timestamp alignment and coordinate system normalization on the data of each slave satellite. The coordinate system transformation and information mapping module regards the sensors of neighboring satellites as remote sensors of the current satellite. Using the inter-satellite relative position and attitude information processed by the multi-source data receiving and alignment module, a multi-satellite coordinate system transformation matrix is constructed. The component measurement information of satellite n is mapped to the coordinate system of satellite m through the multi-satellite coordinate system transformation matrix, thereby realizing the unified representation of heterogeneous data. The multi-information fusion diagnostic module adopts a multi-satellite multi-information fusion method to cross-validate and correct the data output by the multi-source data receiving and alignment module. For faults that cannot be diagnosed by the information of a single satellite, the neighboring satellite data mapped by the coordinate system transformation and information mapping module is used for supplementary diagnosis, and the final single-satellite fault diagnosis result is output.
[0007] Furthermore, the "cloud" fault diagnosis unit includes a fault prediction module, a health status analysis and assessment module, and a remaining life prediction module; The fault prediction module makes predictions based on the received on-orbit operation data of each satellite, when a fault has not yet occurred or is in its early stages of development. The health status analysis and assessment module constructs a hierarchical health status assessment index system for components, individual stars, and star clusters, and conducts health status assessments layer by layer to obtain health status scores and fault prediction results for individual stars and star clusters. The remaining lifetime prediction module, based on health status scores and fault prediction results, performs probability predictions of remaining lifetime at various levels in accordance with the requirements of the constellation mission.
[0008] Furthermore, the fault prediction module utilizes statistical or deep learning methods to establish an evolutionary model from the initial stage of a fault to the point where symptoms become apparent, based on historical data and satellite measurement information. :
[0009] In the formula, For a single star i upper part j The initial amount of degradation, For a single star i upper part j The rate of degradation, It is random noise. In-orbit operating time; Calculate the mean function Describe the overall trend of the degradation process:
[0010] In the formula, Represents a single star i Initial degradation level, For a single star i Degradation rate; Kernel function The estimated model parameters describe the correlation between the degradation values and the credit limit at different time points:
[0011] In the formula, The signal variance controls the fluctuation amplitude of the degradation process; l The length scale controls the duration of temporal correlation. It is an exponential function; The three fault prediction functions obtained above are used to analyze the evolution process of faults.
[0012] Furthermore, the health status analysis and assessment module constructs a hierarchical health status assessment index system for components, individual stars, and star clusters, and conducts health status assessments layer by layer, specifically as follows: Component layer: Construct component health performance degradation functions:
[0013] In the formula, , and For prior estimation of degradation parameters, Based on the basic degradation rate, For load sensitivity coefficient, This is a comprehensive correction item for other factors; For task load intensity, It is random noise; Rate the health status of components; Based on component failure thresholds Assess whether the component has malfunctioned; Single-star layer: Based on component health status scores, combined with component failure conditions and redundancy relationships, calculate the overall health performance degradation function of a single star.
[0014] In the formula, For a single star i The number of key components For a single star i upper part j Importance weights For a single star i upper part j Redundancy factor, Assess the health status of a single star; Based on single-star failure threshold Assess whether a single satellite has malfunctioned; Cluster layer: Based on the health status score of individual stars, combined with the fault status of individual stars and the status of inter-satellite links, the overall health performance degradation function of the cluster is calculated.
[0015] In the formula, For a single star i Task criticality weight, For a single star i Inter-satellite link state factor, Assess the health status of the constellation; According to the star cluster failure threshold To assess whether the constellation has malfunctioned.
[0016] Furthermore, the health status analysis and assessment module establishes a three-tiered hierarchical degradation model at the component level, single-star level, and star cluster level to achieve probabilistic prediction of remaining lifespan at each level. The component layer uses a Gaussian process for single stars upper part Modeling the degradation trajectory:
[0017] In the formula, For a single star i upper part j The predicted degradation value, t For on-orbit operating time, For the true potential degradation process, To measure noise, This is a Gaussian process used to fit nonlinear degenerate morphology; The single-star layer represents the actual degradation process of the component layer. Combined with individual star workload and individual differences:
[0018] In the formula, Represents a single star i Initial degradation level, For a single star i Degradation rate, For task workload importance, Task load intensity; The cluster layer aggregates the degradation states of all individual stars within a cluster, forming an overall health indicator for the cluster. :
[0019] In the formula, The cluster layer intercept term is estimated by Bayesian inference; For the first Individual star weight coefficient; For the stellar cluster level random effects term, n This represents the total number of individual stars within the star cluster.
[0020] Furthermore, using historical data, measurement information acquired by onboard sensors, and fault diagnosis data, a three-tiered hierarchical degradation model at the component level, single-satellite level, and constellation level is trained. Based on the trained three-tiered hierarchical degradation model at the component level, single-satellite level, and constellation level, the remaining lifetime is predicted, specifically as follows: Component layer remaining lifetime probability distribution predict:
[0021] In the formula, Indicates the infimum, Indicates the predicted time. The component failure threshold. Components at time t health status, Leading terms for predicting future degradation; Single-plane remaining lifetime probability distribution predict:
[0022] In the formula, The single-satellite fault threshold is output by the health status analysis and assessment module. Probability distribution of remaining lifetime in the star cluster layer predict:
[0023] In the formula, This is the cluster fault threshold, output by the health status analysis and assessment module.
[0024] Furthermore, the probability distribution of the remaining lifetime of the star cluster layer... The Star Cluster mission requires remaining lifespan. In comparison, if If it meets the requirements for performing the task, then it is considered to satisfy the requirements; if Then, feedback will be provided on tasks at the constellation level, including task reallocation, redundancy switching, and degraded operation.
[0025] Furthermore, regarding the first Each single-satellite task solves for the remaining lifetime requirement of a single satellite. To minimize the value of the following expression:
[0026] Feedback on single-satellite mission adjustment suggestions based on the remaining lifespan requirements of each single satellite.
[0027] The advantages of this invention compared to the prior art are: (1) This invention proposes a diagnostic architecture and functional decoupling allocation mechanism based on a three-layer collaboration of "cloud-edge-device". The satellite is deployed as a "device" diagnostic unit, the main satellite as an "edge" diagnostic unit, and the ground system as a "cloud" diagnostic unit. A bidirectional data flow and diagnostic command interface are defined between the three layers. This solves the problems of fragmented diagnostic functions at each level, severe information silos, and the inability to form a collaborative closed loop in existing technologies, achieving full coverage and collaborative optimization of diagnostic objects, diagnostic information, and diagnostic functions. (2) This invention establishes a hierarchical Bayesian degradation model and a mission constraint backpropagation closed-loop mechanism for lifetime prediction, from single stars to star clusters. At the "end" unit, health features of single stars are extracted; at the "edge" unit, an inter-satellite fault diagnosis system is constructed; and at the "cloud" unit, a degradation model is trained using big data. The backpropagation mechanism then decomposes the remaining lifetime prediction results layer by layer into mission constraint suggestions for components, single stars, and star clusters. This solves the problems in existing technologies where lifetime predictions are independent, cannot be correlated across levels, and are difficult to support star cluster mission decisions. It enables real-time guidance of on-orbit diagnostic strategies based on lifetime prediction results, and quantitative support for star cluster mission planning based on remaining lifetime. Attached Figure Description
[0028] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1This is a schematic diagram of a constellation fault diagnosis system based on cloud-edge-device collaboration, according to an embodiment of the present invention. Figure 2 This is a data interaction diagram between modules of the "end" fault diagnosis unit in an embodiment of the present invention; Figure 3 This is a data interaction diagram between the "edge" fault diagnosis unit modules in an embodiment of the present invention; Figure 4 This is a data interaction diagram between modules of the "cloud" fault diagnosis unit in an embodiment of the present invention. Detailed Implementation
[0029] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
[0030] This invention proposes a constellation fault diagnosis system based on cloud-edge-device collaboration, such as... Figure 1 As shown, it includes a "terminal" fault diagnosis unit, an "edge" fault diagnosis unit, and a "cloud" fault diagnosis unit.
[0031] I. Terminal Fault Diagnosis Unit The "end" fault diagnosis unit is deployed on each slave satellite in the constellation to achieve preliminary fault detection and location based on single-satellite operation data under conditions of low information redundancy and limited resources.
[0032] (1) Module composition and function The "end" fault diagnosis unit consists of the following three modules: Data preprocessing module: To address issues such as inconsistent periods, incomplete information, and outliers in satellite on-orbit data, the module employs methods such as cleaning, dimension enhancement, and resampling to improve data quality and output standardized time series data.
[0033] Feature extraction module: Based on the preprocessed data, two types of features are extracted: single-parameter features, which extract time-domain and frequency-domain features with physical meaning; and multi-parameter features, which use methods such as principal component analysis (PCA) or autoencoder to extract implicit correlation features between multi-source data.
[0034] Fault diagnosis module: Based on extracted data features, it first monitors whether each component has malfunctioned through threshold testing or distance calculation (such as Euclidean distance or Mahalanobis distance). Once a fault is detected, it identifies the faulty component by employing analytical model methods (such as observers) or knowledge / data-driven methods (such as rule matching or lightweight neural networks) for components of the same or different types. Under the premise of meeting the constraints of onboard computer resources, a relatively simple intelligent fault diagnosis method can be selected.
[0035] (2) Final output information Measurement information, compressed feature vector, faulty component number, fault type, fault occurrence time, and diagnostic confidence level.
[0036] (3) Data interaction relationship Data flow between modules, such as Figure 2 As shown: Raw running data → Data preprocessing module → Standardized data → Feature extraction module → Feature vector → Fault diagnosis module → Diagnosis results.
[0037] Interaction with "edge" units: Send diagnostic data (feature vectors and measurement information) and diagnostic results to "edge" units.
[0038] II. "Side" Fault Diagnosis Unit The "edge" fault diagnosis unit is deployed on the main star in the constellation. It is used to improve and correct the fault diagnosis results of the "end" by using the fusion of multi-source information between stars, and solve the problems of poor diagnosability or inaccurate positioning caused by insufficient information from a single star.
[0039] (1) Module composition and function The "edge" fault diagnosis unit consists of the following three modules: Multi-source data receiving and alignment module: Receives measurement information, diagnostic results and feature vectors output from each slave satellite "end" unit, and simultaneously collects the satellite's (master satellite's) operational data and information such as relative position and attitude between satellites, and performs timestamp alignment and coordinate system normalization on the data of each satellite.
[0040] Coordinate system transformation and information mapping module: Treat the sensors of neighboring satellites as remote sensors of the current satellite, and use the inter-satellite relative position and attitude information to construct a multi-satellite coordinate system transformation matrix to map the component measurement information of satellite n to the coordinate system of satellite m, thereby realizing a unified representation of heterogeneous data.
[0041] Multi-information fusion diagnostic module: Employs multi-satellite multi-information fusion methods (such as majority voting, Bayesian fusion, or DS evidence theory) to cross-validate and correct the "end" diagnostic results. For faults that cannot be diagnosed by the information of a single satellite, supplementary diagnosis is performed using mapped neighboring satellite data, outputting the final single-satellite fault diagnosis result.
[0042] (2) Final output information Final fault diagnosis results for each satellite and the satellite itself (faulty component, type, time); correction records (original diagnosis results, correction basis, and corrected results); intermediate data (multi-satellite alignment data, coordinate transformation parameters, fusion evidence, and link quality information, etc.); explanation of improved diagnostic confidence.
[0043] (3) Data interaction relationship Data flow between modules, such as Figure 3 As shown: Diagnostic results of each satellite + local satellite data + inter-satellite measurement information → data alignment module → aligned data → coordinate transformation module → unified coordinate system data → fusion diagnostic module → corrected diagnostic results.
[0044] Interaction with the "end" unit: Receive diagnostic results and feature vectors from each slave satellite via inter-satellite links.
[0045] Interaction with the "cloud" unit: Package the fused diagnostic results and intermediate data and send them to the ground "cloud" unit via the satellite-to-ground link.
[0046] III. Cloud Fault Diagnosis Unit The "cloud" fault diagnosis unit is deployed on the ground system and utilizes powerful computing and storage resources to achieve big data-driven fault prediction, health status assessment, remaining life prediction, and intelligent diagnostic model training and updating.
[0047] (1) Module composition and function The cloud-based fault diagnosis unit consists of the following four modules: Fault prediction module: Predicts faults before they occur or in their early stages. Based on historical data and edge / end measurement information, it uses statistical methods (such as ARIMA) or deep learning methods (such as LSTM, Transformer) to build an evolutionary model from the initial stage of a fault to significant symptoms. :
[0048] In the formula, This is the initial degradation amount. For degradation rate, It is random noise. In-orbit operating time; Calculate the mean function Describe the overall trend of the degradation process:
[0049] In the formula, Represents the initial level of degradation. The degradation rate; Kernel function The estimated model parameters describe the correlation between the degradation values and the credit limit at different time points:
[0050] In the formula, The signal variance controls the fluctuation amplitude of the degradation process; l The length scale controls the duration of temporal correlation. It is an exponential function; The three fault prediction functions obtained above are used for subsequent analysis of the fault evolution process and to predict abnormal state changes in advance.
[0051] Health Status Analysis and Assessment Module: A hierarchical health status assessment indicator system is constructed, consisting of "components → individual stars → constellations." Among these: Component layer: Construct a component health performance degradation function using physical mechanism analysis, statistical methods, or machine learning methods. The value ranges from 0 to 100:
[0052] In the formula, , and For prior estimation of degradation parameters, The basic degradation rate represents the inherent degradation rate that is independent of time. The load sensitivity coefficient represents the degree to which task load affects degradation. This is a comprehensive correction item for other factors. The task load intensity is obtained from measurement information. This is random noise. Set a component failure threshold. (usually based on) (Set the size to 60), output component health status score.
[0053] Single-satellite layer: Based on component-level health scores, combined with component failure conditions and redundancy relationships, calculate the overall health performance degradation function of a single satellite. The value ranges from 0 to 100:
[0054] In the formula, For a single star i The number of key components For components j Importance weights For components j Redundancy factor.
[0055] Set single-satellite fault threshold (usually based on) (Set the size to 60), output the health status score of a single star.
[0056] Cluster layer: Based on the health score of a single satellite, combined with the fault status of a single satellite and the status of inter-satellite links, calculate the overall health performance degradation function of the cluster. The value ranges from 0 to 100.
[0057] In the formula, The mission criticality weight of a single star. For a single star i Inter-satellite link state factor.
[0058] Set cluster failure threshold (usually based on) (Set the size to 60), output the constellation health status score.
[0059] Remaining lifetime prediction module: Based on health status assessment and failure prediction results, and addressing the requirements of constellation missions, a remaining lifetime prediction method based on a hierarchical Bayesian degradation model and backpropagation of mission constraints is proposed. This method establishes a three-layer hierarchical degradation model—component layer → single-star layer → constellation layer—to achieve probabilistic prediction of remaining lifetime at each level. Furthermore, the backpropagation mechanism decomposes the mission requirements at the constellation layer into specific recommendations for the constellation and individual stars. The specific steps are as follows: (i) Construction of hierarchical Bayesian degeneracy model Establish a three-level hierarchical degradation model consisting of component layer, single star layer, and star cluster layer.
[0060] The component layer is processed using a Gaussian process. i The first single star j Modeling the degradation trajectory of each component:
[0061] In the formula, This is the predicted value for component degradation. t For on-orbit operating time, For the true potential degradation process, To measure noise, It is a Gaussian process used to fit nonlinear degenerate morphology.
[0062] The single-star layer represents the actual degradation process of the component layer. Combined with individual star workload and individual differences:
[0063] In the formula, For task workload importance, This refers to the workload intensity of the task.
[0064] The cluster layer aggregates all individual stars within the cluster ( n The degradation state of individual stars forms the overall health indicator of the star cluster. :
[0065] In the formula, The cluster layer intercept term is estimated by Bayesian inference; This is the weighting coefficient for a single star, which is preset according to the importance of the single star. The stochastic effect term for the star cluster layer is estimated from historical data and follows a normal distribution.
[0066] (ii) Probability prediction of remaining lifespan Based on the trained hierarchical Bayesian degradation model, the remaining lifetime is predicted for each of the three levels.
[0067] Component layer remaining lifetime probability distribution predict:
[0068] In the formula, Indicates the infimum; Indicates the predicted time; This is the component failure threshold, output by the health assessment module; For components at time t ; health status; The term used for predicting future degradation comes from the fault prediction module and serves as a guide for the mean function during Gaussian process prediction.
[0069] Single-plane remaining lifetime probability distribution predict:
[0070] In the formula, This is the single-satellite fault threshold, output by the health assessment module.
[0071] Probability distribution of remaining lifetime in the star cluster layer predict:
[0072] In the formula, This is the cluster failure threshold, output by the health assessment module.
[0073] (iii) Backpropagation of task suggestions The mission requirements of the constellation layer are broken down layer by layer, and specific recommendations for individual stars and components are generated based on lifetime predictions.
[0074] First, the probability distribution of the remaining lifetime of the star cluster is... The Star Cluster mission requires remaining lifespan. In comparison, if If it meets the requirements for performing the task, then it is considered to satisfy the requirements; if Then, feedback will be provided regarding tasks at the constellation level, including task reallocation, redundancy switching, and degraded operation.
[0075] Solving for the remaining lifetime requirement of a single satellite for a single-satellite mission. To minimize the value of the following expression:
[0076] Feedback should be provided on adjusting the tasks assigned to a single satellite based on its remaining lifespan requirements. For example, if the predicted remaining lifespan of a single satellite is low, the task allocation should be reduced accordingly.
[0077] Fault diagnosis model training and update module: For different fault diagnosis tasks on the satellite, select the corresponding measurement information to generate training data; for different intelligent methods (such as neural networks, decision trees, support vector machines), set the network type, number of layers, parameters, etc., and train the intelligent fault diagnosis model offline; support irregular updates of the models deployed in the "end" and "edge" units (through satellite-to-ground injection).
[0078] (2) Final output information Fault prediction report (predicted fault type, probability of occurrence, and expected time window), health status assessment curve, remaining life prediction value, task allocation suggestions, and updated intelligent diagnostic model parameters.
[0079] (3) Data interaction relationship Data flow between modules, such as Figure 4 As shown: Historical data + "edge" / "end" measurement information → Fault prediction module → Fault prediction result; "edge" / "end" fault diagnosis result → Health assessment module → Health status result; Fault prediction result + Health status result → Remaining lifetime prediction module → Lifetime prediction result → Task allocation suggestion; "edge" / "end" measurement information → Full training data → Model training module → Intelligent diagnostic model → Uploaded to "end" / "edge" unit.
[0080] Interaction with "edge" units: Receive diagnostic results and measurement information uploaded by "edge" units; distribute updated intelligent diagnostic models.
[0081] Interaction with "End" Units: Update the intelligent diagnostic model of "End" units through relaying from "Edge" units or direct injection.
[0082] The embodiments described above are merely preferred embodiments of the present invention. Ordinary variations and substitutions made by those skilled in the art within the scope of the technical solution of the present invention should be included within the protection scope of the present invention.
Claims
1. A constellation fault diagnosis system based on cloud-edge-device collaboration, characterized in that, It includes "end" fault diagnosis unit, "edge" fault diagnosis unit and "cloud" fault diagnosis unit; The "end" fault diagnosis unit is deployed on each slave satellite in the constellation. Each slave satellite performs preliminary fault detection and location based on on-orbit operation data, and sends the on-orbit operation data and the located fault data of the slave satellite to the "side" fault diagnosis unit through the inter-satellite link. The "edge" fault diagnosis unit is deployed on the main star in the constellation. It uses inter-satellite multi-source information fusion to improve and correct the results of the "end" fault diagnosis unit. The improved and corrected results and the on-orbit operation data of each star are sent to the "cloud" fault diagnosis unit through the satellite-to-ground link. The "cloud" fault diagnosis unit is deployed on the ground system to perform fault prediction, health status analysis and assessment, and remaining lifetime prediction for each star in the constellation.
2. The constellation fault diagnosis system based on cloud-edge-device collaboration according to claim 1, characterized in that, The "end" fault diagnosis unit includes a data preprocessing module, a feature extraction module, and a fault diagnosis module; The data preprocessing module cleans, augments, and resamples the on-orbit operating data collected by the satellite's sensors, and outputs standardized time-series data. The feature extraction module extracts data related to fault determination from the time series data and forms a feature vector; The fault diagnosis module monitors whether any component on the satellite has malfunctioned based on feature vectors. If a malfunction is detected, the module locates the faulty component and outputs the located data, including on-orbit operation data, compressed feature vectors, faulty component number, fault type, fault occurrence time, and diagnostic confidence level.
3. The constellation fault diagnosis system based on cloud-edge-device collaboration according to claim 1, characterized in that, The "edge" fault diagnosis unit includes a multi-source data receiving and alignment module, a coordinate system transformation and information mapping module, and a multi-information fusion diagnosis module. The multi-source data receiving and alignment module receives data output from the fault diagnosis unit of each slave satellite, and simultaneously collects the on-orbit operation data of the satellite and the relative position and attitude data between satellites, and performs timestamp alignment and coordinate system normalization on the data of each slave satellite. The coordinate system transformation and information mapping module regards the sensors of neighboring satellites as remote sensors of the current satellite. Using the inter-satellite relative position and attitude information processed by the multi-source data receiving and alignment module, a multi-satellite coordinate system transformation matrix is constructed. The component measurement information of satellite n is mapped to the coordinate system of satellite m through the multi-satellite coordinate system transformation matrix, thereby realizing the unified representation of heterogeneous data. The multi-information fusion diagnostic module adopts a multi-satellite multi-information fusion method to cross-validate and correct the data output by the multi-source data receiving and alignment module. For faults that cannot be diagnosed by the information of a single satellite, the neighboring satellite data mapped by the coordinate system transformation and information mapping module is used for supplementary diagnosis, and the final single-satellite fault diagnosis result is output.
4. A constellation fault diagnosis system based on cloud-edge-device collaboration according to claim 1, characterized in that, The "cloud" fault diagnosis unit includes a fault prediction module, a health status analysis and assessment module, and a remaining life prediction module. The fault prediction module makes predictions based on the received on-orbit operation data of each satellite, when a fault has not yet occurred or is in its early stages of development. The health status analysis and assessment module constructs a hierarchical health status assessment index system for components, individual stars, and star clusters, and conducts health status assessments layer by layer to obtain health status scores and fault prediction results for individual stars and star clusters. The remaining lifetime prediction module, based on health status scores and fault prediction results, performs probability predictions of remaining lifetime at various levels in accordance with the requirements of the constellation mission.
5. A constellation fault diagnosis system based on cloud-edge-device collaboration according to claim 4, characterized in that, The fault prediction module uses statistical or deep learning methods to establish an evolutionary model from the initial stage of a fault to significant symptoms, based on historical data and satellite measurement information. : In the formula, For a single star i upper part j The initial amount of degradation, For a single star i upper part j The rate of degradation, It is random noise. In-orbit operating time; Calculate the mean function Describe the overall trend of the degradation process: In the formula, Represents a single star i Initial degradation level, For a single star i Degradation rate; Kernel function The estimated model parameters describe the correlation between the degradation values and the credit limit at different time points: In the formula, The signal variance controls the fluctuation amplitude of the degradation process; l The length scale controls the duration of temporal correlation. It is an exponential function; The three fault prediction functions obtained above are used to analyze the evolution process of faults.
6. A constellation fault diagnosis system based on cloud-edge-device collaboration according to claim 4, characterized in that, The health status analysis and assessment module constructs a hierarchical health status assessment index system for components, individual stars, and star clusters, and conducts health status assessments layer by layer, specifically as follows: Component layer: Construct component health performance degradation functions: In the formula, , and For prior estimation of degradation parameters, Based on the basic degradation rate, For load sensitivity coefficient, This is a comprehensive correction item for other factors; For task load intensity, It is random noise; Rate the health status of components; Based on component failure thresholds Assess whether the component has malfunctioned; Single-star layer: Based on component health status scores, combined with component failure conditions and redundancy relationships, calculate the overall health performance degradation function of a single star. In the formula, For a single star i The number of key components For a single star i upper part j Importance weights For a single star i upper part j Redundancy factor, Assess the health status of a single star; Based on single-star failure threshold Assess whether a single satellite has malfunctioned; Cluster layer: Based on the health status score of individual stars, combined with the fault status of individual stars and the status of inter-satellite links, the overall health performance degradation function of the cluster is calculated. In the formula, For a single star i Task criticality weight, For a single star i Inter-satellite link state factor, Assess the health status of the constellation; According to the star cluster failure threshold To assess whether the constellation has malfunctioned.
7. A constellation fault diagnosis system based on cloud-edge-device collaboration according to claim 4, characterized in that, The health status analysis and assessment module establishes a three-tiered hierarchical degradation model at the component level, single-star level, and star cluster level to predict the probability of remaining lifespan at each level. The component layer uses a Gaussian process for single stars upper part Modeling the degradation trajectory: In the formula, For a single star i upper part j The predicted degradation value, t For on-orbit operating time, For the true potential degradation process, To measure noise, This is a Gaussian process used to fit nonlinear degenerate morphology; The single-star layer represents the actual degradation process of the component layer. Combined with individual star workload and individual differences: In the formula, Represents a single star i Initial degradation level, For a single star i Degradation rate, For task workload importance, Task load intensity; The cluster layer aggregates the degradation states of all individual stars within a cluster, forming an overall health indicator for the cluster. : In the formula, The cluster layer intercept term is estimated by Bayesian inference; For the first Individual star weight coefficient; For the stellar cluster level random effects term, n This represents the total number of individual stars within the star cluster.
8. A constellation fault diagnosis system based on cloud-edge-device collaboration according to claim 7, characterized in that, Using historical data, measurement information acquired by onboard sensors, and fault diagnosis data, a three-tiered hierarchical degradation model at the component level, single-satellite level, and constellation level is trained. Based on the trained three-tiered hierarchical degradation model, remaining lifetime is predicted, specifically as follows: Component layer remaining lifetime probability distribution predict: In the formula, Indicates the infimum, Indicates the predicted time. The component failure threshold. Components at time t health status, Leading terms for predicting future degradation; Single-plane remaining lifetime probability distribution predict: In the formula, The single-satellite fault threshold is output by the health status analysis and assessment module. Probability distribution of remaining lifetime in the star cluster layer predict: In the formula, This is the cluster fault threshold, output by the health status analysis and assessment module.
9. A constellation fault diagnosis system based on cloud-edge-device collaboration according to claim 8, characterized in that, Probability distribution of remaining lifetime of star cluster layer The Star Cluster mission requires remaining lifespan. In comparison, if If it meets the requirements for performing the task, then it is considered to satisfy the requirements; if Then, feedback will be provided on tasks at the constellation level, including task reallocation, redundancy switching, and degraded operation.
10. A constellation fault diagnosis system based on cloud-edge-device collaboration according to claim 8, characterized in that, Regarding the first Each single-satellite task solves for the remaining lifetime requirement of a single satellite. To minimize the value of the following expression: Feedback on single-satellite mission adjustment suggestions based on the remaining lifespan requirements of each single satellite.