Electric propulsion fault analysis method and system based on system simulation

Abstract

The invention relates to the technical field of spacecrafts, in particular to an electric propulsion fault analysis method and system based on system simulation, and the method comprises the steps: classifying, organizing and storing test data of an electric propulsion vacuum bin, managing the test data through a database, carrying out null cleaning on the data, filtering the data, dividing working conditions, carrying out statistical analysis, constructing features, and carrying out dimensionality reduction. The method comprises the steps of building a system simulation model, building an anomaly detection model, mining threshold rules, building a trend prediction model and building a comprehensive judgment model. According to the method, a deterministic model in a design process, a ground ignition test and a large amount of on-orbit telemetry data are utilized, real-time accurate interpretation and state evaluation of on-orbit data are realized through logic simulation operation, and a potential fault mode is effectively identified. Compared with a traditional electric propulsion on-orbit state evaluation and fault diagnosis method, the electric propulsion on-orbit state evaluation and fault diagnosis method has higher accuracy and efficiency.

Description

Technical Field

[0001] The present invention relates to the field of spacecraft technology, and in particular to an electric propulsion fault analysis method and system based on system simulation. Background Art

[0002] Electric propulsion is an advanced space propulsion technology. Compared to traditional chemical propulsion, it offers higher specific impulse, higher thrust, and longer lifespan. An increasing number of satellites are using electric propulsion to maintain their original orbital altitude and perform orbital maneuvers. However, assessing and managing the health of electric propulsion systems to ensure their reliable and stable operation for tens of thousands of hours in orbit presents a significant challenge. As a core component of satellite technology, electric propulsion systems are responsible for providing stable and precise thrust to ensure satellites maintain stable operation in their intended orbits and execute their missions. However, due to the highly complex and variable space environment and the limitations of ground-based testing and verification conditions, unexpected failures in electric propulsion systems are still possible during in-orbit operation. These failures could not only impact the normal operation of the satellite but could even lead to the failure of the entire mission.

[0003] Spacecraft are exposed to the complex space environment for extended periods, and anomalies, malfunctions, degradation, and even failures are inevitable during their service. These can impact the execution of high-standard space missions and even lead to catastrophic accidents. However, due to the unique operating environment, the inability to directly observe and the difficulty in acquiring data, their maintenance and support are a major challenge worldwide. As the power source for spacecraft orbit changes and maintenance, analyzing the normal and faulty states of electric propulsion on-orbit is a particularly pressing issue. Currently, there are two types of spacecraft fault analysis technologies: traditional methods and their automated improvements, and emerging artificial intelligence and data-driven technologies.

[0004] Improvements to existing methods include interactive telemetry data analysis tools and third-generation monitoring and diagnostic expert systems. However, these technologies still rely on experience and historical fault information at the underlying level. They simply replace manual data review and fault judgment with automatic software, and still suffer from limited efficiency and low fault detection capabilities. Summary of the Invention

[0005] The purpose of the present invention is to provide an electric propulsion fault analysis method and system based on system simulation, aiming to solve the problems of limited efficiency and low fault detection capability of traditional methods.

[0006] To achieve the above objectives, the present invention provides an electric propulsion fault analysis method based on system simulation, comprising the following steps:

[0007] The electric propulsion vacuum chamber test data is classified, organized and stored, and managed using a database;

[0008] Perform null value cleaning, data filtering, working condition division, statistical analysis, feature construction and dimensionality reduction on the data;

[0009] Build system simulation models;

[0010] Build anomaly detection models; Threshold rule mining; the threshold rule mining includes: selecting the 3sigma method or the kernel density estimation algorithm to perform threshold rule mining based on data distribution characteristics; expanding the input parameter range of the system simulation model, comparing it with the test data, and estimating the upper and lower thresholds;

[0011] Build trend prediction models;

[0012] Construct a comprehensive judgment model.

[0013] The step of constructing a system simulation model further includes:

[0014] Model-based system engineering builds the module definition diagram, state machine diagram, use case diagram, and activity diagram of the electric propulsion system, reflecting the logical relationship and parameter transfer between modules;

[0015] Use logical judgment and functions to establish a definite relationship between parameters and form a system simulation model.

[0016] The model-based system engineering constructs the module definition diagram, state machine diagram, use case diagram, and activity diagram of the electric propulsion system to reflect the logical relationship and parameter transfer between modules. The steps also include:

[0017] After establishing the top-level context module in the module definition diagram, analyze the interactions at different levels;

[0018] The state machine diagram describes the system life cycle states and the transition directions and conditions between states;

[0019] Use case diagrams are refined from functional requirements to obtain scenario use cases, which are further decomposed into included sub-scenarios and extended scenarios triggered by specific situations;

[0020] The activity diagram decomposes the use case into activity processes, and further decomposes the key steps to obtain functions at each level, and uses swimlanes to assign modules to which each behavior in the task belongs.

[0021] The step of building an anomaly detection model further includes:

[0022] Separate data samples in different states based on characteristic indicators and construct a linear or nonlinear mapping relationship between characteristic samples and states;

[0023] Combined with the system simulation model, the independent variable part data corresponding to the abnormal data is used as input, and the simulation results are compared with the abnormal data to determine whether it is abnormal and locate the faulty components and causes.

[0024] The threshold rule mining step further includes:

[0025] Select appropriate threshold rule mining method based on data distribution characteristics;

[0026] Expand the input parameter range of the system simulation model, compare it with the experimental data, and estimate the upper and lower thresholds.

[0027] Wherein, a suitable threshold rule mining method is selected according to the data distribution characteristics, and the steps also include:

[0028] The method includes a 3sigma method and a kernel density estimation algorithm.

[0029] The step of constructing a trend prediction model further includes:

[0030] Use long short-term memory neural network to predict time series and perform time series analysis on parameter values ​​that meet the characteristics of time series;

[0031] Predict the data values ​​for the next time period based on the series of the current time period to evaluate whether there is any unreasonableness in the data.

[0032] The step of constructing a comprehensive judgment model further includes:

[0033] Integrate system simulation models, anomaly detection models, threshold rule mining algorithms, and trend prediction models to form an electric propulsion fault analysis system;

[0034] Make real-time judgments on remote sensing and telemetry data, analyze faults, and output fault analysis results.

[0035] A system simulation-based electric propulsion fault analysis system is applicable to the system simulation-based electric propulsion fault analysis method, comprising a data preprocessing module, a system simulation model construction module, an anomaly detection model construction module, a threshold rule mining module, a trend prediction model construction module, and a fault analysis system integration module, wherein the system simulation model construction module is connected to the data preprocessing module, the anomaly detection model construction module is connected to the data preprocessing module, the threshold rule mining module is connected to the data preprocessing module, the trend prediction model construction module is connected to the data preprocessing module, and the fault analysis system integration module is respectively connected to the system simulation model construction module, the anomaly detection model construction module, the threshold rule mining module, and the trend prediction model construction module.

[0036] The data preprocessing module is used to classify, organize and store the vacuum chamber test data and on-orbit telemetry data of the electric propulsion system, and perform null value cleaning, data filtering and working condition division, statistical analysis, feature construction and dimensionality reduction.

[0037] The system simulation model construction module is used to construct a module definition diagram, a state machine diagram, a use case diagram and an activity diagram of the electric propulsion system through model-based system engineering to form a system simulation model.

[0038] The anomaly detection model building module is used to distinguish data samples in different states based on characteristic indicators, and in combination with the system simulation model, detect abnormal data and locate faulty components and causes.

[0039] The threshold rule mining module is used to select a suitable threshold rule mining method according to data distribution characteristics and estimate upper and lower thresholds.

[0040] The trend prediction model building module is used to use a long short-term memory neural network to perform time series prediction and evaluate whether there is any irrationality in the data.

[0041] The fault analysis system integration module is used to integrate the system simulation model, anomaly detection model, threshold rule mining algorithm and trend prediction model to form an electric propulsion fault analysis system, perform real-time judgment on remote sensing and telemetry data, and analyze faults.

[0042] The present invention provides a system simulation-based electric propulsion fault analysis method and system. By leveraging precise system simulation models and test data at various scales and dimensions, the method rationally applies statistical analysis, novel outlier detection, and pattern recognition algorithms to anomaly detection and telemetry signal trend prediction. The characteristic indicators representing the state are combined with the physical mechanisms of the electric propulsion ignition process itself, thereby exploring more explanatory threshold rules and algorithmic models. Through simulation operations with physical mechanisms combined with real data, the method achieves real-time and accurate interpretation and state assessment of on-orbit data, effectively identifying potential fault modes. This addresses the problem of weak interpretability of single data analysis techniques, uses test data to compensate for insufficient data in the initial stages of operation, and addresses the limited efficiency and low fault detection capabilities of traditional methods, thereby improving the accuracy of electric propulsion fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art.

[0044] Figure 1 4 is a step diagram of an electric propulsion fault analysis method based on system simulation according to a first embodiment of the present invention.

[0045] Figure 2 4 is a schematic structural diagram of an electric propulsion fault analysis system based on system simulation according to a second embodiment of the present invention.

[0046] In the figure: 201 - data preprocessing module, 202 - system simulation model construction module, 203 - anomaly detection model construction module, 204 - threshold rule mining module, 205 - trend prediction model construction module, 206 - fault analysis system integration module. DETAILED DESCRIPTION

[0047] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, but should not be understood as limiting the present invention.

[0048] The first embodiment of this application is:

[0049] The present invention provides an electric propulsion fault analysis method based on system simulation, comprising the following steps:

[0050] S101: Classify and organize the electric propulsion vacuum chamber test data and store them in a database for management;

[0051] S102: Perform null value cleaning, data filtering, working condition division, statistical analysis, feature construction and dimensionality reduction on the data;

[0052] S103: Build a system simulation model;

[0053] S104: Build an anomaly detection model;

[0054] S105: mining the variation patterns of threshold values ​​of key parameters of fault conditions;

[0055] S106: Build a trend prediction model;

[0056] S107: Constructing a comprehensive judgment model;

[0057] 1. First, the electric propulsion vacuum chamber test data is classified, organized and stored, and managed using a database.

[0058] 2. Data preprocessing:

[0059] 2.1 Working condition division based on status switch word: The change of working condition is marked by the jump point of the status switch word. For example, each thruster ignition process can be divided into four working conditions through the status word: cathode heating, cathode ventilation, cathode ignition, and anode ignition.

[0060] 2.2 Operating Condition Classification Based on Data Fluctuation Characteristics: Because equipment data fluctuations are affected by changes in the ignition process, operating conditions can be classified based on the data fluctuation characteristics of telemetry parameters that are relatively sensitive to operating condition changes. For example, operating condition classification can be based on the numerical fluctuation of the anode power supply output voltage.

[0061] 2.3 Statistical analysis: During the ignition phase of the electric propulsion system, there is a correlation between the voltage and current data and a certain chronological relationship with time. The voltage and current values ​​during the entire stable ignition period should meet certain statistical analysis requirements. Therefore, according to the above-mentioned properties and physical principles of the telemetry data of the electric propulsion system, corresponding statistical analysis needs to be carried out, including correlation analysis, distribution test and similarity analysis. The results of the statistical analysis serve as the basis for feature construction, threshold rule mining algorithm, anomaly detection algorithm and trend prediction algorithm selection. Combined with the characteristics of the data, the algorithm model and threshold rule information used for anomaly detection and trend prediction are mined.

[0062] 2.4 Feature Construction and Dimensionality Reduction: Extract key and effective feature parameters to characterize the state of parameters such as current and voltage. For parameters with stable fluctuations and low frequencies, construct time-domain statistical feature indicators, including mean, variance, median, effective value, kurtosis, and skewness. For signals with high-frequency and non-stationary variations, extract wavelet packet features in the time-frequency domain. Principal component analysis (PCA) and variance analysis are used to reduce the dimensionality of high-dimensional feature space samples.

[0063] 3. Build a system simulation model;

[0064] 3.1 Construct the module definition diagram of electric propulsion and each subsystem to accurately reflect the logical relationship and transferred parameters between modules.

[0065] The electric propulsion power subsystem module definition is shown in the figure below: Figure 1 shown.

[0066] 3.2 Construct use case and activity diagrams to categorize and describe the application scenarios and extended scenarios of electric propulsion. Then, decompose the use cases into activity flows, further decompose the key steps, and use swimlanes to assign components to the various behaviors within the tasks. Use functions to express the relationships between the parameters of the physical processes corresponding to these scenarios (such as activation and ignition).

[0067] 3.3 Construct a state machine and form a system simulation model. Use the state machine to describe the conversion direction and conditions between the electric propulsion system and its component states. Express these conversions using AND or NOT logic. Combined with use case diagrams and activity diagrams, a deterministic system simulation model is obtained.

[0068] 4. Build an anomaly detection model;

[0069] 4.1 Separate the data samples of different states in step 2 based on the characteristic indicators, construct a linear or nonlinear mapping relationship between the characteristic samples and the states, and use a supervised classification algorithm to obtain possible abnormal data and the corresponding electric propulsion state.

[0070] 4.2 Combined with the system simulation model established in step 3, the independent variable data corresponding to these abnormal data are used as the system simulation input, and the simulation results are compared with the abnormal data to determine whether it is abnormal. If it is abnormal, the system simulation model will locate the faulty components and the cause of the fault, and establish a data anomaly judgment logic chain to enable it to detect the abnormal status of different telemetry data.

[0071] 5. Threshold rule mining;

[0072] 5.1 Select the corresponding threshold rule mining method according to whether the data obeys Gaussian distribution or whether it is concentrated distribution. In the intermediate state, two methods are selected at the same time and the threshold is finally cross-validated. The threshold is selected according to the anomaly detection results of the corresponding samples of historical data, and the appropriate threshold is saved and used as the criterion for real-time detection.

[0073] 5.2 For data that obeys normal distribution, select the 3sigma method to obtain the upper and lower threshold limits: calculate the mean and variance of the historical data distribution, first perform a normality test on the data, and if it conforms to the normal distribution, obtain the threshold boundary based on the difference and sum of the mean and three times the variance. For data that does not obey the normality test, perform a normal transformation. According to the formula Get the upper and lower limits.

[0074] 5.3 For data that do not obey the normal distribution, select the kernel density estimation (KDE) algorithm to obtain the upper and lower limits of the threshold: calculate a density value based on the distance between the estimated point and each sample point, and take the weighted average of all density values ​​to obtain a probability density value of the estimated point in the sample distribution. Calculate the cumulative probability density from the minimum value to each test point. When it satisfies the requirement of less than (1-confidence interval (generally 0.95)) / 2, it can be used as an option for the lower threshold. When it is greater than the confidence interval, it can be used as an upper threshold. Finally, the values ​​closest to the two boundaries are used as the upper and lower thresholds respectively.

[0075] 6. Trend forecasting;

[0076] 6.1 Dataset Division: Divide the classified data in step 1 into training set data and test set data in a ratio of 3:1, and construct a supervised dataset based on the input and output step size requirements of the two parts of data to obtain the training set and test set respectively;

[0077] 6.2 Based on the long short-term memory neural network, a trend prediction model is constructed on the training set. After parameter optimization and model training, the test set is used to evaluate the model prediction effect, so as to continuously iterate and train to obtain a trend prediction model that meets the prediction accuracy.

[0078] The trend forecast process is shown in the figure.

[0079] 6.3 Compare the future data obtained by trend prediction with the abnormal judgment and threshold value obtained in steps 4 and 5 to obtain a comprehensive judgment model of the current parameters.

[0080] 7. Integrate the models and judgment logic obtained in steps 3, 4, and 5 to form an electric propulsion fault analysis system, perform real-time judgment on remote sensing and telemetry data, and analyze faults.

[0081] When using this method for analyzing electric propulsion faults based on system simulation, statistical analysis, novel outlier detection, and pattern recognition algorithms are rationally applied to anomaly detection and telemetry signal trend prediction through precise system simulation models and test data at various scales and dimensions. The characteristic indicators representing the state are combined with the physical mechanisms of the electric propulsion ignition process itself, thereby exploring more explanatory threshold rules and algorithmic models. Through simulation operations with physical mechanisms combined with real data, real-time and accurate interpretation and state assessment of on-orbit data are achieved, and potential fault modes are effectively identified. This solves the problem of weak interpretability of single data analysis techniques, uses test data to compensate for insufficient data in the initial stage of operation, and simultaneously addresses the limited efficiency and low fault detection capabilities of traditional methods, thereby improving the accuracy of electric propulsion fault diagnosis.

[0082] The second embodiment of the present application is:

[0083] Based on the first embodiment, please refer to Figure 2 , Figure 2 4 is a schematic structural diagram of an electric propulsion fault analysis system based on system simulation according to a second embodiment of the present invention.

[0084] The present invention provides an electric propulsion fault analysis system based on system simulation, which also includes a data preprocessing module 201, a system simulation model construction module 202, an anomaly detection model construction module 203, a threshold rule mining module 204, a trend prediction model construction module 205 and a fault analysis system integration module 206. The system simulation model construction module 202 is connected to the data preprocessing module 201, the anomaly detection model construction module 203 is connected to the data preprocessing module 201, the threshold rule mining module 204 is connected to the data preprocessing module 201, the trend prediction model construction module 205 is connected to the data preprocessing module 201, and the fault analysis system integration module 206 is respectively connected to the system simulation model construction module 202, the anomaly detection model construction module 203, the threshold rule mining module 204 and the trend prediction model construction module 205.

[0085] The data preprocessing module 201 is used to classify, organize and store the vacuum chamber test data and on-orbit telemetry data of the electric propulsion system, and perform null value cleaning, data filtering and working condition division, statistical analysis, feature construction and dimensionality reduction.

[0086] The system simulation model construction module 202 is used to construct a module definition diagram, a state machine diagram, a use case diagram and an activity diagram of the electric propulsion system through model-based system engineering to form a system simulation model.

[0087] The anomaly detection model building module 203 is used to distinguish data samples in different states based on characteristic indicators, and detect abnormal data and locate faulty components and causes in combination with the system simulation model.

[0088] The threshold rule mining module 204 is used to select an appropriate threshold rule mining method according to data distribution characteristics and estimate upper and lower thresholds.

[0089] The trend prediction model building module 205 is used to use a long short-term memory neural network to perform time series prediction and evaluate whether there is any irrationality in the data.

[0090] The fault analysis system integration module 206 is used to integrate the system simulation model, the anomaly detection model, the threshold rule mining algorithm and the trend prediction model to form an electric propulsion fault analysis system, perform real-time judgment on remote sensing and telemetry data, and analyze faults.

[0091] The data preprocessing mold is used to classify, organize and store the vacuum chamber test data and on-orbit telemetry data of the electric propulsion system, and perform null value cleaning, data filtering and working condition division, statistical analysis, feature construction and dimensionality reduction. Data preprocessing includes conventional null value cleaning and data filtering. At the same time, it also considers the impact of different environments and vacuum degrees on parameter values ​​such as current and voltage between vacuum chamber test and telemetry data, as well as the situation that the current will fluctuate untimely after some relays in the power management control unit (PPCU) are turned on and off, resulting in some wild values. Working condition division, data conversion and wild value elimination are performed for these special situations; statistical analysis takes into account the relative values ​​of voltage and current telemetry values. There is correlation between them and a certain order of magnitude in the time dimension. The distribution test is a hypothesis test on the distribution of parameter values, and the subsequent appropriate threshold rule mining algorithm is selected according to its distribution characteristics; feature construction and dimensionality reduction are to construct features of current and voltage values ​​with different fluctuation forms and natural frequencies according to different methods, and at the same time, the correlation and similarity between parameters are extracted in the form of a sliding window in the time dimension to characterize the state of the electric propulsion system; sample balancing takes into account that the cases where normal ignition cannot be performed or the current and voltage values ​​cannot reach the normal range after ignition are rare, so it is necessary to undersample the samples with a larger proportion or oversample the negative samples with a smaller proportion.

[0092] The system simulation model construction module 202 is used to construct the electric propulsion system's module definition diagram, state machine diagram, use case diagram, and activity diagram through model-based systems engineering to form a system simulation model. These diagrams, along with the module definition diagram, state machine diagram, use case diagram, and activity diagram, are constructed according to the model-based systems engineering approach. These diagrams form the internal logic description of the electric propulsion system: the module definition diagram establishes the top-level context module and analyzes interactions at different levels; the state machine diagram describes the system lifecycle states and the transition directions and conditions between them; the use case diagram refines functional requirements to generate scenario use cases, which are then further decomposed into sub-scenarios and extended scenarios triggered by specific scenarios. The activity diagram decomposes the use case into activity flows, further decomposing key steps to define hierarchical functions, and using swimlanes to assign modules to which each behavior in the task belongs. Based on these descriptions, logical reasoning and functions are used to establish deterministic relationships between parameters for system simulation.

[0093] The anomaly detection model construction module 203 is used to distinguish data samples in different states based on characteristic indicators, and in combination with the system simulation model, detect abnormal data and locate faulty components and causes. The main principle is to distinguish data samples in different states based on characteristic indicators, construct a linear or nonlinear mapping relationship between characteristic samples and states, and then combine the system simulation model, use time domain independent variable data as model input, compare and verify dependent variable data with model output, and establish a data anomaly judgment logic chain to detect abnormal states of different telemetry data.

[0094] The threshold rule mining module 204 is used to select an appropriate threshold rule mining method based on the data distribution characteristics, estimate the upper and lower thresholds, expand the range of the system simulation model input parameters, compare with the test data, estimate the upper and lower thresholds based on the distribution and convergence of the original current and voltage test data or the extracted eigenvalues, and work together with the anomaly detection algorithm model to complete data anomaly detection, thereby obtaining data-based state assessment anomaly detection.

[0095] The trend prediction model construction module 205 is used to use the long short-term memory neural network to perform time series prediction and evaluate whether there are any unreasonable aspects in the data. The main principle of using the long short-term memory neural network to perform time series prediction is to perform time series analysis on parameter values ​​that conform to the characteristics of the time series, predict the data values ​​of the next time period based on the sequence of the current time period, and make relevant decisions in advance based on the predicted future trend to evaluate whether there are any unreasonable aspects in the data.

[0096] The fault analysis system integration module 206 is used to integrate the system simulation model, the anomaly detection model, the threshold rule mining algorithm and the trend prediction model to form an electric propulsion fault analysis system, perform real-time judgment on remote sensing and telemetry data, and analyze faults.

[0097] The above disclosure is merely one or more preferred embodiments of the present application and is not intended to limit the scope of the present application. A person skilled in the art will understand that all or part of the processes of the above embodiments and equivalent changes made in accordance with the claims of the present application are still within the scope of the present application.

Claims

1. An electric propulsion fault analysis method based on system simulation, characterized in that: The following steps are involved: The electric propulsion vacuum chamber test data is classified, organized and stored, and managed using a database; Perform null value cleaning, data filtering, working condition division, statistical analysis, feature component analysis and dimensionality reduction on data; Build system simulation models; Build anomaly detection models; Threshold rule mining; The threshold rule mining includes: selecting the 3sigma method or the kernel density estimation algorithm for threshold mining according to the data distribution characteristics; expanding the input parameter range of the system simulation model, comparing it with the test data, and estimating the upper and lower thresholds; Build trend prediction models; Construct a comprehensive judgment model.

2. The electric propulsion fault analysis method based on system simulation according to claim 1, characterized in that: Constructing a system simulation model, the steps also include: Model-based system engineering builds the module definition diagram, state machine diagram, use case diagram, and activity diagram of the electric propulsion system, reflecting the logical relationship and parameter transfer between modules; Use logical judgment and functions to establish a definite relationship between parameters and form a system simulation model.

3. The electric propulsion fault analysis method based on system simulation according to claim 2, characterized in that: The model-based system engineering constructs a module definition diagram, state machine diagram, use case diagram, and activity diagram for the electric propulsion system, reflecting the logical relationship and parameter transfer between modules. The steps also include: After establishing the top-level context module in the module definition diagram, analyze the interactions at different levels; The state machine diagram describes the system life cycle states and the transition directions and conditions between states; The use case diagram is refined from functional requirements to obtain scenario use cases, which are further decomposed into sub-scenarios and extended scenarios triggered by specific scenarios. The extended scenarios triggered by specific scenarios include but are not limited to cathode ignition failure, anode ignition failure, and low cylinder pressure. The activity diagram decomposes the use case into activity processes, and further decomposes the key steps to obtain functions at each level, and uses swimlanes to assign modules to which each behavior in the task belongs.

4. The electric propulsion fault analysis method based on system simulation according to claim 1, characterized in that: Building an anomaly detection model, the steps also include: Separate data samples in different states based on characteristic indicators and construct a linear or nonlinear mapping relationship between characteristic samples and states; Combined with the system simulation model, the independent variable part data corresponding to the abnormal data is used as input, and the simulation results are compared with the abnormal data to determine whether it is abnormal and locate the faulty components and causes.

5. The electric propulsion fault analysis method based on system simulation according to claim 1, characterized in that: Constructing a trend prediction model, the steps also include: Use long short-term memory neural network to predict time series and perform time series analysis on parameter values ​​that meet the characteristics of time series; Predict the data values ​​for the next time period based on the series of the current time period to evaluate whether there is any unreasonableness in the data.

6. The electric propulsion fault analysis method based on system simulation according to claim 1, characterized in that: Constructing a comprehensive judgment model, the steps also include: Integrate system simulation models, anomaly detection models, threshold rule mining algorithms, and trend prediction models to form an electric propulsion fault analysis system; Make real-time judgments on remote sensing and telemetry data, analyze faults, and output fault analysis results.

7. An electric propulsion fault analysis system based on system simulation, applicable to the electric propulsion fault analysis method based on system simulation according to claim 1, characterized in that: It includes a data preprocessing module, a system simulation model construction module, an anomaly detection model construction module, a threshold rule mining module, a trend prediction model construction module and a fault analysis system integration module. The system simulation model construction module is connected to the data preprocessing module, the anomaly detection model construction module is connected to the data preprocessing module, the threshold rule mining module is connected to the data preprocessing module, the trend prediction model construction module is connected to the data preprocessing module, and the fault analysis system integration module is connected to the system simulation model construction module, the anomaly detection model construction module, the threshold rule mining module and the trend prediction model construction module respectively. The data preprocessing module is used to classify, organize and store the vacuum chamber test data and on-orbit telemetry data of the electric propulsion system, and perform null value cleaning, data filtering and working condition division, statistical analysis, feature component and dimensionality reduction; The system simulation model construction module is used to construct a module definition diagram, a state machine diagram, a use case diagram, and an activity diagram of the electric propulsion system through model-based system engineering to form a system simulation model; The anomaly detection model building module is used to distinguish data samples in different states based on characteristic indicators, and in combination with the system simulation model, detect abnormal data and locate faulty components and causes; The threshold rule mining module is used to select an appropriate threshold mining method according to data distribution characteristics and estimate upper and lower thresholds; The trend prediction model building module is used to use a long short-term memory neural network to perform time series prediction and evaluate whether there are any unreasonable aspects in the data; The fault analysis system integration module is used to integrate the system simulation model, anomaly detection model, threshold rule mining algorithm and trend prediction model to form an electric propulsion fault analysis system, perform real-time judgment on remote sensing and telemetry data, and analyze faults.

Images