Fault diagnosis method for reaction flywheel

Abstract

The invention provides a fault diagnosis method for a reaction flywheel, and the method comprises the following steps: collecting the operation data of the reaction flywheel, and carrying out the preprocessing of the operation data; obtaining possible faults and fault occurrence probability of the reaction flywheel according to the preprocessed operation data; constructing a fault diagnosis path of the reaction flywheel; processing the fault diagnosis paths based on possible faults and fault occurrence probabilities by adopting a grey wolf optimization algorithm to obtain a sequence of the fault diagnosis paths; performing fault diagnosis based on the sorting result; according to the method, after fault diagnosis paths are formed through random composition of parts, used for fault diagnosis, of the reaction flywheel, the grey wolf optimization algorithm is adopted to compare and sort the fault diagnosis paths, and when fault diagnosis is conducted according to the sorting result, the fault diagnosis efficiency can be greatly improved, the fault position can be rapidly positioned, and the fault diagnosis efficiency is improved. Therefore, repair and maintenance can be carried out in time, and normal operation of the spacecraft is prevented from being affected.

Description

Technical Field

[0001] The invention relates to the field of aerospace technology, and in particular to a fault diagnosis method for a reaction flywheel. Background Art

[0002] In the high-tech field of aerospace technology, the reaction flywheel is undoubtedly an extremely critical component in satellite attitude control and orbit adjustment. It cleverly adjusts its own angular momentum and accurately generates reaction torque, thereby achieving fine adjustment of the satellite's attitude and flexible orbital maneuverability. It is the core guarantee for the satellite to operate stably in space and successfully perform various tasks. Its stability and reliability directly determine the ultimate success or failure of the satellite mission.

[0003] However, the space environment is full of complex and changeable harsh factors. If the reaction flywheel operates in such an extreme environment for a long time, it will inevitably suffer from problems such as wear and aging. Various faults also occur from time to time during its operation. For example, communication abnormalities will lead to obstruction of information transmission between the ground control center and the flywheel, and failure to receive and send key instructions in time; there is also the dangerous situation of out-of-control speed, which may cause the satellite attitude to lose balance instantly and fall into chaos. If these faults are not detected and properly handled in time, the safety of the satellite's operation will be greatly threatened, and the execution of its mission will be out of the question, and all efforts may be wasted.

[0004] The traditional method of diagnosing reaction flywheel faults mainly relies on ground telemetry and remote control technology and manual analysis. Ground control personnel receive limited data sent back by satellites and make analysis and judgments based on their experience and professional knowledge. However, this method has obvious disadvantages. Its response speed is often slow, and it is unable to reflect the actual operating status of the flywheel in real time and accurately. Especially in extremely challenging mission scenarios such as deep space exploration, due to the communication delays caused by long distances and the constraints of factors such as limited data transmission, traditional fault diagnosis methods are even more stretched and unable to meet mission requirements. They are gradually unable to keep up with the rapid development of aerospace technology.

[0005] In recent years, with the rapid rise of artificial intelligence technology, new hope has been brought to reaction flywheel fault diagnosis. By installing various advanced sensors on the reaction flywheel, massive real-time data can be collected. After in-depth mining and analysis by artificial intelligence algorithms, it can be preliminarily confirmed whether the reaction flywheel has hidden fault hazards. However, due to the large number of components and complex structure of the reaction flywheel itself, fault diagnosis can only be carried out randomly, and a clear and orderly diagnostic path has not yet been formed. This has led to a significant increase in diagnostic time, forcing the reaction flywheel to run difficultly in a faulty state for a long time. As a result, its service life will inevitably be greatly reduced. At the same time, it will also have a serious negative impact on the normal operation of the entire spacecraft, which may trigger a series of chain reactions and threaten the safety and smooth progress of the entire space mission. Summary of the invention

[0006] In view of this, the present invention proposes a fault diagnosis method for a reaction flywheel, which can select a fault diagnosis path to improve the efficiency of fault diagnosis.

[0007] The technical solution of the present invention is achieved in this way:

[0008] A method for fault diagnosis of a reaction flywheel comprises the following steps:

[0009] Step S1, collecting the operating data of the reaction flywheel and preprocessing the operating data;

[0010] Step S2, obtaining possible faults of the reaction flywheel and the probability of the fault occurring according to the preprocessed operating data;

[0011] Step S3, constructing a fault diagnosis path of the reaction flywheel;

[0012] Step S4: using the Grey Wolf Optimization Algorithm to process the fault diagnosis paths based on possible faults and fault occurrence probabilities, and obtain the ranking of the fault diagnosis paths;

[0013] Step S5: perform fault diagnosis based on the sorting result.

[0014] Preferably, the specific steps of step S1 include:

[0015] Step S11, collecting real-time speed, motor current, bearing temperature and vibration data of the reaction flywheel;

[0016] Step S12: performing data cleaning, denoising and feature extraction on the collected real-time speed, motor current, bearing temperature and vibration data.

[0017] Preferably, the specific steps of step S2 include:

[0018] Step S21, constructing a neural network model and performing training to map the normal range of the reaction flywheel operation data into the neural network model;

[0019] Step S22, inputting the operation data into the neural network model, and the neural network model processes the operation data;

[0020] Step S23: The neural network model outputs possible faults of the reaction flywheel and the probability of the fault occurring.

[0021] Preferably, the specific steps of constructing the neural network model and training it in step S21 are: obtaining historical data of the reaction flywheel, wherein the historical data includes historical operation data and fault records of the reaction flywheel, taking the historical operation data and fault records as a group of data, and dividing them into a training set and a test set in a ratio of 7:3, training the neural network with the training set, and after the training is completed, testing it with the test set, and obtaining a trained neural network model after the test accuracy reaches a set threshold.

[0022] Preferably, the specific steps of step S3 are:

[0023] Step S31, obtaining several diagnostic components of the reaction flywheel for fault diagnosis;

[0024] Step S32, constructing a virtual sphere, and evenly mapping the diagnostic components onto the surface of the virtual sphere;

[0025] Step S33: starting from the center of the virtual sphere, randomly selecting at least one diagnostic component on the surface of the virtual sphere and connecting them in sequence to form a plurality of fault diagnosis paths.

[0026] Preferably, the step S3 further comprises:

[0027] Step S34: according to the possible fault of the reaction flywheel, remove the irrelevant parts in the fault diagnosis path.

[0028] Preferably, the specific steps of step S4 include:

[0029] Step S41, selecting possible faults based on the probability of fault occurrence and the severity of the fault;

[0030] Step S42, initializing the gray wolf population and setting the number of iterations;

[0031] Step S43: randomly select a fault diagnosis path and calculate the fitness value based on possible faults;

[0032] Step S44, selecting another fault diagnosis path, calculating the fitness value, comparing it with the previous calculation result, and obtaining a sorting result after sorting according to the comparison result.

[0033] Preferably, the specific steps of step S41 include:

[0034] Step S411, assessing the severity of the impact of possible faults and quantifying them;

[0035] Step S412: assign values ​​to the fault occurrence probability and the quantified impact degree respectively, and sum them up;

[0036] Step S413: select possible faults from large to small according to the summation result.

[0037] Preferably, the number of iterations of step S42 is the number of fault diagnosis paths.

[0038] Preferably, the specific steps of step S5 are: selecting a fault diagnosis path from large to small according to the sorting result to perform fault diagnosis.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] 1. The present invention provides a fault diagnosis method for a reaction flywheel. After collecting the operation data of the reaction flywheel, the possible faults of the reaction flywheel and the probability of the faults occurring can be evaluated according to the operation data. Then, after constructing a complete fault diagnosis path, the fault diagnosis path is processed by using a gray wolf optimization algorithm. The processing is based on the possible faults of the reaction flywheel, so that the sorting results of multiple fault diagnosis paths for diagnosing possible faults can be obtained. The largest fault diagnosis path in the sorting results is the path that is most likely to diagnose the fault, so that the time for fault diagnosis can be greatly reduced and the efficiency of fault diagnosis can be improved, so as to quickly locate the fault point of the reaction flywheel, so that timely repair and maintenance can be carried out to ensure the normal operation of the spacecraft.

[0041] 2. When constructing the fault diagnosis path of the reaction flywheel, a virtual sphere is first constructed, and then several diagnostic components used for fault diagnosis are mapped to the surface of the virtual sphere. Starting from the center of the virtual sphere, at least one diagnostic component on its surface is connected in sequence to form several fault diagnosis paths. The virtual sphere is set up to ensure that the distance to the first diagnostic component reached after starting from the center is consistent, and the formed fault diagnosis path can be displayed intuitively, which is convenient for subsequent query of the constructed fault diagnosis path for data analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only preferred embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0043] Figure 1 is a flow chart of a fault diagnosis method for a reaction flywheel of the present invention;

[0044] Figure 2 is a flow chart of step S1 of a fault diagnosis method for a reaction flywheel of the present invention;

[0045] Figure 3 is a flow chart of step S2 of a fault diagnosis method for a reaction flywheel of the present invention;

[0046] Figure 4 is a flow chart of step S3 of a fault diagnosis method for a reaction flywheel of the present invention;

[0047] Figure 5 is a flow chart of step S4 of a fault diagnosis method for a reaction flywheel of the present invention;

[0048] Figure 6 is a flow chart of step S41 of a fault diagnosis method for a reaction flywheel of the present invention;

[0049] Figure 7 A schematic diagram of a virtual sphere for a fault diagnosis method for a reaction flywheel according to the present invention. DETAILED DESCRIPTION

[0050] In order to better understand the technical content of the present invention, a specific embodiment is provided below, and the present invention is further described in conjunction with the accompanying drawings.

[0051] See also Figures 1 to 7 The present invention provides a method for diagnosing a fault of a reaction flywheel, comprising the following steps:

[0052] Step S1, collecting the operating data of the reaction flywheel and preprocessing the operating data;

[0053] The specific steps include:

[0054] Step S11, collecting real-time speed, motor current, bearing temperature and vibration data of the reaction flywheel;

[0055] Step S12: performing data cleaning, denoising and feature extraction on the collected real-time speed, motor current, bearing temperature and vibration data.

[0056] The reaction flywheel is an important part of the spacecraft, which is used to adjust and control the attitude and orbit of the spacecraft. Therefore, the normal use of the reaction flywheel is crucial. By installing several sensor groups on the reaction flywheel, the daily operation data of the reaction flywheel can be collected. The daily operation data can be used to quickly evaluate whether the reaction flywheel has a fault. The real-time speed, motor current, bearing temperature and vibration data of the reaction flywheel are the key to the collected operation data. When the speed is abnormal, it may be that the bearing is stuck, the coupling is worn, the motor is faulty, the power supply system is faulty, etc. When the motor current is abnormal, it may be a motor fault, a drive circuit fault, a control system fault, etc. When the bearing temperature is abnormal, it may be bearing wear, poor lubrication, stuck or motor failure, etc. When the vibration data is abnormal, it may be a flywheel fault, a loose mounting base, a motor fault or a bearing fault, etc. After collecting the operation data of the reaction flywheel, it can be roughly judged whether the reaction flywheel has a fault based on the operation data.

[0057] In order to ensure that the running data can be processed by the subsequent neural network model, the present invention pre-processes the running data, including data cleaning, denoising and feature extraction, wherein data cleaning includes missing value processing: checking the missing values ​​in the data, and selecting appropriate processing methods according to the attributes of the data set and the location of the missing values, such as deletion, interpolation or model filling; outlier detection and processing: detecting outliers through visualization methods such as histograms and box plots, and selecting appropriate processing methods according to business needs, such as deletion or correction; duplicate value detection and processing: detecting duplicate values ​​by comparing the unique values ​​and the total number in the data set, and using processing methods such as deletion, merging or modification; data format unification: ensuring that the data format in the data set is unified, such as date format, numerical type, etc., to facilitate subsequent data analysis and processing; data cleaning can improve the accuracy and credibility of the data, provide a reliable basis for subsequent data analysis, and remove redundant and erroneous data, simplify the data processing process, and improve efficiency.

[0058] Denoising includes at least one of mean filtering, median filtering, Gaussian filtering and adaptive filtering, which can remove some irrelevant noise in the running data, reduce the interference of noise on the data analysis results, make the data smoother, and facilitate subsequent data analysis and modeling

[0059] Feature extraction can extract the most representative information from the original data, provide strong support for subsequent data analysis and modeling, reduce the dimension of the data, reduce redundant information, improve data processing efficiency and model performance, so that the neural network model can be processed quickly.

[0060] Step S2, obtaining possible faults of the reaction flywheel and the probability of the fault occurring according to the preprocessed operating data;

[0061] The specific steps include:

[0062] Step S21, constructing a neural network model and performing training to map the normal range of the reaction flywheel operation data into the neural network model;

[0063] Step S22, inputting the operation data into the neural network model, and the neural network model processes the operation data;

[0064] Step S23: The neural network model outputs possible faults of the reaction flywheel and the probability of the fault occurring.

[0065] After obtaining the operation data, the present invention uses a deep learning model to process the operation data, wherein the deep learning model uses a neural network model. The neural network model is a complex network system formed by widely interconnecting a large number of simple processing units (called neurons). It is a highly complex nonlinear dynamic learning system. These neurons transmit information to each other through weighted connections and can adjust their connection weights according to the input information and the current state of the network, thereby realizing learning and adaptation. The neural network has large-scale parallel, distributed storage and processing, self-organization, self-adaptation and self-learning capabilities, and is particularly suitable for processing inaccurate and fuzzy information processing problems that require considering many factors and conditions at the same time. After pre-training, a trained neural network model can be obtained, and then the normal value of the corresponding operation data is mapped to the neural network model, and the operation data can be processed by the neural network model. The operation data is the input of the neural network model, and the output of the neural network model is the possible faults and the probability of occurrence of each possible fault. The present invention introduces a neural network, and can identify the characteristic mode of the reaction flywheel fault by learning a large amount of fault data. Even in a complex spatial environment, the reaction flywheel fault can be quickly and accurately detected, and the neural network model can automatically extract fault features through learning and training, thereby reducing the complexity of fault diagnosis. This makes the troubleshooting process simpler, more efficient and easier to implement.

[0066] Preferably, the specific steps of constructing the neural network model and training it in step S21 are: obtaining historical data of the reaction flywheel, wherein the historical data includes historical operation data and fault records of the reaction flywheel, taking the historical operation data and fault records as a group of data, and dividing them into a training set and a test set in a ratio of 7:3, training the neural network with the training set, and after the training is completed, testing it with the test set, and obtaining a trained neural network model after the test accuracy reaches a set threshold.

[0067] The rich historical data accumulated in the past of the reaction flywheel were systematically sorted out, and divided rigorously and meticulously according to specific fixed ratios to form two key data sets: training set and test set. Both the training set and the test set comprehensively cover the massive operating data generated by the reaction flywheel in its long historical operation process. These data record in detail the changes in various parameters of the flywheel under different working conditions and different operating stages. At the same time, they also include corresponding fault records. These fault records clearly point out the specific circumstances and relevant data characteristics of the faults that have occurred in past operations.

[0068] After the neural network model is built, its core training phase is immediately started, and it is engaged in in-depth learning of the training set data. In this process, the model will be repeatedly trained on the data in the training set based on a complex algorithm architecture. Each training is a fine adjustment and optimization of the model's internal parameters, continuously improving the model's understanding and grasp of data features, so that it can gradually and accurately capture the potential correlation and patterns between operating data and faults. As training continues, the performance of the neural network model is also gradually improving and evolving.

[0069] When the training process reaches a certain stage, the test set can be used for accuracy testing. If the test accuracy of the neural network model on the test set can successfully reach the preset threshold standard, it can be officially put into use. The neural network model can receive various data during the operation of the reaction flywheel in real time, and quickly and accurately determine whether there are any hidden faults, providing strong technical support and guarantee for the stable operation of the reaction flywheel and even the entire satellite system. In addition, during the operation of the reaction flywheel, real-time data can be obtained to continuously update the neural network model, gradually improving its fault diagnosis accuracy.

[0070] Step S3, constructing a fault diagnosis path of the reaction flywheel;

[0071] The specific steps are:

[0072] Step S31, obtaining several diagnostic components of the reaction flywheel for fault diagnosis;

[0073] Step S32, constructing a virtual sphere, and evenly mapping the diagnostic components onto the surface of the virtual sphere;

[0074] Step S33, starting from the center of the virtual sphere, randomly selecting at least one diagnostic component on the surface of the virtual sphere and connecting them in sequence to form a plurality of fault diagnosis paths;

[0075] Step S34: according to the possible fault of the reaction flywheel, remove the irrelevant parts in the fault diagnosis path.

[0076] After confirming the possible faults of the reaction flywheel and the probability of the fault occurring, it is necessary to confirm which components of the reaction flywheel to start the fault diagnosis. Some components are associated with each other. For some faults, it is necessary to confirm that the previous component is fine before the fault can be determined through the next component. Therefore, it is necessary to confirm the fault diagnosis path for different components. When constructing the fault diagnosis path, the present invention uses a spherical construction method to first build a virtual sphere, take the center of the virtual sphere as the initial node, obtain several diagnostic components for the reaction flywheel for fault diagnosis, and then use the diagnostic components as other nodes to evenly distribute them on the surface of the virtual sphere. Starting from the initial node, it is possible to reach the node where any diagnostic component is located. At least two nodes can form a fault diagnosis path, and then from the node where the first diagnostic component is located, it is possible to reach other nodes, and finally several fault diagnosis paths can be formed. The fault diagnosis path contains at least two nodes, and there will be some noise in the selected fault diagnosis path, that is, a path that is completely irrelevant to the possible fault. After the corresponding fault diagnosis path is eliminated, the number of fault diagnosis paths can be reduced, and the subsequent calculation amount can be reduced.

[0077] The distance from the center of the virtual sphere to the surface of the sphere is the same, that is, the distance from the initial node to the first node of all fault diagnosis paths is also the same. Since the initial node does not require fault diagnosis, the time between the initial node and the first node is set to be unified through the setting of the virtual sphere, so as to facilitate unified comparison of the subsequent Grey Wolf optimization algorithm.

[0078] Step S4: using the Grey Wolf Optimization Algorithm to process the fault diagnosis paths based on possible faults and fault occurrence probabilities, and obtain the ranking of the fault diagnosis paths;

[0079] The specific steps of S4 include:

[0080] Step S41, selecting possible faults based on the probability of fault occurrence and the severity of the fault;

[0081] The specific steps include:

[0082] Step S411, assessing the severity of the impact of possible faults and quantifying them;

[0083] Step S412: assign values ​​to the fault occurrence probability and the quantified impact degree respectively, and sum them up;

[0084] Step S413: select possible faults from large to small according to the summation result.

[0085] Before the Grey Wolf Optimization Algorithm performs sorting of fault diagnosis paths, it is first necessary to determine the possible faults to be selected. Since there are multiple possible faults, each possible fault will be processed to obtain the corresponding impact severity, so the selection order of the possible faults is crucial. After the failure probability of each possible fault can be obtained through the neural network model, the present invention will also evaluate the severity of the impact of the possible faults, and then comprehensively evaluate the importance of the possible faults based on the severity and the failure probability. For faults with a higher probability of occurrence and a more serious impact, they need to be processed first. If the selection order is inappropriate, it may lead to a lot of energy being spent on some relatively minor fault investigation and diagnosis under limited resources, and delay the timely processing of key faults, thereby causing serious consequences.

[0086] When evaluating the severity of the impact of possible faults, the empirical analysis method can be used to evaluate the severity of the impact in combination with historical data, and then the quantified severity of the impact can be obtained after quantification through expert scoring. Since the numerical range, measurement standard and data nature involved in the probability of fault occurrence and the severity of impact are often not unified, it is necessary to convert the two into the same comparable numerical system based on specially designed assignment rules. For example, for the probability of fault occurrence, the corresponding numerical value can be assigned after comprehensive consideration of multiple factors such as its frequency of occurrence in historical data and the fluctuation range under the influence of relevant environmental factors; for the quantified severity of the impact, it is also reasonably assigned based on its importance weight in the entire system operation architecture and the degree of interference with the execution of key tasks. After the assignment is completed, the two sets of data are summed. Through this summation result, a numerical indicator that comprehensively reflects the importance of the fault can be obtained, that is, the importance of the fault. When selecting possible faults for processing, the gray wolf optimization algorithm selects them from large to small according to the importance of the fault, so as to obtain the fault diagnosis path corresponding to the fault with a greater degree of importance for fault diagnosis.

[0087] Step S42, initializing the gray wolf population and setting the number of iterations, where the number of iterations is the number of fault diagnosis paths;

[0088] Step S43: randomly select a fault diagnosis path and calculate the fitness value based on possible faults;

[0089] Step S44, selecting another fault diagnosis path, calculating the fitness value, comparing it with the previous calculation result, and obtaining a sorting result after sorting according to the comparison result.

[0090] Then, the gray wolf optimization algorithm is used to calculate the fitness values ​​of all fault diagnosis paths for the same possible fault. The fitness value is a key indicator to measure the quality of this fault diagnosis path. It comprehensively considers many factors such as the effectiveness, accuracy and efficiency of the fault diagnosis path in dealing with various possible faults. For example, for a fault diagnosis path, if it can quickly and accurately detect faults with high probability and serious impact, and is relatively economical in resource consumption, then its fitness value will be relatively high; conversely, if a fault diagnosis path has a high missed detection rate or false detection rate when detecting faults, and requires a lot of time and resources, its fitness value will be lower. After multiple comparisons, the comparison results are sorted, and the fault diagnosis path with the largest fitness value has the highest priority.

[0091] Step S5: perform fault diagnosis based on the sorting results, and select fault diagnosis paths from large to small according to the sorting results to perform fault diagnosis.

[0092] During the entire iterative calculation process, the gray wolf optimization algorithm gives full play to the advantages of group intelligence. The individuals in the gray wolf population cooperate with each other, share information, and work together towards the optimal solution. Fault diagnosis paths with larger fitness values ​​will stand out and be given higher priority. In actual fault diagnosis work, once the system detects a possible fault, it will prioritize these fault diagnosis paths in order from large to small for investigation and diagnosis, quickly locate possible faults in the reaction flywheel, greatly improve the efficiency and accuracy of fault diagnosis, effectively ensure the stable operation of the entire system, and reduce potential risks and losses caused by delayed fault diagnosis.

[0093] A fault diagnosis method for a reaction flywheel of the present invention collects the operating data of the reaction flywheel and then uses a neural network model to process the pre-processed operating data, so as to quickly identify possible faults of the reaction flywheel and the probability of the fault occurring, then obtain components of the reaction flywheel for fault diagnosis and map them onto the surface of a virtual sphere, connect the nodes on the surface of the virtual sphere and the center of the circle to form a number of fault diagnosis paths, and then use the gray wolf optimization algorithm to compare and sort the fault diagnosis paths based on possible faults, and finally select the fault diagnosis path to perform fault diagnosis on the reaction flywheel according to the sorting result, quickly locate the fault location, improve the fault diagnosis efficiency, and avoid affecting the normal operation of the spacecraft.

[0094] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for fault diagnosis of a reaction flywheel, characterized in that: The following steps are involved: Step S1, collecting the operating data of the reaction flywheel and preprocessing the operating data; Step S2, obtaining possible faults of the reaction flywheel and the probability of the fault occurring according to the preprocessed operating data; Step S3, constructing a fault diagnosis path of the reaction flywheel; Step S4: using the Grey Wolf Optimization Algorithm to process the fault diagnosis paths based on possible faults and fault occurrence probabilities, and obtain the ranking of the fault diagnosis paths; Step S5: perform fault diagnosis based on the sorting result.

2. A method for fault diagnosis of a reaction flywheel according to claim 1, characterized in that: The specific steps of step S1 include: Step S11, collecting real-time speed, motor current, bearing temperature and vibration data of the reaction flywheel; Step S12: performing data cleaning, denoising and feature extraction on the collected real-time speed, motor current, bearing temperature and vibration data.

3. A method for fault diagnosis of a reaction flywheel according to claim 1, characterized in that: The specific steps of step S2 include: Step S21, constructing a neural network model and performing training to map the normal range of the reaction flywheel operation data into the neural network model; Step S22, inputting the operation data into the neural network model, and the neural network model processes the operation data; Step S23: The neural network model outputs possible faults of the reaction flywheel and the probability of the fault occurring.

4. A method for fault diagnosis of a reaction flywheel according to claim 3, characterized in that: The specific steps of constructing the neural network model and training it in step S21 are as follows: obtaining historical data of the reaction flywheel, wherein the historical data includes historical operation data and fault records of the reaction flywheel, taking the historical operation data and fault records as a group of data, and dividing them into a training set and a test set in a ratio of 7:3, training the neural network through the training set, and after the training is completed, testing it through the test set, and obtaining a trained neural network model after the test accuracy reaches a set threshold.

5. A method for fault diagnosis of a reaction flywheel according to claim 1, characterized in that: The specific steps of step S3 are: Step S31, obtaining several diagnostic components of the reaction flywheel for fault diagnosis; Step S32, constructing a virtual sphere, and evenly mapping the diagnostic components onto the surface of the virtual sphere; Step S33: starting from the center of the virtual sphere, randomly selecting at least one diagnostic component on the surface of the virtual sphere and connecting them in sequence to form a plurality of fault diagnosis paths.

6. A method for fault diagnosis of a reaction flywheel according to claim 5, characterized in that: The step S3 further comprises: Step S34: according to the possible fault of the reaction flywheel, remove the irrelevant parts in the fault diagnosis path.

7. A method for fault diagnosis of a reaction flywheel according to claim 1, characterized in that: The specific steps of step S4 include: Step S41, selecting possible faults based on the probability of fault occurrence and the severity of the fault; Step S42, initializing the gray wolf population and setting the number of iterations; Step S43: randomly select a fault diagnosis path and calculate the fitness value based on possible faults; Step S44, selecting another fault diagnosis path, calculating the fitness value, comparing it with the previous calculation result, and obtaining a sorting result after sorting according to the comparison result.

8. A method for fault diagnosis of a reaction flywheel according to claim 7, characterized in that: The specific steps of step S41 include: Step S411, assessing the severity of the impact of possible faults and quantifying them; Step S412: assign values ​​to the fault occurrence probability and the quantified impact degree respectively, and sum them up; Step S413: select possible faults from large to small according to the summation result.

9. A method for fault diagnosis of a reaction flywheel according to claim 7, characterized in that: The number of iterations of step S42 is the number of fault diagnosis paths.

10. A method for fault diagnosis of a reaction flywheel according to claim 1 or 7, characterized in that: The specific steps of step S5 are: selecting a fault diagnosis path from large to small according to the sorting result to perform fault diagnosis.

Images