CN113567858A - Control moment gyro residual life prediction system - Google Patents
Control moment gyro residual life prediction system Download PDFInfo
- Publication number
- CN113567858A CN113567858A CN202110867310.3A CN202110867310A CN113567858A CN 113567858 A CN113567858 A CN 113567858A CN 202110867310 A CN202110867310 A CN 202110867310A CN 113567858 A CN113567858 A CN 113567858A
- Authority
- CN
- China
- Prior art keywords
- control moment
- moment gyro
- residual life
- life prediction
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/34—Testing dynamo-electric machines
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses a control moment gyro residual life prediction system, which mainly comprises a control moment gyro fault mechanism modeling module, a signal characteristic analysis module and a residual life prediction module: the control moment gyro fault mechanism modeling adopts a multi-signal flow diagram mode to describe the multi-level fault and test transfer relation of the control moment gyro; the signal characteristic analysis module is mainly responsible for extracting the signal characteristics of the control moment gyro sensor and screening the characteristics suitable for predicting the residual life based on indexes; and the residual life prediction module evaluates the residual life of the control moment gyroscope based on the screening signal characteristics. According to the method, the Django framework is used for building a browser/server architecture reasoning platform, the structure of the reasoning platform is provided in the form of Restful API, a residual life prediction model is built in a data-driven mode, the difficulty of building a physical model caused by increasing complexity of a system is reduced, and the problem of model migration is relieved to a certain extent.
Description
Technical Field
The invention relates to a residual life prediction system for controlling moment gyro degradation characteristics, in particular to a fault mechanism modeling, degradation characteristic analysis and residual life prediction module for a controlling moment gyro.
Background
With the rapid development of the aerospace technology, higher requirements are put forward on the long-life operation of the electromechanical products for the spacecraft. In recent decades, large-scale spacecrafts represented by remote sensing satellites have been developed rapidly, and the requirements for high precision and long service life are met while the working conditions are complex. The control moment gyroscope is a key component for spacecraft attitude control as a representative spacecraft electromechanical device, and the performance of the control moment gyroscope is related to the attitude control precision, maneuvering indexes and the service life of the spacecraft.
According to design requirements, the control moment gyro needs to continuously work under extreme working conditions during the service period of the rail (more than 8 years): the moment gyro product is controlled to work in a vacuum environment, and the heat dissipation condition is poor. Meanwhile, the high-speed rotor keeps constant high-speed operation for a long time, and when the spacecraft is in maneuvering, the control moment gyro product can be acted by huge coupling moment caused by star rotation. Therefore, according to the above conditions, under normal and extreme conditions, the control moment gyro product is always subjected to the action of multiple physical fields (thermal, force, vibration, etc.): the high-speed rotor of the control moment gyroscope always works under the action of unbalanced moment generated during operation and under the action of coupling moment caused by star rotation; the control moment gyroscope is always in a relatively complex thermal environment under the heat generated during high-speed rotation and an external thermal vacuum environment; due to the structural complexity of the high-speed rotor product and the uninterrupted unbalanced moment of the product during high-speed operation, the control moment gyro product is always in a vibration field.
In combination with the background, at present, domestic research on control moment gyro service life prediction is in a starting stage, the control moment gyro product is taken as a research object, multi-performance combined degradation analysis and service life prediction of the control moment gyro are realized under the action of on-orbit extreme working conditions (long-term high-speed operation, vacuum environment, frequent application of coupling moment and the like) and multiple physical fields (heat, force, vibration and the like), and a corresponding platform is developed to realize technical application.
Disclosure of Invention
The technical content of the subject mainly centers on the failure mechanism of the control moment gyroscope, combines the roles of all components of the control moment gyroscope and the structural relationship of a system, and adopts a method of a multi-signal flow diagram to realize the modeling of the component level and the complete machine level of the control moment gyroscope.
Through the system-level model correlation of the control moment gyro, the fault mechanism of the control moment gyro is analyzed according to the mutual influence relation among the low-speed frame shaft temperature, the low-speed frame rotating speed, the high-speed rotor shaft temperature, the high-speed rotor rotating speed, the high-speed rotor current and the voltage, and a typical fault mode and a measuring point in the control moment gyro are obtained through sorting. On the basis, a multi-signal flow diagram model of the control moment gyro is constructed, and the transfer relation of the sensor signals in the model is described.
According to the common data of a plurality of groups of bearing parts, the influence of extreme working conditions and multi-field coupling is considered, and a deep learning model is established. And the migration from the common data to the data of the single product is realized through the online updating of the model, and the prediction of the degradation trend of the single product is completed.
And developing a control moment gyro residual life prediction system to realize core functions of real-time data receiving, fault diagnosis, life prediction and the like. A distributed service platform with a browser/server model as a basic framework is designed, decoupling of an algorithm platform and a monitoring platform is achieved, and online updating of a health management algorithm is guaranteed.
The invention has the advantages that:
1. the control moment gyro fault mechanism is constructed in a multi-signal flow graph mode, and compared with a form and document carding mode, on one hand, the man-machine interaction is optimized; on the other hand, the graphical description mode has clearer description on the transmission relation of the faults at the same level and multiple levels.
2. The control moment gyro degradation characteristic analysis and residual life prediction method adopts a data-driven deep learning model, effectively avoids the problem that a physical model is difficult to construct due to the aggravation of the complexity of a control moment gyro system, and effectively realizes the problems of model sharing and switching among different single-machine products.
3. The control moment gyro residual life prediction system adopts a browser/server architecture, has the characteristics of light system, strong real-time performance, strong portability and multi-task parallelism compared with a client/server architecture, and is particularly suitable for multi-node and multi-task on-orbit task monitoring.
4. And the control moment gyro residual life prediction system ground reasoning platform accesses data resources by adopting RESTful API. The RESTful architecture follows a uniform interface principle, comprising a restricted set of predefined operations, with access to all resources by using the same interface. Meanwhile, according to the semantic exposed resources of the HTTP method, the interface has the characteristic of idempotency, and the safety of the server when the access amount is large and the network is unstable is guaranteed.
Drawings
FIG. 1 is a system hardware framework diagram
FIG. 2 is a system framework flow diagram
FIG. 3 is a multi-signal flow graph modeling and fault detection
FIG. 4 is a block diagram of a multi-signal flow diagram design and display module
FIG. 5 is a feature selection flow chart
FIG. 6 is a table of common signal characteristics
FIG. 7 is a schematic diagram of an extreme learning machine
FIG. 8 is a life prediction neural network framework
FIG. 9 is a flow chart of the operation
FIG. 10 is a flowchart of inference platform work
Detailed description of the preferred embodiments
The following result figures illustrate the control moment gyro remaining life prediction system described in the present invention in detail:
1 System framework
1.1 System Overall design
The control moment gyro life prediction system architecture is shown in fig. 1. According to the requirement of application requirements on a system platform, a data receiving and displaying function is used as one of core functions of a data monitoring platform, real-time performance is an important index of the data monitoring platform, and performance conflict exists between the data receiving and displaying function and life prediction reasoning under the background that the data volume of a control moment gyro is large. Therefore, in order to ensure the real-time performance of data receiving and displaying, and also ensure the effective execution of the state evaluation and life prediction algorithm, and simultaneously based on the principles of convenient deployment, flexible configuration and adaptation to low-performance hardware, the system adopts a software structure based on a browser/server architecture, so as to realize the decoupling of a data monitoring platform and a life prediction inference engine module, ensure the real-time performance of data receiving and visualization, and simultaneously meet the basic configuration environment requirements of the algorithm with relatively large resource consumption.
This section describes a software architecture deployed in a ground control center for control moment gyro health management, whose functions include information management, telemetry data management, outlier detection, and fault diagnosis.
The system comprises two subsystems. The data center subsystem receives real-time data of the on-orbit satellite and provides functions of data management, visualization and the like. In addition, as a data concentration node, it sends data to the computing platform subsystems of the system for complex data processing and health assessment algorithms. Fig. 2 shows the overall structure of the system.
1.2 data center subsystems
The functions of the data center subsystem include data receiving, data storing and man-machine interaction. A data center is considered as a combination of a database and a function server.
First, considering that a plurality of computing nodes and a plurality of clients access the system at the same time, all information, including real-time monitoring data, historical monitoring data, anomaly detection algorithms, diagnostic algorithms and expert knowledge, is stored in the data center.
Historical data stored in the database is used to train outlier detection models and diagnostic models. During the flight mission, the ground control center will receive sensor data from the satellites. These data will indicate the satellite health status. At the same time, the data center provides a human-machine interface for the customer to check the health status of the satellites, data curves, and other information related to the flight mission.
In addition, the data center provides rich human-computer interfaces, including functional structures such as drawing of multi-signal flow diagrams, feature extraction and selection of a residual life prediction method, data import and data export and management interfaces such as user management, project management and data management.
The real-time receiving and displaying of the data comprises receiving, storing and displaying the data in real time. And the data receiving part adopts a RESTful API mode, and a data sender actively sends data to the system to the data center according to an HTTP protocol. On one hand, the system man-machine interaction module and the data receiving module can be decoupled by doing so; on the other hand, considering the requirement of real-time performance, when the Django framework cannot meet the requirement of higher-speed data transmission, the data can be analyzed by developing data receiving modules of other languages, and the data can be transmitted to the data center to store data according to the requirement of the HTTP protocol, so that the data can be transmitted in time.
1.3 computing platform subsystems
In order to decouple from the data center subsystem and provide a more efficient calculation execution mode, the calculation platform preprocesses data, diagnoses faults in the data received from the data center, extracts features, realizes a calculation part of degradation feature analysis and residual life prediction, and returns calculation and analysis results to the data center.
First, the computing platform will send a request to the data center for processing environment configuration, including the computing model and knowledge database, at the beginning of deployment. After deployment, the historical data will be sent to the new computing platform for model training. Since model training is a time consuming process, it is performed off-line.
Real-time data from the data center will be used for online outlier detection and diagnosis, as well as degradation signature analysis and remaining life prediction while performing the flight mission. The computing platform processes the data with the model and knowledge, packages the result in a fixed format, and sends the result back to the data center, so that a conclusive result is provided for the data center, and the requirement of a user on monitoring the control moment gyro is met.
2. Control moment gyro fault mechanism modeling based on multiple signal flow diagrams
Aiming at the defects of the traditional modeling method, the invention adopts a multi-signal flow diagram method to model the component level and the whole machine level of the control moment gyro. And the multi-signal flow diagram takes the signal flow as a research object, analyzes the signal flow propagated by the fault mode of each component of the control moment gyro lubrication system and the rotor motor and the signal flow detected by each measuring point, and adds the description of the signal dependency relationship on the system structure mode on the basis of the signal flow, thereby completing the modeling process of the multi-signal flow diagram. The method integrates structural modeling and dependency modeling methods, overcomes the using defects of the structural modeling and the dependency modeling, and has better model expression.
The multi-signal flow diagram model adopts a directed graph to describe the fault propagation dependency relationship among components of the control moment gyro, the composition, performance characteristics and test flow relationship of the control moment gyro are represented by a layered directed graph, a system signal system is divided into component related attributes and detection related attributes, causal relationship exists between the component related attributes and the detection related attributes, and the catastrophic fault and the general fault are obtained by rearrangement according to the influence range of the fault and in combination with fault type description.
The multi-signal flow diagram model is a signal flow situation model in a system, a system signal is divided into two-dimensional attributes from components and from testing, and flow and causal connection between the signals is established, wherein the signal attributes from the components generally comprise fault modes and relevant parameters, and the signal attributes from the testing are signal data detected by a measuring point. The main elements in the multi-signal flow diagram include modules, subsystems, tests, switches, and nodes, connecting lines. The module component corresponds to a failure mode, the subsystem module is used for realizing multi-level modeling of the system, the switch component can be used for representing multi-mode characteristics of the system, the node component is used for representing redundancy characteristics of the system, and the connecting line represents the flow direction of signals. The multi-signal model consists of the following eight parts:
1) l sets of system components C ═ { C ═ C1,c2……cL}
2) K sets of system test signals S ═ S1,s2……sK}
3) N-dimensional finite test set T ═ T1,t2……tn}
4) P-dimensional measurement point set TP ═ { TP ═ TP1,TP2……TPp}
5) Any measuring point TPtCorresponding set of test sets SP (TP)t)
6) Any component ciA corresponding set of semaphores SC (c)i)
7) Any test tjDetecting a set of signals ST (t)j)
8) The directed graph constitutes DG ═ C, TP, E, where the directed edge E represents the physical connection of the system.
The fault diagnosis process is as shown in fig. 3, for each measuring point in the multi-signal flow diagram, the upper and lower limits of the time series threshold are obtained by a data driving method and the idea of the SPOT clustering algorithm, in the actual fault diagnosis process, real-time data is received and compared with the upper and lower values to obtain the state of the parameter, and then the abnormal conditions of the system, the subsystem and the single-machine product are obtained by inference step by step with how the real-time data is received and a rule method.
The multi-signal flow diagram design interface is shown in fig. 4, a user can place nodes in the interface in a dragging mode, all modules of the system are linked through design, a design tool provides nodes of three layers of a single machine product, a subsystem system and test, a JSON format file for describing the nodes can be derived through model storage, and the file can be used as a source file loaded by a model and can also be used as an inference basis for deducing faults of the single machine product and the subsystem through a measuring point in fault diagnosis.
3. Characteristic parameter analysis and remaining life prediction
3.1 feature analysis
As a key part for controlling the health condition of the moment gyro, the motor current in the high-speed rotor bearing is taken as an important health factor, and can be selected as a life prediction basis of the control moment gyro on the premise that the vibration data of the bearing cannot be acquired. Since the current signal appears as a vibration-like signal in the time domain, the original time series needs to be processed to select the appropriate characteristics as the health factor. Generally, for a sensor signal presenting vibration characteristics, the original signal can be subjected to feature extraction from three angles of a time domain, a frequency domain and a time-frequency domain, and common signal features mainly include the following parts:
for the above features, it is theoretically possible to extract all the features and use the extracted features as health factors, but in application, on one hand, the dimensionality of the signal is increased, so that the calculation amount is increased invisibly in the training of subsequent data, and on the other hand, not all the features have the degradation tendency according to the characteristics of the signal features. Therefore, adding these features to data training does not, but does not, increase the accuracy of life prediction, but rather increases the computational load invisibly. This reduces the efficiency of the system in online life prediction of the control moment gyro with high real-time performance and large data volume. Therefore, in the data set training stage, the degradation trend of the high-speed rotor of the control moment gyro needs to be analyzed, and further, in the invention, the data is analyzed, and a proper characteristic is selected as a health factor for predicting the service life of the high-speed rotor.
The working flow of the high-speed rotor current signal characteristic selection is as shown in the following figure 5
Aiming at an original signal, due to the existence of sensor noise and external influence factors, the acquired signal needs to be denoised, and the method adopts a wavelet analysis denoising method to denoise the data in a frequency domain; the denoised data is grouped according to a time sequence, and a characteristic extraction method in a common signal characteristic table in fig. 6 is used for extracting characteristics from each group of collected data sequences to obtain a plurality of groups of sequences with different characteristics. If a feature has a tendency to degrade, then theoretically the time series of the feature can exhibit an approximately monotonic tendency to decline or rise; extracting an adaptive regression model from the sequence of each feature, constructing a recursive model mode, wherein an output value is the next signal value in the current model, and if the model is established, the model can be approximately regarded as that the data has certain trend change; and constructing an extreme learning machine model, evaluating the efficiency and the action of the autoregressive model, selecting proper indexes and threshold values, and extracting the first N characteristics which most accord with the indexes. Theoretically, these characteristics have a relatively good correlation with the remaining life, so that in an actual environment, without considering the influence of an external environment and other signals, depending on historical data and experimental data, a health factor capable of representing the remaining life can be found.
(1) Feature extraction
As a candidate for feature extraction in the sequence, different time sequences are obtained, and the core feature selection definition is shown in FIG. 6
Aiming at the characteristics in the figure 6, the data are respectively subjected to characteristic processing to obtain a proper characteristic selection image
(2) Autoregressive model
The autoregressive model is a statistical method of processing time series using the previous stages of the same variable, e.g., x, i.e., x1To xt-1To predict xtAnd assume that they have a linear relationship. Since this is developed from linear regression in regression analysis, but instead of predicting y with x, x predicts x (itself); so called autoregressive.
In the invention, an autoregressive model is constructed for each feature, and if a model window is p, the model of the framework is represented as follows:
wherein a iskFor model parameters, p is the model order, i.e. the window of value prediction, enIn the invention, the model parameters are determined by a Yule-Walker method, and meanwhile, in order to measure the effectiveness degree of the model, coefficients of Akaike Information Criterion (AIC) are used as labels, and the calculation mode is as follows:
wherein
In order to measure the window value, the invention still adopts a traversal mode, selects a value from 1 to 100 as a candidate value, takes AIC as a measurement standard, and finally selects 10 as a value prediction window.
(3) Extreme learning machine model
After the autoregressive model is built, the relationship between the autoregressive model and the residual service life is built through an ELM model, the network structure of an extreme learning machine model is the same as that of a single hidden layer feedforward neural network, and the extreme learning machine model is not a gradient-based algorithm (back propagation) which is frequently tried in the traditional neural network in a training stage, but random input layer weight and deviation are adopted, and the output layer weight is calculated through a generalized inverse matrix theory. And the training of the extreme learning machine is completed after the weights and the deviations on all the network nodes are obtained, and the data can be predicted by the network output completion by utilizing the output layer weights obtained just before the test data comes.
Assume training set { xi,yi|xi∈RD,ti∈RD,i＝1,2,3,4……N},xiDenotes the ith data instance, yiThe number of hidden layer nodes of the extreme learning machine is L, the network structure of the extreme learning machine is as shown in FIG. 7
3.2 remaining Life prediction
The problem of difficult frequency domain analysis of the rotor current signal is caused by relatively low real-time sampling rate of the control moment gyroscope. Therefore, a specific means is required for high frequency reconstruction of the signal. The compressed sensing technology is a signal compression technology, and the initial purpose is to reduce the dimension of a signal on the basis that information is kept as much as possible in a sparse signal. With further research, the compressed sensing technology can achieve reconstruction of low-frequency signals to a certain extent according to selection of a specific observation matrix and a specific compression matrix. Due to the weak constraint of the sparsity of the current signal in the frequency domain, the current signal can realize the high-frequency reconstruction of the signal in a certain frequency domain range. Meanwhile, the integrity constraint of the information after signal reconstruction is reduced by considering the energy description of the signal, so that the compressed sensing technology can be used as a core part of signal preprocessing.
The life prediction model architecture is shown in fig. 8. Along with the increasing complexity of the system structure, the accurate physical model is increasingly difficult to construct, and correspondingly, along with the development of the computer technology, the acquisition quantity of data is increased increasingly, and through the deep learning technology, the black box type model construction is possible by combining a large amount of data. As an important tool for time series analysis, the long-time neural network (LSTM) can effectively realize time series prediction. Meanwhile, based on a Dynamic Bayesian (DBN) thought, the service life at a certain moment can be considered to be related to the hidden state and output of the LSTM, and the probability relation between the state and the service life can be roughly estimated through a large amount of data.
According to the model block diagram, the invention adopts a combined model of a compressed sensing technology and a recurrent neural network to predict the service life. Firstly, low-frequency current time data are mapped to a frequency domain with the bandwidth of 0-100 Hz through a compressed sensing technology, the bandwidth is divided into 10 parts through Pasval theorem, and energy is respectively taken
After a 10-dimensional vector is obtained, each dimension represents the energy of a certain bandwidth. And (3) transmitting the time sequence forming the 10-dimensional vector into a model, outputting a hidden state, and predicting the residual life based on a dynamic Bayesian updating method.
The self-learning life prediction inference engine module flow is shown in fig. 9, and the complete model training comprises two parts of online training and offline training.
In the off-line training, the model is completely trained by depending on experimental data and historical data, so that the process of the model from scratch is realized, the training process is off-line training and is basically aimed at constructing a relatively universal model, so that the requirement on the training data volume is high, and meanwhile, the requirement on the real-time performance of time is not high due to the early preparation stage.
Unlike the on-line training requirement, since the model is already applied to the actual requirement, two aspects need to be considered: 1. due to the characteristics of online operation, the model updating has the requirement of real-time performance, so that the dynamic updating of the model can be realized on the basis of not great requirement on data volume. Aiming at the requirement, the method adopts a mode of reducing the historical data window, and carries out fine adjustment on the weight parameter on the basis of the model trained under the on-line condition, thereby not only reducing the requirement of the historical data, but also ensuring the real-time update of the model. 2. How the designed model is deployed into an existing system. In past software designs, it was common practice to shut down the system and restart it after loading the model. This approach can result in, on the one hand, a short loss of system functionality, which is unacceptable in critical tasks, and, on the other hand, requires specialized personnel for redeployment, which is difficult to operate. Aiming at the phenomenon, the invention adopts an online loading mode to reserve the backup model, thereby ensuring that an error model can be backed up and realizing the switching of the model in a working period.
4 flow of operation
The operation flow of the control moment gyro residual life prediction system is shown in fig. 10. After logging in the system, a user uploads or selects a test object and an inference knowledge base and submits the test object and the inference knowledge base, a browser front end generates a form according to user selection and sends the form to a main control server rear end in an HTTP POST mode, a server Django rear end view receives and analyzes parameters, and an HTTP POST request is generated according to a user instruction and a required data URL and sent to a ground inference platform. And after receiving the request, the ground inference platform starts an inference thread, sends a data request to the data assurance center through RESTful API, executes an algorithm according to an inference knowledge base after acquiring JSON format data resources, converts an operation result into a JSON format and returns the JSON format to the main control service terminal in the form of HTTP Response, and the main control service terminal renders the data into HTML pages according to the format and returns the HTML pages to the client browser for the user to browse. The specific process is described as follows:
step 1: the ground inference engine loads a database and related configuration through a python manager.
Step 2: starting an inference machine thread, and serializing a test operation object example;
and step 3: loading a corresponding inference algorithm file and a knowledge base required by an inference algorithm in a database according to the inference engine and the knowledge base selected by the master control server;
and 4, step 4: receiving the telemetering parameters sent by the main control server in an HTTP POST mode, and quantizing the test result after analysis;
and 5: and inputting the test quantification result into a fault diagnosis algorithm to obtain a fault diagnosis result.
Step 6: sending the real-time parameters into a computing platform subsystem to complete feature extraction, degradation feature analysis and control moment gyro residual life prediction
And 7: and integrating the test result into a JSON file format, and returning the test result to the main control server through HTTP Response.
And 8: and closing the thread, writing the operation information into the log file, and destroying the test operation example.
Claims (4)
1. Control moment top residual life prediction system, its characterized in that: the system mainly comprises a control moment gyro fault mode modeling module, a signal characteristic analysis module and a residual life prediction module: the control moment gyro fault mechanism modeling adopts a multi-signal flow diagram mode to describe the multi-level fault and test transfer relation of the control moment gyro; the signal characteristic analysis module is mainly responsible for extracting the signal characteristics of the control moment gyro sensor and screening the characteristics suitable for predicting the residual life based on indexes; and the residual life prediction module evaluates the residual life of the control moment gyroscope based on the screening signal characteristics.
2. The control moment gyro remaining life prediction system of claim 1, wherein: the control moment gyro fault mechanism modeling adopts a multi-signal flow graph modeling mode, provides a visual modeling tool for a user, and describes the connection relation and the fault transmission relation between single-layer and multi-layer fault modes of the control moment gyro and a test node; the control moment gyro signal characteristic screening adopts an autoregressive model and a single hidden layer neural network method, and is based on the Chichi information criterion as the basis of the characteristic screening; and the residual life prediction adopts a mode of combining compressed sensing and a cyclic neural network, and is evaluated based on the parameter degradation trend.
3. The control moment gyro remaining life prediction system of claim 1, wherein: the control moment gyro residual life prediction system can adopt a browser/server architecture and can be loaded on a windows system and a Linux system, wherein a prediction algorithm and a diagnosis and inference knowledge base are stored by adopting MySQL according to real-time sensor signals.
4. The process of predicting residual life of a control moment gyro according to claim 1, wherein:
the first step is as follows: and (3) running a server of the inference system through a python manager.
The second step is that: and operating the browser, logging in an inference engine platform interface, and setting an inference engine algorithm model and an inference engine knowledge base.
The third step: and other data platforms send the telemetering data to the inference engine platform through the URL, and test results are obtained according to the set upper limit and the set lower limit.
The fourth step: and the data platform receives the fault diagnosis result and carries out subsequent operation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110867310.3A CN113567858A (en) | 2021-07-30 | 2021-07-30 | Control moment gyro residual life prediction system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110867310.3A CN113567858A (en) | 2021-07-30 | 2021-07-30 | Control moment gyro residual life prediction system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113567858A true CN113567858A (en) | 2021-10-29 |
Family
ID=78169291
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110867310.3A Pending CN113567858A (en) | 2021-07-30 | 2021-07-30 | Control moment gyro residual life prediction system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113567858A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114919559A (en) * | 2022-07-05 | 2022-08-19 | 中南大学 | System and method for predicting residual service life of braking system based on digital twinning |
| CN115628758A (en) * | 2022-12-21 | 2023-01-20 | 四川图林科技有限责任公司 | Method for detecting and correcting key process parameters under multi-physical-field coupling condition |
| CN115824261A (en) * | 2023-01-15 | 2023-03-21 | 北京理工大学 | Control moment gyroscope fault detection method and device and related storage medium |
| CN116202558A (en) * | 2023-05-04 | 2023-06-02 | 中国西安卫星测控中心 | CMG rotating part working condition detection method based on incremental data statistics |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2017142153A (en) * | 2016-02-10 | 2017-08-17 | セイコーエプソン株式会社 | Life prediction method, life prediction device, and life prediction system |
| CN110007662A (en) * | 2019-04-28 | 2019-07-12 | 北京航空航天大学 | The full aircraft health evaluating of self -support technical identification system and management ground Inference Platform |
| CN110378052A (en) * | 2019-07-25 | 2019-10-25 | 北京航空航天大学 | It is looked to the future the equipment method for predicting residual useful life of operating condition based on Recognition with Recurrent Neural Network |
| US20200067969A1 (en) * | 2018-08-22 | 2020-02-27 | General Electric Company | Situation awareness and dynamic ensemble forecasting of abnormal behavior in cyber-physical system |
| CN112784501A (en) * | 2021-03-23 | 2021-05-11 | 中国核电工程有限公司 | Modeling system and method for residual life prediction model of equipment and prediction system |
| CN112818490A (en) * | 2021-03-18 | 2021-05-18 | 中国人民解放军陆军工程大学 | Cloud-edge-architecture equipment digital twin parallel simulation system and method |
-
2021
- 2021-07-30 CN CN202110867310.3A patent/CN113567858A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2017142153A (en) * | 2016-02-10 | 2017-08-17 | セイコーエプソン株式会社 | Life prediction method, life prediction device, and life prediction system |
| US20200067969A1 (en) * | 2018-08-22 | 2020-02-27 | General Electric Company | Situation awareness and dynamic ensemble forecasting of abnormal behavior in cyber-physical system |
| CN110007662A (en) * | 2019-04-28 | 2019-07-12 | 北京航空航天大学 | The full aircraft health evaluating of self -support technical identification system and management ground Inference Platform |
| CN110378052A (en) * | 2019-07-25 | 2019-10-25 | 北京航空航天大学 | It is looked to the future the equipment method for predicting residual useful life of operating condition based on Recognition with Recurrent Neural Network |
| CN112818490A (en) * | 2021-03-18 | 2021-05-18 | 中国人民解放军陆军工程大学 | Cloud-edge-architecture equipment digital twin parallel simulation system and method |
| CN112784501A (en) * | 2021-03-23 | 2021-05-11 | 中国核电工程有限公司 | Modeling system and method for residual life prediction model of equipment and prediction system |
Non-Patent Citations (5)
| Title |
|---|
| DANYANG HAN: "A Remaining Useful Life Prediction Approach", 《IEEE SENSORS JOURNAL》 * |
| YUE ZHANG: "Aeroengine Remaining Life Prediction Algorithm Based on Improved Differntial Time Domain Features and LSTM", 《INFORMATION SYSTEM AND COMPUTER ENGINEERING》 * |
| 栾家辉，代永德: "某型控制力矩陀螺寿命与可靠性评估方法研究", 《设计·工艺》 * |
| 王庐华: "基于深度学习和压缩感知的图像重构算法研究", 《无线互联网科技》 * |
| 王竣德: "基于性能退化的控制力矩陀螺寿命预测", 《科学技术与工程》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114919559A (en) * | 2022-07-05 | 2022-08-19 | 中南大学 | System and method for predicting residual service life of braking system based on digital twinning |
| CN115628758A (en) * | 2022-12-21 | 2023-01-20 | 四川图林科技有限责任公司 | Method for detecting and correcting key process parameters under multi-physical-field coupling condition |
| CN115628758B (en) * | 2022-12-21 | 2023-03-31 | 四川图林科技有限责任公司 | Method for detecting and correcting key process parameters under multi-physical-field coupling condition |
| CN115824261A (en) * | 2023-01-15 | 2023-03-21 | 北京理工大学 | Control moment gyroscope fault detection method and device and related storage medium |
| CN116202558A (en) * | 2023-05-04 | 2023-06-02 | 中国西安卫星测控中心 | CMG rotating part working condition detection method based on incremental data statistics |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113567858A (en) | Control moment gyro residual life prediction system | |
| Li et al. | Intelligent fault diagnosis of rolling bearings under imbalanced data conditions using attention-based deep learning method | |
| CN109766583B (en) | Aircraft engine life prediction method based on unlabeled, unbalanced and initial value uncertain data | |
| CN111340282B (en) | DA-TCN-based method and system for estimating residual service life of equipment | |
| CN109471698B (en) | System and method for detecting abnormal behavior of virtual machine in cloud environment | |
| CN114266278B (en) | Dual-attention network-based equipment residual service life prediction method | |
| Gültekin et al. | A novel deep learning approach for intelligent fault diagnosis applications based on time-frequency images | |
| Nasser et al. | A hybrid of convolutional neural network and long short-term memory network approach to predictive maintenance | |
| Wang et al. | Deep learning domain adaptation for electro-mechanical actuator fault diagnosis under variable driving waveforms | |
| CN112906775A (en) | Equipment fault prediction method and system | |
| Shi et al. | A dual attention LSTM lightweight model based on exponential smoothing for remaining useful life prediction | |
| Kayode et al. | Lirul: A lightweight lstm based model for remaining useful life estimation at the edge | |
| RU2600099C1 (en) | Method of neural network forecasting of change of values of function with its complementary wavelet processing and device for its implementation | |
| Maher et al. | Survey on deep learning applied to predictive maintenance. | |
| CN112560252B (en) | Method for predicting residual life of aeroengine | |
| Yang et al. | Dynamic spatial–temporal graph-driven machine remaining useful life prediction method using graph data augmentation | |
| Li et al. | Deep learning-based QoS prediction for manufacturing cloud service | |
| CN110555008B (en) | Generator comprehensive diagnosis and state evaluation cloud platform | |
| Peng et al. | Remaining useful life prediction for aircraft engines based on grey model | |
| Soni et al. | Predictive maintenance of gas turbine using prognosis approach | |
| Lu et al. | A Novel Bearing Fault Diagnosis Algorithm Based on Adaptive Weighted Multi-view Learning | |
| Yangyang et al. | Research on parallel lstm algorithm based on spark | |
| Colace et al. | Unsupervised Learning Techniques for Vibration-Based Structural Health Monitoring Systems Driven by Data: A General Overview | |
| Ovseenko et al. | The possibility of artificial neural network application in prototyping in instrument making industry | |
| Moradi et al. | Developing health indicators for composite structures based on a two-stage semi-supervised machine learning model using acoustic emission data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20211029 |
|
| WD01 | Invention patent application deemed withdrawn after publication |