Upper stage open electrical intelligent health monitoring and management system

Abstract

The present invention designs an upper stage open electrical intelligent health monitoring and management system, aiming at the health management requirements such as fault diagnosis and prediction evaluation of the upper stage electrical system. On the basis of analyzing the working principle of the upper stage electrical system, through the analysis of the fault mechanism of the electrical system, typical fault modes and fault manifestations are sorted out. Aiming at the fault diagnosis and prediction evaluation requirements of the upper stage electrical system, research and verification of algorithms such as rapid fault diagnosis and fault prediction of the electrical system are carried out to provide technical support for the functional implementation of the health monitoring of the upper stage electrical system.

Description

Technical Field

[0001] The present invention relates to an upper stage open electrical intelligent health monitoring and management system, belonging to the technical research field of intelligent fault diagnosis and health management. Background Art

[0002] The upper stage is generally a relatively independent stage (or multiple stages) added to the basic stage launch vehicle, with strong mission adaptability and capable of completing tasks such as orbital maneuvering and payload separation. The upper stage generally has characteristics such as multiple starts, long-term operation, and autonomous flight, and has the ability of multi-satellite launch and orbital deployment, which is one of the effective ways to improve the rocket performance and mission adaptability.

[0003] Due to the particularity of the mission, the upper stage needs to deploy high specific impulse propulsion technology, high-precision autonomous navigation and integrated navigation control technology, and other advanced technologies, which greatly improves the complexity and intelligence level of the upper stage electrical system. It has developed from the early simple command sending to a multi-processor, automatic fault diagnosis and reconfiguration multi-functional complex intelligent system, providing a hardware basis for the agile control of the launch vehicle, equipment testing, and data analysis automation. The upper stage electrical system of the new generation launch vehicle adopts an integrated design, which makes the electrical system scheme have good versatility, independence, and adaptability.

[0004] The upper stage electrical system mainly consists of a control system, a GNC system, a measurement and communication system, an on-board power supply and power distribution system, and a ground test system. The functional requirements of each system are as follows:

[0005] 1) The control system realizes the autonomous control and management of the upper stage, monitors, redundantly, fault-tolerantly, and reconfigures the management of the health state of the electrical system, and implements the clock management function of the upper stage;

[0006] 2) The GNC system completes the tasks of attitude stabilization and control, navigation, and orbital deployment of the upper stage;

[0007] 3) The measurement and communication system realizes the functions of telemetry, remote control, and tracking measurement;

[0008] 4) The on-board power supply and power distribution system realizes the functions of power supply and power distribution energy management for the upper stage equipment;

[0009] 5) The ground test system realizes the flight control and real-time full-process monitoring of the upper stage.

[0010] The upper stage electrical system of the new generation launch vehicle mainly adopts a 1553B bus architecture, and most of the on-board equipment are intelligent devices with independent processing functions. The overall framework of its electrical system is as Figure 1 shown.

[0011] Due to the complex, harsh and changeable environment in space, the upper stage electrical system is likely to experience component damage, communication interruption, etc. During the upper stage electrical system test, the faults can be divided into safety faults and non-safety faults according to their impact. The specific fault modes can be seen in Table 2. Safety faults are relatively serious faults. On the one hand, they will cause personal injury to relevant personnel and evolve into technical safety accidents; on the other hand, they will cause irreparable damage to the test equipment and even the products on the rocket, and lose their basic functions.

[0012] Table 1 Common failure modes of upper stage electrical system

[0013]

[0014] Due to the complexity of the upper stage electrical system, the location and type of faults during system testing are difficult to predict, and there is no fixed pattern for fault diagnosis. Therefore, emergency response plans for corresponding problems should be formulated before testing. Since the degree of harm caused by safety faults and non-safety faults is different, the treatment of the two types of faults is also different.

[0015] The common steps for handling safety failures are: 1) Cut off the power supply, cut off the source of danger, and prevent the danger from expanding. 2) Fully record the on-site situation: including the time, location, timing, environmental conditions of the failure, the name, model (code), drawing number, number, batch and other comprehensive information of the faulty product; summarize, verify the situation, make records, and take photos and videos of the fault phenomenon when necessary. 3) Analyze and handle problems. Discover and handle problems in a timely manner to reduce equipment losses. When a failure occurs, the platform, arrow machine, servo system and other on-board equipment and each single machine on the ground will have different manifestations: voltage, current, frequency, feedback indication, status indication, etc. Problems should be discovered and reported in a timely manner. When encountering serious faults such as servo system swinging and oil injection, in order to ensure the safety of on-board equipment and personnel, the power can be directly cut off. Since such problems occur suddenly and have serious effects, the power supply of the on-board and ground equipment must be cut off immediately. Multimedia equipment should be used as much as possible to record the entire system test process to provide a reliable basis for analyzing and handling problems.

[0016] The common steps for handling non-safety faults are: 1) Protect the site and freeze the status; fully record the fault phenomenon; call relevant personnel for preliminary analysis and unified description. 2) Reproduce the fault according to the situation. 3) According to the fault phenomenon, reproduction results, and analysis results, formulate a treatment plan and implement troubleshooting. Non-safety issues account for the majority of system test failures, and there is sufficient time for analysis and troubleshooting of such issues. Try to maintain the status of the onboard and ground equipment, protect the test site, do not easily change the interface and electrical connector, and do not make the fault phenomenon disappear, so as to facilitate the search and analysis of the cause of the fault.

[0017] With the development of space technology and the requirements of space missions, the upper stage electrical system will develop towards undertaking multiple space missions in the future. This poses higher requirements for the reliability of the upper stage electrical system. At the same time, the ability of the electrical system to complete relevant work under fault conditions is also becoming increasingly important. Summary of the Invention

[0018] The technical problem solved by the present invention is: overcoming the deficiencies of the prior art, and providing an open electrical intelligent health monitoring and management system for the upper stage, which has the ability to diagnose, locate, and alarm faults in the upper stage electrical system, aiming at the working conditions of long-term on-orbit multi-mission requirements of the upper stage.

[0019] The technical solution of the present invention is: an open electrical intelligent health monitoring and management system for the upper stage, which includes a health status data acquisition subsystem, an upper stage health management subsystem, and a ground health management subsystem;

[0020] The health status data acquisition subsystem converts and synchronizes the time of the health status data of each device in the electrical system, and then sends it to the upper stage health management subsystem. The device health status data includes the device operation status data collected by sensors and the electrical system control calculation data. The device operation status data includes temperature, pressure, displacement, strain, voltage and current, and magnetic field intensity;

[0021] The upper stage health management subsystem stores the health status data of each device in the electrical system into the upper stage database; performs BIT tests on the single devices in the electrical system; according to the health status data of each device in the electrical system and the BIT test results, uses an inference engine to diagnose and isolate faults for each single device respectively, and at the same time sends the BIT test results of each single device and the obtained health status data of each device in the electrical system to the ground health management subsystem;

[0022] The ground health management subsystem stores the health status data and BIT test results of each device in the electrical system; according to the health status data and BIT test results of each device in the electrical system, uses an intelligent diagnosis algorithm to comprehensively analyze and perform secondary diagnosis on the health status of the electrical system during flight, realizes accurate fault detection and location, and gives the overall health status information of the electrical system.

[0023] The beneficial effects of the present invention compared with the prior art are:

[0024] (1) The present invention adopts an open and hierarchical design method to design an overall architecture for open health monitoring and management of an upper-stage intelligent electrical system, forming a hierarchical and distributed inference mechanism of bottom-layer sensor and BIT testing, regional-level inference diagnosis, and platform-level inference diagnosis, and solving the problem of function expansion during the compatible operation of the system with the principle prototype of intelligent advanced electrical technology and other equipment.

[0025] (2) The present invention adopts health management analysis by integrating multiple algorithms. Compared with traditional health detection methods, the present invention integrates and analyzes fault diagnosis and evaluation algorithms based on signal processing, hidden Markov model, multi-signal model, and reliability analysis, etc., to detect, analyze, and locate faults in electrical system equipment, and improves the fault detection rate of the electrical system to more than 95%.

[0026] (3) In response to the requirements of fault diagnosis and prediction evaluation for the upper-stage electrical system, research and verification on algorithms such as rapid fault diagnosis of the electrical system are carried out to provide technical support for the functional implementation of health monitoring of the upper-stage electrical system. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is the overall architecture of the upper-stage electrical system according to an embodiment of the present invention;

[0028] Figure 2 is the architecture design diagram of the upper-stage health monitoring and management system according to an embodiment of the present invention;

[0029] Figure 3 is the block diagram of wavelet transform according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0030] The following further describes in detail the specific embodiments of the present invention with reference to the drawings.

[0031] 1. Architecture of the upper-stage health monitoring and management system

[0032] The upper-stage health monitoring and management system can be divided into three levels according to the spatial distribution, namely, the health status data acquisition subsystem, the upper-stage health management subsystem, and the ground health management subsystem.

[0033] The health status data acquisition subsystem includes various types of sensors (such as temperature, pressure, displacement, strain, voltage and current, magnetic field intensity, etc.), which can comprehensively collect the operation status data of each device in the electrical system; the control and calculation data of the electrical system and the operation status data of each device measured by individual sensors are used as health status data. After operations such as data conversion and time synchronization, the health status data is transmitted through the bus to the upper-stage health management subsystem;

[0034] The upper-stage health management subsystem stores the health status data of each device in the electrical system into the upper-stage database; conducts BIT tests on the single devices in the electrical system; based on the operation status data and BIT test results of each device in the electrical system, uses an inference engine to perform fault diagnosis and isolation on each single device respectively, and at the same time sends the BIT test results of each single device and the obtained health status data of each device in the electrical system to the ground health management subsystem;

[0035] The ground health management subsystem stores the health status data and BIT test results of each device in the electrical system; based on the health status data and BIT test results of each device in the electrical system, uses an intelligent diagnosis algorithm to conduct comprehensive fault detection and precise positioning on the entire electrical system, and gives the overall health status information of the electrical system.

[0036] The intelligent diagnosis and evaluation algorithm for the upper-stage electrical system work has two research contents. One is the judgment of fault symptoms, that is, it is required to have a sufficiently high resolution sensitivity to weak fault information; the other is the effective detection and precise positioning of faults. The specific research contents are as follows:

[0037] The fault detection and isolation ability, including the effective detection and precise positioning of faults, as few false alarms and missed detections as possible, accurate judgment of fault symptoms, sufficiently high resolution sensitivity to tiny fault information, and strong fault detection and isolation capabilities and high robustness;

[0038] The detection and isolation speed should be as fast as possible, that is, the delay time is short. The delay time not only means reporting to the equipment operator in advance the upcoming faults, but also is related to whether a sufficient time window can be provided for the execution of the prediction algorithm;

[0039] Sufficiently stable to background noise and working condition changes;

[0040] Set a highly reliable confidence level.

[0041] The intelligent diagnosis and evaluation algorithm includes various fault diagnosis algorithms such as signal processing-based, hidden Markov, and multi-signal model fault diagnosis algorithms. The present invention, aiming at the working principle of the electrical system, mines more detailed fault information from the operation data, discovers the hidden change trends in the data, and infers the upper-stage health assessment and life prediction based on the hidden change trends in the data, providing reliable decision support for the on-orbit maintenance and design optimization of the upper stage.

[0042] The architecture of the upper-stage health monitoring and management system is as Figure 2 shown. Each subsystem is as follows:

[0043] (1) Health status data collection subsystem

[0044] Health status data is the foundation of the entire health management, so the quality of the selected data and the quality of the data collection technology will directly affect the health prediction and diagnosis part of the entire health management system. A comprehensive health management system should have the basic functions of obtaining and integrating information from multiple system elements and compiling this information into a knowledge base of system health.

[0045] The data of the upper stage health monitoring and management system comes from various sensors and control solution data of the upper stage. After data format conversion, A / D conversion, time synchronization and other signal processing, it is sent to the upper stage health management subsystem for analysis. During the execution of the task, the sensor continuously obtains the health status data of the upper stage electrical equipment, and sends it to the knowledge database of the upper stage health management subsystem through the data transmission device for further analysis.

[0046] During the use of the health monitoring and management system, the health status data collection subsystem needs to be improved and updated in a timely manner according to usage. The improvements include adding new sensor types, optimizing sensor location distribution, and improving data collection accuracy.

[0047] 2) Onboard health management subsystem

[0048] The upper-level health management system includes two main functions: BIT and inference engine.

[0049] The working modes of the BIT module of the upper stage onboard health management system include BIT during operation and BIT before operation: BIT during operation is mainly used to obtain various performance index parameters of each device in the electrical system when it is working and store the status data in real time, and timely feedback to the inference engine; BIT before operation is mainly used to check whether the technical status of each device in the electronic system before working is normal, which corresponds to the test, launch and control process before the launch of the upper stage. The structure of the upper stage BIT is centralized, arranged on various devices of the upper stage electrical system such as switch amplifiers, fiber optic inertial groups, star sensors, etc., and is used to monitor the underlying equipment. The BIT module of the upper stage onboard health management system runs in the upper stage core processor, which controls the diagnostic procedures of each subsystem and transmits the sensor signal to the core processor.

[0050] The inference engine is composed of a group of computer function program modules, which receives the performance indicators of each device, extracts fault signs from the sensor signals output by BIT detection, and performs fault reasoning through knowledge matching. Finally, it enters the knowledge base for storage and transmits it to the ground.

[0051] When the inference engine of the upper-stage onboard health management subsystem is running, it is necessary to perform fault analysis on the continuously optimized fault diagnosis algorithm. During the verification process, according to the relationship between the fault diagnosis and prediction capabilities and the specific use guarantee tasks, the functional requirements analysis of the fault diagnosis and prediction capabilities is carried out. On the basis of considering the timeliness and accuracy requirements, the acceptable thresholds are set for the performance metrics. When the diagnosis and prediction performance metrics do not meet the threshold requirements, the verification results should be returned to the algorithm developers for targeted adjustments.

[0052] In the diagnostic and prediction performance measurement index system, different performance measurement indicators describe the ability of the algorithm from different aspects. For example, indicators such as the success detection rate, false alarm rate, and accuracy represent the correct diagnosis ability of the system algorithm; the average fault detection time represents the operating efficiency of the system algorithm, and the average fault detection time has greater reference value in tasks with real-time requirements; working condition sensitivity and noise sensitivity represent the adaptability of the system algorithm to different usage environments, and have a higher reference value in complex tasks. Therefore, when the indicator does not meet the threshold requirements and the system algorithm needs to be adjusted, it is necessary to design and adjust the strategy according to the specific indicators and specific tasks.

[0053] 3) Ground health management subsystem

[0054] The ground health management system module is required to have the ability to comprehensively analyze and conduct secondary diagnosis of the health status data of the flight process, and to formulate targeted maintenance strategies based on the fault and health status information of each flight mission to support information-based maintenance and guarantee work. Based on the above-mentioned platform data analysis results and maintenance evaluation results, it completes the health status assessment of the entire aircraft, gives recommendations on whether to enter the next flight mission, and assists commanders in making decisions.

[0055] The process management system in the ground health management subsystem can optimize the fault diagnosis method of the entire system and improve the fault diagnosis capability by continuously updating or adding new modules or fault information; the ground health management subsystem also has the ability to exchange and integrate information with other systems (such as comprehensive maintenance information of existing equipment), and can continuously obtain fault information data from the outside world. The main functions of the ground health management subsystem include five parts, namely, aircraft-ground collaborative health management, daily inspection and maintenance, pre-launch test decision-making, health management system maintenance, and process management. The functions cover the entire process of pre-launch testing, aircraft-ground collaboration during flight, and maintenance support after return.

[0056] ① Aircraft-local collaborative health management

[0057] During the period from launch to return to the ground, the upper stage health management subsystem completes data acquisition, data processing, status monitoring and health assessment, transmits the status information to the ground, and the ground health management system predicts the health status.

[0058] The memory of the upper stage on-board health management subsystem imports the data obtained by the inference engine calculation into the ground health management system for comprehensive analysis, and then iteratively optimizes the diagnostic algorithm of the upper stage on-board health management subsystem according to the health detection results of the comprehensive analysis of the information.

[0059] ② Daily detection and maintenance

[0060] After returning to the ground and before the next mission, a large amount of detection and maintenance work is required. One of the important functions of the ground health management subsystem is the daily detection and maintenance function, which needs to realize the secondary diagnostic analysis of flight process data and faults, and based on this, put forward suggestions for this overhaul and test to improve the efficiency of overhaul and test. And analyze and manage the data of overhaul and test to form a complete health management database.

[0061] ③ Pre-launch test decision-making

[0062] The pre-launch test decision-making system conducts full-system fault diagnosis during the upper stage test launch section and puts forward decision-making information according to the diagnosis results. Conducting measurement point detection, fault diagnosis before the upper stage launch, and giving decision-making information for launch are effective means to reduce launch accidents and improve launch safety.

[0063] The full-system test and health assessment functions before launch are the core functions of the ground subsystem, which realize the comprehensive analysis and quantitative assessment of the health status of the whole machine, and give suggestions on whether to enter the next flight, providing decision-making reference for launch commanders.

[0064] ④ Health management system maintenance

[0065] The maintenance of the health management system is the basis for ensuring the normal functioning of the health management system. Due to the increase in the number of flights and the continuous increase in the number of repairs, various data volumes are continuously accumulated, and the health management system needs to be optimized and improved. At the same time, with the in-depth understanding of problems and technologies, the model library, knowledge base, and decision-making library inevitably face continuous maintenance and upgrading, enabling system engineers to conveniently maintain and upgrade this system and continuously strengthen the capabilities of the health management system.

[0066] ⑤ Process management module

[0067] During the process of realizing the functions of the upper stage health monitoring and management system, it needs to interact with other systems frequently, and the system itself has a complex functional structure, a large amount of data, and covers the entire mission profile. Therefore, a process management module is required to monitor the system process, manage the tasks uniformly, schedule according to task requirements, and overall manage the system resources to achieve the efficient operation of the upper stage health monitoring and management system.

[0068] 2. Electrical System Quick Fault Diagnosis and Health Assessment Algorithm

[0069] The upper-stage electrical system includes control equipment, navigation and guidance equipment, measurement and communication equipment, power supply and power distribution equipment;

[0070] For the control equipment and navigation and guidance equipment, a fault diagnosis algorithm based on the hidden Markov model is used for fault diagnosis;

[0071] For the measurement and communication equipment, a fault diagnosis algorithm based on signal processing is used for fault diagnosis, and the fault diagnosis algorithm based on signal processing includes time-frequency analysis method;

[0072] For the power supply and power distribution equipment, a fault diagnosis and health assessment algorithm based on the Gaussian mixture model or a fault diagnosis and health assessment algorithm based on the deep sequential network is used for fault diagnosis.

[0073] 2.1 Fault Diagnosis Algorithm Based on Markov Model

[0074] (1) Data preprocessing

[0075] Perform data standardization processing on the health status data obtained from the control equipment and navigation and guidance equipment, that is, process all data into a unified data format;

[0076] Since different parameters have different dimensions and dimension units, such a situation will affect the results of data analysis. In order to eliminate the dimension influence between parameters, data standardization is required, and the original data in this algorithm is processed by data standardization. The data standardization processing method is: process all data into a unified data format. When processing all original data, the frame header is 2 bytes (1C1D), the data identification bit is 4 bytes (the first byte represents the subsystem that sends the signal, the second byte represents the subsystem that receives the signal, and the third and fourth bytes both represent the frame type), the data length is 2 bytes (determined by the length of the data content in the next column), the data content is X bytes (determined by the transmitted data content, represented by 16-bit hexadecimal data), the check code is 2 bytes (Y = X16 + X12 + X5 + 1, the initial value is 1), and the frame tail is 2 bytes (C3C3).

[0077] (2) Data smoothing processing

[0078] Smooth the health status data after standardization processing by using the exponential moving average smoothing method;

[0079] After the original data is standardized, although its changing trend can be seen, there are still some problems in the data that have a great impact on the construction of the model and the guarantee of accuracy. Among them, the mutation of data points and the change of data trend in the original data will have a great impact on the health assessment work. Therefore, the exponential moving average smoothing method is adopted in the present invention for parameter smoothing processing.

[0080] (3), Feature extraction

[0081] Due to the large number of original parameter data, there will inevitably be coupling phenomena among the parameters, and the characterization capabilities of different parameters for the health state are also different. If the original parameter data is directly used for model construction, it will not only lead to an excessive model complexity but also a situation where the model cannot be fitted. Therefore, for the health state data after smoothing processing, the PCA principal component analysis method is selected for feature extraction to obtain the feature parameter set after PCA dimensionality reduction.

[0082] (4), Fault preprocessing

[0083] The feature parameter set after PCA dimensionality reduction is used to perform a threshold judgment method to initially judge the control equipment and the navigation and guidance equipment, and estimate the health state of the control equipment and the navigation and guidance equipment. The estimated health state includes three types: normal working state, control equipment failure, and navigation and guidance equipment failure;

[0084] (5), Health state assessment

[0085] Based on the health state estimation results of the control equipment and the navigation and guidance equipment, the feature parameter set after PCA dimensionality reduction is sent into the hidden Markov models in different health states to obtain the possible health states in the current state, and the health state assessment is completed.

[0086] The hidden Markov model library in different health states includes the hidden Markov model library in the normal working state, the hidden Markov model library after the control equipment fails, and the hidden Markov model library after the navigation and guidance equipment fails.

[0087] The present invention uses the feature parameter set after PCA dimensionality reduction to construct a hidden Markov model and perform parameter training to construct a hidden Markov model library in different health states.

[0088] In the present invention, due to the limitation of the original data volume, white noise is added to the original data to construct a test data set to complete the verification of the algorithm.

[0089] 2.2, Fault diagnosis algorithm based on signal processing

[0090] Non-stationary signals generally have the characteristics that their amplitudes change particularly significantly, and the frequency diagrams obtained after performing spectral transformation on non-stationary signals do not have obvious periodic characteristic frequencies. As is well known, there are a large number of non-stationary components such as offsets, trends, and mutations in the vibration signals generated in actual working conditions, and the important characteristics of the signals are often reflected by these non-stationary components. Therefore, an analysis method that can simultaneously have time resolution and frequency domain resolution becomes particularly important. Therefore, the fault diagnosis algorithm based on signal processing in the present invention refers to the time-frequency analysis method.

[0091] Wavelet transform is a time-frequency analysis method in which both the time window and the frequency window can be changed. It has high frequency resolution and low time resolution in the low-frequency part, and high time resolution and low frequency resolution in the high-frequency part. Therefore, wavelet transform has strong ability to represent the local characteristics of signals in both the time domain and the frequency domain. Through wavelet transform, the signal can be analyzed with different scales from coarse to fine in a multi-resolution manner, which is a typical characteristic of wavelet transform, and the quality factor is constant at different resolutions. Therefore, wavelet transform is also known as the mathematical microscope; another characteristic is its ability to highlight the local characteristics of the signal in the time domain and frequency domain, so it is used as a good tool in detecting signal transients and edges.

[0092] The core idea of wavelet transform is to first decompose the original signal into an approximation signal and a detail signal, and then continue to decompose them into an approximation signal and a detail signal. And so on, finally, it can be decomposed into approximation signals and detail signals of n layers. In order to improve the time resolution of the decomposed low-frequency and high-frequency signals to the time resolution of the original signal, the decomposed signal is reconstructed, and the inverse process of decomposition is actually its reconstruction process. The low-frequency part of the signal after wavelet decomposition is reflected in the approximation signal, and the high-frequency part of the signal is reflected in the detail signal. In vibration signals, the characteristics of the signal itself are manifested in the low-frequency part, while the subtle differences of the signal are manifested in the high-frequency part. As Figure 3 shown.

[0093] The method based on signal processing avoids the difficulty of extracting the mathematical model of the object, and the fault diagnosis of electrical equipment based on wavelet transform can provide local descriptions of the signal in the time and frequency domains by using wavelet transform, so as to analyze non-stationary signals. Wavelet analysis can also divide the spectrum of the interference signal after adding noise into a useful frequency part and an interference part, so as to effectively suppress noise. It can effectively extract fault characteristics and achieve effective and reliable on-line fault diagnosis.

[0094] Since the local changes involved in the continuous wavelet transform are variable, higher time resolution is manifested in the high-frequency part, while higher frequency resolution is manifested in the low-frequency part, that is, the wavelet transform has the characteristic of "zooming". All this characteristic makes the wavelet transform particularly suitable when processing sudden signal.

[0095] At the same time, due to the good video localization characteristics, wavelet analysis can accurately present the characteristics of dynamic signals, showing obvious advantages in the analysis of dynamic signals, and is suitable for online fault diagnosis of communication measurement equipment.

[0096] 2.3 Fault diagnosis and health assessment algorithm based on Gaussian mixture model

[0097] Based on the fault diagnosis and health assessment of the upper-level electrical system, the performance parameters of the electrical system are used as input to establish a Gaussian mixture model to obtain the health of the power supply and distribution equipment. The overall health assessment scheme is shown in the figure above, and the specific implementation process of the algorithm is described below.

[0098] (1) Obtaining power supply performance parameter data

[0099] For upper-stage power supply and distribution equipment, degradation-sensitive parameters mainly include signals such as the maximum discharge capacity of the power supply and the cut-off voltage of the power battery pack. According to the built-in data interface of the software, the required parameter data is read from the historical parameter database or simulation model, parsed according to the protocol, and converted into the standard data format mentioned above.

[0100] (2) Data preprocessing

[0101] Since the upper stage power supply will be affected by the environment, operating condition fluctuations, measurement errors and other interferences during operation, the actual parameter data usually contains a large number of interference factors such as wild values ​​and noise, which affect the training convergence and evaluation accuracy of the health assessment model. Therefore, before using the data, it is necessary to perform certain preprocessing operations on the actual parameter data. Statistical criteria such as the Wright criterion, the Nair criterion, and the Grubbs criterion are used to remove wild value points in the original data. If the statistical index of the data point exceeds the threshold range required by the above criteria, it is considered that the data point belongs to the wild value and should be removed or the five-point smoothing method, linear angle smoothing method, positive axis parabola weighted average method, oblique axis parabola weighted average method, local weighted regression scatter point smoothing method (LOWESS) and other methods are used to smooth and reduce the noise of the parameter data to remove the local noise or interference contained in the data. In addition, in order to adapt to the performance characteristics of the Gaussian mixture model, the parameter data needs to be normalized.

[0102] (3) Using the degradation-sensitive parameter after the current normalization processing as the data sample, establish the first GMM model, and calculate the overlap degree between the first GMM model and the second GMM model established by using the degradation-sensitive parameter in the normal state as the data sample;

[0103] The overlap degree function of the two GMM models is as follows:

[0104]

[0105] Among them, g1(x) is the density distribution function of the first GMM model;

[0106] g2(x) is the density distribution function of the second GMM model, and x is the operating parameter of the device.

[0107] (4) Distance normalization to health

[0108] After obtaining the overlap degree calculation result, the relative magnitude of the distance metric result reflects the health status of the upper-level power supply, but its absolute value cannot characterize its health degree. Therefore, a normalization method is needed to map the overlap degree to the 0-1 interval and convert it into a health CV value to quantitatively characterize the health degree of the power supply.

[0109] The following is the normalization formula using the arctangent function to normalize the overlap degree to obtain the CV value.

[0110]

[0111] Among them, overlap is the overlap degree of the two GMM models, and a is the maximum floating value of the error parameter.

[0112] (5) Evaluate the health status of the power supply and distribution equipment by the health degree. The closer the health degree value is to 1, the healthier the power supply and distribution equipment is, and the closer the health value is to 0, the less healthy the power supply and distribution equipment is.

[0113] The construction process of the above Gaussian mixture model is as follows:

[0114] The performance parameters of the upper-level power supply are multi-dimensional features with a complex statistical distribution, and it is difficult to effectively fit with a single statistical distribution. The method of the Gaussian mixture model GMM is a linear combination of multiple Gaussian distribution functions, which can achieve effective fitting of any type of distribution and is usually used to solve the situation where the data in the same set contains multiple different distributions. Therefore, for the health assessment of power distribution equipment, a deviation measurement method based on the Gaussian mixture model is adopted. After obtaining the performance parameters of typical components after preprocessing, the parameter vector composed of the performance parameters at each moment is used as the feature, and the EM algorithm introduced in the algorithm principle is used to estimate the parameters in the GMM model. The parameter space composed of the estimated model parameters is used as the high-dimensional space in the healthy state of the key components.

[0115] The data to be evaluated are measured parameter data, including upper stage position and attitude data, time series data, and power output voltage and current signals. After obtaining the actual parameter data according to the data acquisition method, the same preprocessing method is used for data preprocessing, and then the feature vectors composed of performance parameters at each moment are used as inputs to estimate the parameters of the Gaussian mixture model, which is used as the high-dimensional space where the state of the data to be evaluated is located.

[0116] 2.5 Multi-signal model fault diagnosis algorithm

[0117] Before building the multi-signal model, it is necessary to clarify information such as tests / test points, the constituent units of the object under test, fault classes, and fault-test correlations in combination with the specific information of the research object, as follows:

[0118] Tests and test points: A test refers to the operation process taken to determine the performance, characteristics, or whether a system or device can work properly and effectively. Whether the unit under test (UUT, Unit under Test) is in a normal state or a fault state is determined by whether the responses to the excitations and controls used in the test process are as expected. If the expected values are reached, the UUT is considered to be in a normal working state; otherwise, it is considered to be in a fault state. Before conducting a test, it is necessary to determine the test points. The definition of a test point is any physical location where the required state information can be obtained. It should be emphasized that a test can be carried out using one or more test points. Similarly, a test point can also be used by one or more tests. In the process of building a multi-signal model for the upper stage electrical system in the present invention, a test point is only used by one test.

[0119] Constituent units of the object under test and fault classes: A constituent unit refers to the unit that needs to be replaced during repair after a fault occurs. In fact, fault diagnosis mainly focuses on the faults that occur in the constituent units. Therefore, the fault unit can be used to represent the constituent unit. If their performance characteristics are the same or similar, they are called fault classes.

[0120] Fault-test correlation: Faults and tests in a system usually have a correlation. If it is deduced from the occurrence of fault s i that test t j fails, and it can be deduced from the passing of t j that s i has not occurred, then t j and s i are mutually correlated, or t j is a symmetric test. If it can only be deduced from the occurrence of s i that t j fails, and it cannot be deduced from the passing of t j that s i has not occurred, then tj Asymmetric test

[0121] Dependency matrix: The dependency matrix (also known as the D matrix) is a matrix that reflects the relationship between faults and tests, usually denoted as:

[0122]

[0123] In the formula, the matrix element d ij is a binary variable. If the test t j can detect the fault of component s i , then make d ij = 1; otherwise, let d ij = 0;

[0124] Any row vector [d i1 , d i2 , …, d in in the D matrix T means the detection results of each test when component s i has a fault. The result of this row vector is regarded as the symptom of component s i having a fault; Any column vector [d 1j , d 2j , …, d mj in the D matrix T represents all the fault states that test t j can detect, and to a certain extent reflects the fault detection ability of test t j ;

[0125] S2. According to the types, quantities, and working characteristics of electrical system equipment, construct a multi-signal model. The multi-signal model consists of the following elements:

[0126] C = {c1, c2, …, c m}: m unit components that may have faults;

[0127] T = {t1, t2, …, t n}: A finite set of n available tests; PT is a subset of T without alarm tests, and FT is a subset of T with alarm. The mathematical meanings of PT and FT are PT = {t i |t i has no alarm, t i ∈T}, FT = {t j |t j has an alarm, t j ∈T}, T = PT ∪ FT (i = 1, 2, …, m; j = 1, 2, …, n);

[0128] D = [dij :Correlation matrix between electronic system unit components and tests, d ij = 1 indicates that if the electronic system unit component s i fails, then the test t j confirms an alarm; while d ij = 0 means that the failure of the electronic system unit component s i cannot be detected by the test t j ;

[0129] C(t i ): Set of finite elements that can be detected by the test t i ; Each unit or component is associated with four different states: normal (Good), failed (Bad), suspected (Suspected), and unknown (Unknown);

[0130] Initially, the state of all units or components is unknown. If the test detects that a unit or component is normal, the corresponding unit or component state will be updated to normal; otherwise, the state of these units or components will be set to suspected.

[0131] S3. After constructing the multi-signal model, the diagnostic reasoning based on the multi-signal model can be implemented through the following algorithm:

[0132] For t i ∈ PT, G ← U iε∈PT C(t i );

[0133] For t j ∈ FT, the set S is equal to the limit value of C(t j ) - G, denoted by {s j}, since the limit value of the set {s j} tends to zero, so the union of the set Su and {s j} finally tends to itself, as shown in the following formula:

[0134] S = {s j} ← C(t j ), Su ← Su ∪ {s j};

[0135] If |{s j}| = 1, then B ← B ∪ (s j ), Su ← Su - B

[0136] where ← represents updating the set state, S = {s j} is from the set C(t iThe set of suspicious units or components after deleting all normal elements. G represents the set of normal components, Su represents the set of suspected components, and B represents the set of faulty components. |{s j}| is the cardinality of {s j}.

[0137] Intelligent Monitoring and Management of Electrical Systems Based on Information Fusion

[0138] In view of the characteristics of the faults of upper-level electrical equipment being spreadable and the monitoring parameter types being diverse, combined with various fault diagnosis algorithms such as signal processing, hidden Markov, and multi-information models, an ARIMA information fusion computational intelligent monitoring and management algorithm for electrical systems is studied.

[0139] The present invention also studies a method for predicting electrical system faults based on weighted fusion ARIMA. According to the actual data characteristics of the upper-level electrical system, based on the sensitive parameters of the electrical system, by virtue of the robust prediction performance of the weighted fusion algorithm and the uncertainty management ability of the statistical time series modeling algorithm, multi-ARIMA models are used to carry out fusion prediction of electrical system equipment, so as to achieve the suppression of the degradation differences of electrical equipment and provide a fault interval prediction with probabilistic significance, and further realize the extrapolation prediction of the power failure occurrence time with good interpretability.

[0140] After the multi-ARIMA model obtains the fault information of the power system, it is necessary to carry out information fusion technology to analyze the obtained data. Let the information data set obtained by using the multi-ARIMA model be S i , and the credibility data and plausibility data obtained during the information acquisition process are A i , B i respectively, and the corresponding credibility function and plausibility function are m(A i ) and m(B i ). The average support degree of the information data of time, upper-level position and attitude, navigation, etc. obtained by the i-th ARIMA model for the state variable attributes of the electrical system is shown as follows.

[0141]

[0142] In the formula, J(A l ) represents the probability of the model analyzing the credibility of the real data.

[0143] Therefore, the uncertain entropy introduced by the information data obtained by the i-th ARIMA model is defined as follows

[0144]

[0145] The larger the obtained uncertain entropy value is, it proves that the unknown degree of the data information corresponding to the uncertain entropy is greater. Accordingly, the credibility expressions of the information data obtained by multiple ARIMA models can be given as follows

[0146]

[0147] The value α represents the timely credibility data of the information data, which can be obtained through the wavelet transform mentioned above and the time-frequency calculation.

[0148] When performing information data analysis and fusion, the weighted coefficients are defined as follows:

[0149]

[0150] In the formula, M l represents that the information data obtained by the i-th ARIMA model (corresponding to each single machine system of the electrical system) is included in the l-th classification result.

[0151] Perform weighted averaging on all classified synthetic information to obtain the final multi-ARIMA model fusion information, which can reflect the overall health information index of the electrical system.

[0152]

[0153] In the formula is the synthetic information of the i-th ARIMA model under the l-th classification.

[0154] m F corresponds to the column vector state data information obtained from the power system detection.

[0155] Then adopt the FMECA analysis method. Its purpose is to target all possible faults of the product. According to the analysis of the fault modes, determine the impact of each fault mode on the product operation, find out the single-point faults, and determine their harmfulness according to the severity and occurrence probability of the fault modes. The FMECA analysis results can not only assist in selecting high-reliability design methods during the product design process, but also provide a basis for the product test and maintenance work planning.

[0156] When performing FMECA analysis on the upper-stage electrical equipment, the first step is to collect the basic information of the electrical equipment, including: the composition structure diagram of the electrical system; the flow of each functional profile and process step; the functions of the system components and their functional connection relationships with other components; the environmental parameters that may affect the operation; the specific fault modes and their results of the upper-stage electrical equipment.

[0157] Step 2: List the components of the system and confirm the following issues: the ways in which obvious failures occur in each part, the failure mechanisms that cause such failures, the possible impacts of the failures (no obvious impact on the system function or being harmful), and the ways of failure detection.

[0158] Step 3: Classify the failure modes. The classification criteria can adopt the failure severity, which is generally divided into four categories: Category I - Catastrophic accident: causing death or permanent total disability of personnel or irreversible serious environmental damage; Category II - Serious accident: causing permanent partial disability of personnel and reversible environmental damage; Category III - Critical accident: causing injury to personnel or occupational disease or environmental damage without violating laws or regulations; Category IV - Minor accident: causing injury to personnel or occupational disease, but the personnel can still work, or causing minor environmental damage without violating laws or regulations.

[0159] A list of failure modes, failure mechanisms, and their impacts on each component, system, or process step (which may include information on the probability of failure) can provide information on the causes of failures and their impacts on the entire system.

[0160] After constructing the FMECA, use the existing normal data and fault data of electrical equipment to construct a training set. When the collected fault data increases and the failure modes increase, the current fault diagnosis model can be retrained and updated through the fault diagnosis model to achieve the fault diagnosis of the latest faults or improve the diagnosis accuracy of the existing failure modes. When performing fault diagnosis, the corresponding signals should be selected. When it is determined that the current electrical equipment has a fault or the current state of the electrical equipment cannot be clearly analyzed through preliminary analysis, the fault diagnosis method using a convolutional neural network can be used to clarify the failure mode of the current state.

[0161] The technical solutions described in detail above are common knowledge in the art.

Claims

1. An upper stage open electrical intelligent health monitoring and management system, characterized in that It includes a health status data acquisition subsystem, an upper-level health management subsystem, and a ground health management subsystem; The health status data acquisition subsystem converts and synchronizes the time of the health status data of each device in the electrical system and then sends it to the upper-level health management subsystem. The device health status data includes the device operation status data collected by sensors and the electrical system control solution data. The device operation status data includes temperature, pressure, displacement, strain, voltage and current, and magnetic field strength; The upper-level health management subsystem stores the health status data of each device in the electrical system in the upper-level database; performs BIT tests on the single devices in the electrical system; according to the health status data and BIT test results of each device in the electrical system, uses an inference engine to perform fault diagnosis and isolation on each single device respectively, and at the same time sends the BIT test results of each single device and the obtained health status data of each device in the electrical system to the ground health management subsystem; The ground health management subsystem stores the health status data and BIT test results of each device in the electrical system; according to the health status data and BIT test results of each device in the electrical system, uses an intelligent diagnosis algorithm to comprehensively analyze and perform secondary diagnosis on the health status of the electrical system during flight, achieve accurate fault detection and location, and give the overall health status information of the electrical system; The upper-level electrical system includes control devices, navigation and guidance devices, measurement and communication devices, and power supply and distribution devices; For control devices and navigation and guidance devices, the ground health management subsystem uses a fault diagnosis algorithm based on the hidden Markov model for fault diagnosis; For measurement and communication devices, the ground health management subsystem uses a fault diagnosis algorithm based on signal processing for fault diagnosis; For power supply and distribution devices, a fault diagnosis algorithm based on the Gaussian mixture model is used for fault diagnosis; The steps of the fault diagnosis algorithm based on the hidden Markov model are as follows: S1.

1. Standardize the health status data obtained for control devices and navigation and guidance devices, that is, process all data into a unified data format; S1.

2. Smooth the health status data after standardization processing using the exponential moving average smoothing method; S1.

3. For the health status data after smoothing processing, select the PCA principal component analysis method for feature extraction to obtain the feature parameter set after PCA dimensionality reduction; S1.

4. Use the method of threshold judgment on the feature parameter set after PCA dimensionality reduction to initially judge the control devices and navigation and guidance devices, and estimate the health status of the control devices and navigation and guidance devices. The estimated health status includes three types: normal working state, control device failure, and navigation and guidance device failure; S1.

5. According to the health status estimation results of the control devices and navigation and guidance devices, send the feature parameter set after PCA dimensionality reduction into the hidden Markov models under different health states to obtain the possible health states in the current state, and complete the health status assessment; The hidden Markov models under different health states include the hidden Markov model under normal working conditions, the hidden Markov model after the control device fails, and the hidden Markov model after the navigation and guidance device fails.

2. The upper stage open electrical intelligent health monitoring and management system according to claim 1, characterized in that The fault diagnosis algorithm based on signal processing refers to the time-frequency analysis method.

3. The upper stage open electrical intelligent health monitoring and management system according to claim 1, characterized in that The fault diagnosis and health assessment algorithm based on the Gaussian mixture model includes the following steps: S2.

1. Obtain the degradation-sensitive parameters of the power supply and distribution equipment. The degradation-sensitive parameters include the maximum discharge capacity and cut-off voltage of the power supply. S2.

2. Perform smoothing, noise reduction, and normalization processing on the degradation-sensitive parameters of the power supply and distribution equipment. S2.

3. Using the currently normalized degradation-sensitive parameters as data samples, establish a first GMM model, and calculate the overlap degree between the first GMM model and a second GMM model established using the degradation-sensitive parameters under normal conditions as data samples. S2.

4. Normalize the overlap degree between the first GMM model and the second GMM model, map the overlap degree to the 0-1 interval, and obtain the health degree of the power supply and distribution equipment. S2.

5. Evaluate the health state of the power supply and distribution equipment based on the health degree. The closer the health degree value is to 1, the healthier the power supply and distribution equipment is, and the closer the health value is to 0, the less healthy the power supply and distribution equipment is.

4. The upper stage open electrical intelligent health monitoring and management system according to claim 3, characterized in that The smoothing and noise reduction step in step S2.2 includes: using statistical criterion indicators such as the Wright criterion, Neill criterion, or Grubbs criterion to eliminate outliers in the original data, or using methods such as the five-point smoothing method, linear chamfering method, positive-axis parabola weighted average method, oblique-axis parabola weighted average method, and locally weighted regression scatter point smoothing method to perform smoothing and noise reduction on the degradation-sensitive parameters to remove the noise or interference contained in the data.

5. The upper stage open electrical intelligent health monitoring and management system according to claim 1, characterized in that The intelligent diagnosis algorithm also includes a multi-signal model fault diagnosis algorithm.

Images