Fault prediction method based on aviation power supply

By using the SVM model and expert knowledge base, combined with the power log and fault feature vector, the problem of insufficient accuracy in identifying power failure types is solved, and the accuracy of power failure prediction and system stability is improved.

CN119004232APending Publication Date: 2024-11-22ECU ELECTRONICS INDAL

Patent Information

Application Number
CN202411093628.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-09
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

In the prior art, the ELM-CHMM model has poor effect in identifying the actual power failure type, resulting in a decrease in the accuracy of power failure prediction.

Method used

By obtaining the fault feature vector from the database, preprocessing it and building a data set, using the SVM model for training, establishing a fault prediction model, and building a rule base with power logs and expert knowledge to determine the component failure type.

Benefits of technology

It improves the accuracy of power failure prediction and overall system reliability, reduces economic losses and repair costs caused by faults, and ensures the stability and safety of the power supply system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119004232A_ABST
    Figure CN119004232A_ABST
Patent Text Reader

Abstract

The invention discloses a fault prediction method based on an aviation power supply, relates to the technical field of aviation power supply intelligent management, and solves the technical problem that the accuracy of power supply fault prediction is reduced due to the fact that power supply faults are difficult to comprehensively identify when fault prediction is carried out on a power supply through an ELM-CHMM model in the prior art. According to the invention, multiple sensors are arranged to collect measured values; fitting processing is carried out on the measured value through a recursive average filtering method; obtaining a plurality of types of fault feature vectors from a database, preprocessing the fault feature vectors, and constructing a data set; training the SVM model by using the data set to obtain a fault prediction model; acquiring use information of a plurality of elements in the power supply and a power supply log from a database; judging whether the corresponding element has a fault according to the use information, and predicting a possible fault type of the corresponding element according to a fault prediction model; and the accuracy of power supply fault prediction is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of intelligent management of aviation power supplies, relates to intelligent management technology, and specifically is a fault prediction method based on aviation power supplies. Background Art

[0002] By real-time monitoring of the status of the aviation power system, predicting and preventing the occurrence of potential faults, aviation accidents caused by power system failures can be effectively avoided, thereby ensuring the safety of passengers and crew members. When it is predicted that a failure is about to occur, the system can promptly warn the crew, giving them enough time to take countermeasures, such as adjusting the flight plan, finding an alternate airport, etc., to ensure the safe landing of the aircraft. Methods based on fault prediction can detect potential problems in advance, thereby avoiding unplanned maintenance caused by sudden failures of the power system and reducing economic losses caused by grounding for maintenance. Through continuous monitoring and data analysis of the health status of the power system, maintenance plans can be formulated more accurately, maintenance can be carried out on demand, and excessive or insufficient maintenance can be avoided, thereby reducing maintenance costs. Fault prediction methods based on big data and machine learning can provide more accurate and comprehensive data support, making maintenance decisions more scientific and reasonable. Future fault prediction methods will tend to be multi-model fusion, combining the advantages of multiple prediction models to improve the accuracy and reliability of fault prediction, thereby further optimizing maintenance strategies.

[0003] The prior art (invention patent application with publication number CN109615003A) discloses a power supply fault prediction method based on ELM-CHMM; the prior art divides voltage signal data into training and test data sets, preprocesses the training data set, and obtains a reconstructed voltage signal matrix; establishes an ELM model; inputs the test data set into the ELM model, and outputs the voltage signal predicted by the ELM model; extracts the characteristic parameters of the training data set; establishes a CHMM state prediction model; extracts the characteristic parameters of the voltage signal data predicted by the ELM model, and inputs them into the CHMM model respectively; obtains the ELM-CHMM model, obtains the state of the power supply fault to be tested, and thus predicts the possible faults of the power supply; the prior art predicts the fault of the power supply through the ELM-CHMM model, and the ELM-CHMM model has a good effect on identifying the fault type that occurs in the voltage signal of the power supply, but the actual power supply fault types are diverse, and the ELM-CHMM model is difficult to fully identify the power supply fault, resulting in the problem of reduced accuracy of power supply fault prediction.

[0004] The present invention provides a fault prediction method based on aviation power supply to solve the above technical problems. Summary of the invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a fault prediction method based on an aviation power supply, which is used in the prior art to predict the fault of the power supply through the ELM-CHMM model. The ELM-CHMM model has a good effect on identifying the fault type occurring in the voltage signal of the power supply, but the actual power supply fault types are various, and the ELM-CHMM model is difficult to fully identify the power supply fault, resulting in a technical problem that the accuracy of power supply fault prediction is reduced; the present invention obtains several types of fault feature vectors from a database, constructs a data set through the fault feature vectors; and uses the data set to train the SVM model to obtain a fault prediction model to solve the above technical problems.

[0006] To achieve the above object, a first aspect of the present invention provides a fault prediction method based on an aviation power supply, comprising:

[0007] S1: Set up multiple sensors to collect measurement values; perform fitting processing on the measurement values ​​by recursive average filtering method; wherein the measurement values ​​include: voltage, current, and temperature;

[0008] S2: Obtain several types of fault feature vectors from the database and preprocess the fault feature vectors; wherein the preprocessing includes: normalization, dimensionality reduction, and filtering;

[0009] S3: retrieve the fault feature vector to construct a data set, and use the data set to train the SVM model to obtain a fault prediction model;

[0010] S4: Obtain usage information of several components in the power supply and the power supply log from the database; determine whether the corresponding component has a fault based on the usage information, and if so, arrange for a technician to repair it; if not, retrieve the fault prediction model and predict the possible fault type of the corresponding component based on the fault prediction model.

[0011] Preferably, the fitting process of the measured values ​​by the recursive average filtering method includes:

[0012] Retrieve the measured values, integrate them into a queue, mark the queue length as N, put the newly collected data at the end of the queue, and remove the measured values ​​at the head of the queue with the same number;

[0013] The noise is marked as n(i), and the actual value is marked as s(i); wherein i represents the position of the measured value in the queue, i=1, 2, ..., N; N is a positive integer;

[0014] The measured values ​​within a sampling period are obtained by calculating the formula y(i)=s(i)+n(i);

[0015] By formula Calculate the average of the measured values;

[0016] By formula Calculate the error;

[0017] By formula Calculate the minimum value of the sum of squared errors;

[0018] By formula Calculate the extreme value;

[0019] Take the average value of N measurements, merge the average values ​​of the measurements using the least squares method, and fit the data curve.

[0020] It should be noted that when solving the average value, the larger N is, the better the signal smoothing is.

[0021] The present invention can effectively reduce or even remove random interference by taking the average value of N measurement values, merge the measurement values ​​using the least square method, and fit the data curve to gradually approach the real state; the recursive average filtering method is used to calculate the average value of the measurement values ​​over a period of time, which can effectively remove random noise and short-term fluctuations in the signal, making the signal smoother, which is very important for improving the signal-to-noise ratio of the signal and the accuracy of subsequent processing; while removing noise, the recursive average filtering method can also retain the main characteristics and trends of the signal, avoiding signal distortion.

[0022] Preferably, the method of merging the average values ​​of the measured values ​​by using the least square method comprises:

[0023] Take the average value of N measurements and mark it as x, and mark the actual value as y i , mark the error value sum of square function as φ, and mark the number of test samples as m;

[0024] The true value is calculated by the formula y=a0+a1x; where a0 and a1 are arbitrary real numbers;

[0025] By the formula e=y i -y calculates the error value;

[0026] By the formula φ=∑(y i -a0-a1x) 2 Calculate the error value square sum function;

[0027] The sum of squares of the error values ​​is used as the minimum optimization criterion;

[0028] By formula Calculate the partial derivative of the function φ with respect to a0;

[0029] By formula Calculate the partial derivative of the function φ with respect to a1;

[0030] The solution is:

[0031]

[0032] Where m is an integer greater than 0, and xi and yi are any set of experimental values.

[0033] The present invention uses the least square method to obtain an optimization method for the best data matching function by minimizing the sum of squares of errors; by taking the average value of N measurement values, the measurement error caused by random errors can be significantly reduced, because random errors usually offset each other in multiple measurements, while systematic errors remain unchanged, so the average value is closer to the true value; using the average value as the basis for subsequent processing can ensure that the data input into the least square method has a higher quality, thereby improving the reliability and accuracy of the entire data processing flow.

[0034] Preferably, the method of using the data set to train the SVM model to obtain the fault prediction model includes:

[0035] Retrieve the data set, classify the data set and mark the labels; train the classification model through the machine learning classification algorithm; use the newly collected fault feature vector as the input of the classification model, and output the prediction result as the specific fault category; obtain the fault prediction model through the support vector machine classifier.

[0036] The present invention obtains a fault prediction model by training the SVM model using a data set. SVM performs well in processing high-dimensional data and complex classification problems, and can effectively distinguish fault characteristics of different categories, thereby achieving accurate fault identification; an accurate fault prediction model can help the system to promptly detect and handle potential fault problems, thereby avoiding the expansion of faults to cause more serious damage to the system, and helping to improve the overall reliability and stability of the system.

[0037] Preferably, judging whether a corresponding component fails according to the usage information includes:

[0038] A rule base is constructed by collecting expert knowledge; usage information is retrieved, and whether the usage information complies with the rule information in the rule base is determined by matching the usage information with the rule information; if yes, the corresponding component does not have a fault; if no, the corresponding component does have a fault.

[0039] The present invention constructs a rule base by collecting expert knowledge, which can be easily modified and expanded as needed. As new expert knowledge is discovered or old knowledge is eliminated, the rule base can be updated accordingly to adapt to changing needs and environments. This flexibility enables the system to continuously improve and maintain its effectiveness. Since the rule base is built based on expert knowledge, it usually has higher reliability when dealing with known and predictable types of failures.

[0040] Preferably, predicting the possible fault type of the corresponding component according to the fault prediction model includes:

[0041] Retrieve the fault prediction model, analyze the power log using the fault prediction model, and mark the peak value measured in the corresponding component power log as H i , mark the trough value adjacent to the peak value as L i ;Get the fluctuation threshold from the database;

[0042] By formula H i -L i =B i Calculate the fluctuation value in the power log of the corresponding component;

[0043] Determine whether the fluctuation value in the power supply log of the corresponding component is greater than the fluctuation threshold; if yes, it is determined to be an abnormal fluctuation phenomenon; if not, continue monitoring; predict the remaining life of the capacitor device based on the abnormal fluctuation phenomenon in the measured value.

[0044] It should be noted that the life of a capacitor device is related to current, temperature and voltage values.

[0045] The present invention can timely discover and solve potential power supply problems by real-time monitoring of fluctuation values ​​in the power supply log and setting fluctuation thresholds to determine whether the power supply is in normal working condition; through real-time monitoring and prediction of power supply and component status, enterprises can more reasonably allocate maintenance resources; by predicting the remaining life of capacitor components, maintenance personnel and technical support can be arranged in advance to ensure that replacement work is completed before component failure, which can not only reduce production interruptions caused by equipment failures, but also improve the efficiency and quality of maintenance work.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] 1. The present invention performs fitting processing on the measured values ​​through the recursive average filtering method, and uses the least square method to merge the average values ​​of the measured values; since the calculation of the recursive average filtering method is relatively simple, only the arithmetic average operation is required, the calculation amount is small, and it is suitable for systems with high real-time requirements; and the recursive average filtering method can significantly reduce the random noise and short-term fluctuations in the signal by continuously taking N sampling values ​​for arithmetic averaging, making the signal smoother, and can further ensure the fitting accuracy and consistency of the data, which is crucial to improving the signal-to-noise ratio of the signal and the accuracy of subsequent processing.

[0048] 2. The present invention obtains a fault prediction model by training the SVM model using a data set. The SVM performs well in processing high-dimensional data and complex classification problems, and can effectively distinguish the fault characteristics of different categories, thereby achieving accurate fault identification; by collecting expert knowledge to build a rule base, it can be easily modified and expanded as needed. As new expert knowledge is discovered or old knowledge is eliminated, the rule base can be updated accordingly to adapt to the ever-changing needs and environment; by real-time monitoring of the fluctuation value in the power supply log and setting the fluctuation threshold to determine whether the power supply is in normal working condition, potential power supply problems can be discovered and resolved in time. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 A schematic diagram of the overall fault prediction scheme of the present invention;

[0051] Figure 2 A schematic diagram of a fault prediction model constructed for the present invention;

[0052] Figure 3 It is a schematic diagram of the fault prediction process of the present invention. DETAILED DESCRIPTION

[0053] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] See also Figure 1 The first embodiment of the present invention provides a fault prediction method based on an aviation power supply, comprising:

[0055] S1: Set up multiple sensors to collect measurement values; perform fitting processing on the measurement values ​​by recursive average filtering method; wherein the measurement values ​​include: voltage, current, and temperature;

[0056] S2: obtaining several types of fault feature vectors from the database and preprocessing the fault feature vectors; wherein the preprocessing includes: normalization, dimension reduction, and filtering;

[0057] S3: retrieve the fault feature vector to construct a data set, and use the data set to train the SVM model to obtain a fault prediction model;

[0058] S4: Obtain usage information of several components in the power supply and the power supply log from the database; determine whether the corresponding component has a fault based on the usage information, and if so, arrange for a technician to repair it; if not, retrieve the fault prediction model and predict the possible fault type of the corresponding component based on the fault prediction model.

[0059] See also Figure 2 , build a fault prediction model, set up multiple sensors to collect measurement values, integrate the measurement values ​​into a queue, and mark the queue length as N, put the collected new data at the end of the queue, and remove the measurement values ​​at the head of the queue with the same number;

[0060] The noise is marked as n(i), and the actual value is marked as s(i); wherein i represents the position of the measured value in the queue, i=1, 2, ..., N; N is a positive integer;

[0061] The measured values ​​within a sampling period are obtained by calculating the formula y(i)=s(i)+n(i);

[0062] By formula Calculate the average of the measured values;

[0063] By formula Calculate the error;

[0064] By formula Calculate the minimum value of the sum of squared errors;

[0065] By formula Calculate the extreme value;

[0066] Take the average value of N measurements and mark it as x, merge the average value of the measurements using the least squares method, and fit the data curve; mark the measured value as y i , mark the error value sum of square function as φ, and mark the number of test samples as m;

[0067] The true value is calculated by the formula y=a0+a1x; where a0 and a1 are arbitrary real numbers;

[0068] By the formula e=y i -y calculates the error value;

[0069] By the formula φ=∑(y i -a0-a1x) 2 Calculate the error value square sum function;

[0070] The sum of squares of the error values ​​is used as the minimum optimization criterion;

[0071] By formula Calculate the partial derivative of the function φ with respect to a0;

[0072] By formula Calculate the partial derivative of the function φ with respect to a1;

[0073] The solution is:

[0074]

[0075] Wherein m is an integer greater than 0, xi and yi are any set of experimental values; several types of fault feature vectors are obtained from the database, and the fault feature vectors are preprocessed; wherein the preprocessing includes: normalization, dimensionality reduction, and filtering; the data set is retrieved, the data set is classified and labeled; the classification model is trained through a machine learning classification algorithm; the newly collected fault feature vector is used as the input of the classification model, and the output prediction result is a specific fault category; and the fault prediction model is obtained through a support vector machine classifier.

[0076] See also Figure 3 ,Fault prediction process, obtain the usage information of several components in the power supply and the power supply log from the database; build a rule base by collecting expert knowledge; retrieve the usage information, and judge whether the usage information conforms to the rule information by matching the usage information with the rule information in the rule base; if yes, the corresponding component does not have a fault; if no, the corresponding component has a fault; retrieve the fault prediction model, analyze the power supply log through the fault prediction model, and mark the peak value measured in the power supply log of the corresponding component as H i , mark the trough value adjacent to the peak value as L i ;Get the fluctuation threshold from the database;

[0077] By formula H i -L i =B i Calculate the fluctuation value in the power log of the corresponding component;

[0078] Determine whether the fluctuation value in the power supply log of the corresponding component is greater than the fluctuation threshold; if yes, it is determined to be an abnormal fluctuation phenomenon; if not, continue monitoring; predict the remaining life of the capacitor device based on the abnormal fluctuation phenomenon in the measured value.

[0079] For example, obtain the usage information of several components in the power supply and the power supply log from the database; build a rule base by collecting expert knowledge; retrieve the usage information of the power supply, where the current is 3A and the voltage is 5V; the current in the rule base is 4A and the voltage is 5V; the voltage usage information conforms to the rule information, so the corresponding voltage component does not have a fault; the current usage information does not conform to the rule information, so the corresponding current component has a fault; retrieve the fault prediction model, and analyze the power supply log through the fault prediction model. At this time, a set of current peak values ​​measured in the power supply log of the corresponding component is 4.2A, and the valley value is marked as 3.8A; obtain the fluctuation threshold of 0.5A from the database; through H i -L i =B i The calculated fluctuation value in the corresponding component power log is 0.4A;

[0080] Since 0.4A<0.5A, the fluctuation value in the corresponding component power log is less than the fluctuation threshold, and monitoring continues.

[0081] Part of the data in the above formula is calculated by removing the dimension and taking its numerical value. The formula is a formula closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0082] The working principle of the present invention is as follows: the present invention collects measurement values ​​by setting up multiple sensors; the measurement values ​​are fitted by a recursive average filtering method; wherein the measurement values ​​include: voltage, current, and temperature; several types of fault feature vectors are obtained from a database, and the fault feature vectors are preprocessed; wherein the preprocessing includes: normalization, dimensionality reduction, and filtering; the fault feature vectors are retrieved to construct a data set, and the SVM model is trained using the data set to obtain a fault prediction model; the usage information of several components in the power supply and the power supply log are obtained from the database; whether the corresponding component has a fault is determined based on the usage information, and if so, a technician is arranged to perform maintenance; if not, the fault prediction model is retrieved, and the possible fault type of the corresponding component is predicted based on the fault prediction model.

[0083] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A fault prediction method based on aviation power supply, characterized in that: S1: Set up multiple sensors to collect measurement values; perform fitting processing on the measurement values ​​by recursive average filtering method; wherein the measurement values ​​include: voltage, current, and temperature; S2: Obtain several types of fault feature vectors from the database and preprocess the fault feature vectors; wherein the preprocessing includes: normalization, dimensionality reduction, and filtering; S3: retrieve the fault feature vector to construct a data set, and use the data set to train the SVM model to obtain a fault prediction model; S4: Obtain usage information of several components in the power supply and the power supply log from the database; determine whether the corresponding component has a fault based on the usage information, and if so, arrange for a technician to repair it; if not, retrieve the fault prediction model and predict the possible fault type of the corresponding component based on the fault prediction model.

2. A fault prediction method based on aviation power supply according to claim 1, characterized in that: The fitting process of the measured value by the recursive average filtering method includes: Retrieve the measured values, integrate them into a queue, mark the queue length as N, put the newly collected data at the end of the queue, and remove the measured values ​​at the head of the queue with the same number; The noise is marked as n(i), and the actual value is marked as s(i); wherein i represents the position of the measured value in the queue, i=1, 2, ..., N; N is a positive integer; The measured values ​​within a sampling period are obtained by calculating the formula y(i)=s(i)+n(i); By formula Calculate the average of the measured values; By formula Calculate the error; By formula Calculate the minimum value of the sum of squared errors; By formula Calculate the extreme value; Take the average value of N measurements, merge the average values ​​of the measurements using the least squares method, and fit the data curve.

3. A fault prediction method based on aviation power supply according to claim 2, characterized in that: The method of combining the average values ​​of the measured values ​​by using the least square method comprises: Take the average value of N measurements and mark it as x, and mark the actual value as y i , mark the error value sum of square function as φ, and mark the number of test samples as m; The true value is calculated by the formula y=a0+a1x; where a0 and a1 are arbitrary real numbers; By the formula e=y i -y calculates the error value; By the formula φ=∑(y i -a0-a1x) 2 Calculate the error value square sum function; The sum of squares of the error values ​​is used as the minimum optimization criterion; By formula Calculate the partial derivative of the function φ with respect to a0; By formula Calculate the partial derivative of the function φ with respect to a1; The solution is: Where m is an integer greater than 0, and xi and yi are any set of experimental values.

4. A fault prediction method based on aviation power supply according to claim 1, characterized in that: The method of using the data set to train the SVM model to obtain a fault prediction model includes: Retrieve the data set, classify the data set and mark the labels; train the classification model through the machine learning classification algorithm; use the newly collected fault feature vector as the input of the classification model, and output the prediction result as the specific fault category; obtain the fault prediction model through the support vector machine classifier.

5. The fault prediction method based on aviation power supply according to claim 1 is characterized in that: The determining whether a corresponding component fails according to the usage information includes: A rule base is constructed by collecting expert knowledge; usage information is retrieved, and whether the usage information complies with the rule information in the rule base is determined by matching the usage information with the rule information; if yes, the corresponding component does not have a fault; if no, the corresponding component does have a fault.

6. A fault prediction method based on aviation power supply according to claim 1, characterized in that: The predicting of possible fault types of corresponding components according to the fault prediction model includes: Retrieve the fault prediction model, analyze the power log using the fault prediction model, and mark the peak value measured in the corresponding component power log as H i , mark the trough value adjacent to the peak value as L i ;Get the fluctuation threshold from the database; By formula H i -L i =B i Calculate the fluctuation value in the power log of the corresponding component; Determine whether the fluctuation value in the power supply log of the corresponding component is greater than the fluctuation threshold; if yes, it is determined to be an abnormal fluctuation phenomenon; if not, continue monitoring; predict the remaining life of the capacitor device based on the abnormal fluctuation phenomenon in the measured value.

Citation Information

Patent Citations

  • Switching power supply operation state on-line discrimination method

    CN110503004A

  • Lithium battery fault diagnosis method based on support vector machine and K mean value

    CN111090050A

  • Displacement sensor data processing method applied to elevator safety detection system

    CN113213297A

  • Multi-mode double-layer fault diagnosis method for fuel cell system

    CN113962259A

  • Lithium ion battery voltage abnormity sampling detection method based on fault model analysis

    CN116679216A

Cited By

  • Fault prediction method based on aviation power supply

    CN121027908A

  • Method and system for monitoring no-load state of aircraft power supply system

    CN121142386A