Aircraft air pressure reducing valve fault state evaluation method and system based on trend characterization

Abstract

The invention provides an aircraft air pressure reducing valve fault state evaluation method and system based on trend characterization, and relates to the technical field of aviation equipment fault analysis, and the method comprises the steps: S1, determining fault monitoring parameters of an aircraft air pressure reducing valve, and calculating fault monitoring parameter features and generalized life labels; s2, constructing a trend representation neural network model for predicting performance degradation of the aircraft air pressure reducing valve; s3, establishing a loss function used for optimizing the trend representation neural network model, and improving the performance degradation prediction precision of the aircraft air pressure reducing valve; and S4, determining the fault state of the aircraft air pressure reducing valve by using the trend representation neural network model. According to the method, a fault state evaluation process based on air pressure reducing valve fault monitoring parameters is established, and a trend representation space sample distance measurement method, a trend representation neural network model loss function design method and a health state estimation method based on neighbor samples are provided; and the fault state evaluation accuracy of the aircraft air pressure reducing valve can be effectively improved.

Description

Technical Field

[0001] This invention relates to the field of aviation equipment fault analysis technology, specifically to a method and system for assessing the fault status of an aircraft air pressure reducing valve based on trend characterization. Background Technology

[0002] The air pressure relief valve is a critical component for the successful startup of the aircraft's auxiliary power unit (APU). During APU startup, the air pressure relief valve provides stable pressure to the fuel tank through closed-loop pressure regulation. A malfunction could prevent the APU from starting, rendering the aircraft unable to fly and necessitating the use of a backup aircraft or mission cancellation, severely impacting mission execution.

[0003] During aircraft operation, onboard systems continuously record a large number of operating parameters, including key indicators such as pressure, temperature, speed, and current. Each flight, from takeoff to landing, consists of these parameters, which are downloaded and used for operational status analysis after each landing. Some of these parameters are directly related to the air pressure regulator, reflecting its operating characteristics at different stages of flight. Analyzing these parameters allows for the assessment of the air pressure regulator's operational status, providing a reference for maintenance decisions. However, existing fault condition assessment methods often rely on simple judgments based on upper and lower thresholds of the original parameters. While these methods are low-cost, they struggle to accurately represent the true health status of the air pressure regulator, easily leading to misjudgments or omissions. To improve assessment accuracy, more systematic research is necessary.

[0004] Air pressure reducing valve failures are often not one-off events, but rather a gradual process from healthy to deteriorating. Raw sensor samples, such as pressure, temperature, and valve position, represent instantaneous states and cannot directly reflect performance trends. Trend characterization can compress this time-series evolution into several comparable numerical indicators, making it easier for models to capture degradation patterns. This method focuses on fault monitoring parameters related to air pressure reducing valves, establishing a data processing flow including feature extraction, generalized life label creation, trend characterization extraction, distance measurement, loss function design, model training, and state assessment. This invention proposes an air pressure reducing valve fault state assessment method based on adaptive learning of trend characterization, achieving effective evaluation of air pressure reducing valve fault states. Summary of the Invention

[0005] To address the shortcomings of the existing technology, the present invention aims to provide a method and system for assessing the fault status of an aircraft air pressure relief valve based on trend representation. By establishing a fault status assessment process based on air pressure relief valve fault monitoring parameters, the present invention proposes a trend representation spatial sample distance metric method, a trend representation neural network model loss function design method, and a health status estimation method based on nearest neighbor samples, which can effectively improve the accuracy of fault status assessment of aircraft air pressure relief valves.

[0006] Specifically, on the one hand, the present invention provides a method for assessing the fault status of an aircraft air pressure reducing valve based on trend characterization, which includes the following steps: S1: Determine the fault monitoring parameters affecting the aircraft air pressure reducing valve, including pressure fault monitoring parameters, speed fault monitoring parameters, and temperature fault monitoring parameters; calculate the fault monitoring parameter characteristics of the aircraft air pressure reducing valve, determine the generalized life label of the aircraft air pressure reducing valve, and construct a unified dataset of aircraft air pressure reducing valve faults. S2: Construct a trend representation neural network model for predicting the performance degradation of aircraft air pressure reducing valves, and output a fault trend representation vector. ; Calculate the distance metric parameters used to optimize the trend representation neural network model. To quantify unified data on different faults in the fault trend representation vector space Similarities and differences between them; S3: Obtain the positional relationship function output based on the fault trend representation vector in step S2. Establish a loss function for optimizing the trend representation neural network model, and obtain the first... The trend representation comparison loss function for the aircraft air pressure relief valve failure data is as follows: ; in, For the first Trend representation of aircraft air pressure relief valve failure data compared with loss function; To find the maximum value of the function; Output the positional relationship function; Fault trend representation vector and Previous distance metric parameters; Fault trend representation vector and Distance metric between them; This is a parameter for distinguishing trends; Batch number for data collection of aircraft air pressure reducing valve malfunctions; A generalized lifespan label for aircraft air pressure relief valves; Output of the total loss function for calculating aircraft air pressure relief valve fault data Total loss function using aircraft air depressurization valve fault data Using the unified dataset of aircraft air pressure relief valve failure in step S1, the trend representation neural network model in step S2 is optimized and trained. S4: Input the unified fault data of the aircraft air pressure relief valve obtained in step S1 into the trend representation neural network model trained in step S3 to determine the fault status of the aircraft air pressure relief valve.

[0007] Preferably, step S2 specifically includes: S21: Construct a trend representation neural network model; set the input as the feature set of fault monitoring parameters of the aircraft air pressure reducing valve. This includes pressure fault monitoring parameters, speed fault monitoring parameters, and temperature fault monitoring parameters; the output is the corresponding aircraft air pressure reducing valve fault trend representation vector. ; Input dimensions for the trend representation neural network model; The output dimension of the trend representation neural network model; S22: Obtain a unified training set of aircraft air pressure relief valve fault data, calculate the distance metric parameter of random fault unified data samples in the aircraft air pressure relief valve fault trend representation vector, quantify the similarity and difference between different fault unified data samples, constrain and optimize the trend representation neural network model, and improve the output accuracy and stability of the trend representation neural network model.

[0008] Preferably, the distance metric parameter for the fault unified data sample in step S22 is: ; in, For the first unified data sample of faults The first aircraft air pressure relief valve Failure trend representation vector for each flight; For the second fault unified data sample The first aircraft air pressure relief valve Failure trend representation vector for each flight; The distance metric between two fault-compliant data samples; Number the aircraft; This refers to the aircraft sortie number.

[0009] Preferably, step S3 specifically includes: S31: Fault trend representation vector obtained in step S2 The set is the sampling data source, and three fault trend representation vectors are randomly selected with replacement each time. The function outputs the positional relationship of the unified fault data samples. ; S32: Establish the trend representation comparison loss function. The trend representation comparison loss function has a total of... Group, calculate the first Trend representation of aircraft air pressure relief valve failure data compared with loss function ; S33: Calculate the total loss function for aircraft air pressure relief valve fault data by summing the data in batches and output the result. .

[0010] Preferably, step S31 specifically includes: The first, second, and third fault trend representation vectors in the fault trend representation vector space In the middle, respectively use , , This indicates that the corresponding generalized lifespan labels are respectively used , , The calculated positional relationship function output is as follows: ; in, Output the positional relationship function between the 1st, 2nd, and 3rd fault trend characterization vector samples; For the first fault trend representation vector sample, the corresponding first... The first aircraft air pressure relief valve A broad lifespan label for each flight; The second fault trend representation vector sample corresponds to the first... The first aircraft air pressure relief valve A broad lifespan label for each flight; The third fault trend representation vector sample corresponds to the first... The first aircraft air pressure relief valve A broad lifespan label for each flight.

[0011] Preferably, the positional relationship function in step S3 is used to determine the actual position between the three fault trend characterization vector samples; when the positional relationship function outputs... =1 indicates that the actual distance between the first and second fault trend representation vector samples is greater than the actual distance between the first and third fault trend representation vector samples; the positional relationship function output is when =-1 indicates that the actual distance between the first and second fault trend representation vector samples is smaller than the actual distance between the first and third fault trend representation vector samples.

[0012] Preferably, step S4 specifically includes: The distance metric parameter set is obtained based on the trend characterization neural network model. ,Pick Center front A sample of aircraft air depressurization valve failure data forms a similar representative lifespan label reference set. ; Calculate the mean life label of the aircraft air pressure reducing valve failure data sample, which is the generalized life label of the aircraft air pressure reducing valve of the test sample. Set the health assessment threshold for the aircraft's air decompression valve. ,like If the aircraft's air decompression valve is in good working order, then the aircraft's air decompression valve is in good working order. This indicates that the aircraft's air pressure relief valve is about to malfunction, serving as a warning.

[0013] Preferably, step S1 specifically includes: S11: Based on the failure modes of the aircraft air pressure relief valve, screen out the fault monitoring parameters that are highly correlated with the health status of the aircraft air pressure relief valve, and determine the fault monitoring parameters that affect the aircraft air pressure relief valve. S12: Calculate the mean, variance, waveform factor, mean absolute deviation, and peak factor characteristics of the fault monitoring parameters of the aircraft air pressure reducing valve, and obtain the following results. Each fault monitoring parameter feature is composed of a fault monitoring parameter feature set. ; S13: Calculate the generalized life label of the aircraft air pressure reducing valve according to the degradation law of the aircraft air pressure reducing valve; construct a unified dataset of aircraft air pressure reducing valve faults based on the fault monitoring parameter feature set and the generalized life label.

[0014] Preferably, the generalized life label of the aircraft air pressure reducing valve in step S1 is: ; in, For the first The first aircraft air pressure relief valve A broad lifespan label for each sortie. and One-to-one correspondence; For the first The first aircraft air pressure relief valve Fault monitoring parameter characteristics for each flight; This refers to the total number of aircraft sorties.

[0015] On the other hand, the present invention provides an evaluation system for an aircraft air pressure relief valve fault state assessment method based on trend representation, which includes: a fault monitoring parameter extraction and processing module, a generalized life tag creation module, a trend representation neural network model construction module, a trend representation neural network model optimization module, and an aircraft air pressure relief valve fault state assessment module. The fault monitoring parameter extraction and processing module is used to determine the fault monitoring parameters and characteristics affecting the aircraft air pressure reducing valve. The fault monitoring parameters include: pressure fault monitoring parameters, speed fault monitoring parameters, and temperature fault monitoring parameters. The fault monitoring parameter characteristics of the aircraft air pressure reducing valve are: the mean, variance, waveform factor, mean absolute deviation, and peak factor of the fault monitoring parameter samples of the aircraft air pressure reducing valve. The generalized life tag generation module calculates the generalized life tag of the aircraft air pressure reducing valve based on the fault monitoring parameter characteristics of the aircraft air pressure reducing valve, and constructs a unified dataset of aircraft air pressure reducing valve faults by combining the fault monitoring parameter feature set and the generalized life tag. The trend representation neural network model building module fuses and nonlinearly maps the statistical characteristics of the fault monitoring parameters to generate a trend representation that reflects the dynamic change law of the aircraft air pressure reducing valve, which serves as the input to the trend representation neural network model. The trend representation neural network model optimization module uses a loss function and a unified training set of aircraft air depressurization valve fault data to optimize and train the trend representation neural network model until convergence reaches the stopping criterion, thus obtaining the trained trend representation neural network model. The aircraft air pressure relief valve fault status assessment module inputs a unified data sample of aircraft air pressure relief valve faults into a trained trend representation neural network model to determine the health status of the aircraft air pressure relief valve.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) In this invention, the trend representation is directly operated on by the trend representation comparison loss function in the trend representation space, which solves the problem of chaotic distribution of trend representation in the trend representation space in the traditional aircraft air pressure relief valve life prediction method, so that the trend representation is structurally distributed in the space according to the life of the aircraft air pressure relief valve.

[0017] (2) The aircraft air pressure relief valve life prediction framework proposed in this invention is different from the traditional end-to-end method of directly mapping life labels to features. It optimizes the trend characterization by calculating the distance between samples and comparing their positions, and directly calculates the predicted value of the aircraft air pressure relief valve life based on the trend characterization.

[0018] (3) This invention proposes a method for constructing a generalized life label reference sample set for aircraft air pressure relief valves and a method for predicting the life of aircraft air pressure relief valves based on the generalized life label reference set. The reference sample is used to calculate the life value of the aircraft air pressure relief valve in the sample set to be tested, making the calculation results more credible and solving the problem of unreliable prediction results of neural network methods. Attached Figure Description

[0019] Figure 1The flowchart shows the fault status assessment method for aircraft air pressure reducing valves based on trend characterization. Figure 2 This is a generalized lifespan label diagram for an embodiment of the present invention where the total number of aircraft sorties M=10. Figure 3 This is a diagram showing the relationship between the input vector, the trend representation neural network, and the output vector in an embodiment of the present invention. Figure 4 This is a diagram illustrating the optimization process of the positional relationship function in an embodiment of the present invention. Figure 5 This is a schematic diagram of a similar representative lifetime label reference set for the sample to be tested in an embodiment of the present invention. Detailed Implementation

[0020] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

[0021] This invention proposes a method for assessing the fault status of an aircraft air pressure reducing valve based on trend characterization, such as... Figure 1 As shown, the fault monitoring parameters of the aircraft air pressure reducing valve are determined, and the fault monitoring parameter characteristics and generalized life label are calculated. A trend representation neural network model for predicting the performance degradation of the aircraft air pressure reducing valve is constructed. A loss function is established to optimize the trend representation neural network model to improve the accuracy of predicting the performance degradation of the aircraft air pressure reducing valve. The fault state of the aircraft air pressure reducing valve is determined using the trend representation neural network model. The specific steps include: Step S1: Determine the fault monitoring parameters affecting the aircraft air pressure reducing valve, calculate the fault monitoring parameter characteristics of the aircraft air pressure reducing valve, and determine the generalized life label of the aircraft air pressure reducing valve.

[0022] Step S11: Determine the fault monitoring parameters affecting the aircraft air pressure reducing valve. Based on the design principle of the aircraft air pressure reducing valve and the typical fault phenomena exhibited by the aircraft air pressure reducing valve during operation, determine the operational failure mode of the aircraft air pressure reducing valve and screen out the fault monitoring parameters that are highly correlated with the health status of the aircraft air pressure reducing valve. The working process of the aircraft air pressure reducing valve involves multiple aspects such as airflow regulation, pressure control, temperature adaptation, and dynamic response of the actuator. Its performance changes are often reflected in the fluctuation characteristics of related fault monitoring parameters. Therefore, after screening, the three fault monitoring parameters finally determined by this invention include: pressure fault monitoring parameters, speed fault monitoring parameters, and temperature fault monitoring parameters.

[0023] Step S12: Calculate the fault monitoring parameter characteristics of the aircraft air pressure reducing valve. In this embodiment, the following is selected: The fault monitoring parameters of the aircraft air pressure relief valves are collected throughout their entire lifecycle, from installation to failure. The flight data for each aircraft is respectively used , ... This indicates that the data for the pressure fault monitoring parameters, speed fault monitoring parameters, and temperature fault monitoring parameters of the air pressure reducing valve for each aircraft flight are referred to as one sample. The mean characteristic, variance characteristic, waveform factor characteristic, mean absolute deviation characteristic, and peak factor characteristic of the fault monitoring parameter data for each aircraft air pressure reducing valve in step S11 are calculated respectively, resulting in a total of... Each fault monitoring parameter characteristic, in this embodiment, is described here. =3*5=15. For example, choose... The fault monitoring parameters for the air pressure reducing valves of five aircraft are based on their full life cycle data. Here, the full life cycle refers to the time from installation to failure of the air pressure reducing valve. The five aircraft are numbered No.1, No.2, No.3, No.4, and No.5. The number of aircraft for each aircraft's air pressure reducing valve's full life cycle is set as follows: Taking aircraft number No.1 as an example, This indicates that the aircraft flew a total of 180 sorties from the time the air pressure relief valve was installed until it malfunctioned, meaning the air pressure relief valve was considered to have been used 180 times. In this embodiment, if the aircraft uses the air pressure relief valve multiple times during a single flight, only the monitoring data from the first operation of the air pressure relief valve in that single flight is used.

[0024] Setting the first Aircraft air pressure relief valve The fault monitoring parameter feature set for each flight is as follows: The feature sets of pressure fault monitoring parameters, speed fault monitoring parameters, and temperature fault monitoring parameters include Each fault monitoring parameter characteristic, , Indicates the first Aircraft air pressure relief valve The first fault monitoring parameter of the flight Features; Construction Aircraft air pressure relief valve Fault monitoring parameter feature set of each flight For example, in the 10th flight of the air pressure reducing valve on aircraft No. 1, the pressure fault monitoring parameter sequence collected during the operation of the air pressure reducing valve is: {0.11, 0.24, 0.24, 0.15, 0.15, 0.16, 0.14, 0.09, 0.08}. The mean, variance, waveform factor, mean absolute deviation, and peak factor characteristics of this pressure fault monitoring parameter sequence are {0.11, 0.24, 0.24, 0.15, 0.15}, respectively. Note that this calculation only uses pressure fault monitoring parameters as an example to calculate mean characteristics, variance characteristics, waveform factor characteristics, mean absolute deviation characteristics, and peak factor characteristics. Speed ​​fault monitoring parameters and temperature fault monitoring parameters are not calculated. If all three fault monitoring parameters are calculated, then... .

[0025] Step S13: Determine the generalized life label of the aircraft air pressure reducing valve; the aircraft air pressure reducing valve will experience performance degradation during use, therefore, the generalized life label is determined according to the degradation pattern of the aircraft air pressure reducing valve; based on the results obtained in step S12... All of the aircraft's air pressure relief valves Fault monitoring parameter characteristics corresponding to each flight , ,......, The generalized life label for calculating the aircraft air pressure relief valve is: ; in, For the first The first aircraft air pressure relief valve A broad lifespan label for each sortie. and One-to-one correspondence; For the first The first aircraft air pressure relief valve Fault monitoring parameter characteristics for each flight; Total number of aircraft sorties; Number the aircraft; This refers to the aircraft sortie number.

[0026] In the embodiment, when the total number of aircraft sorties At that time, the generalized life label for the air pressure relief valve of aircraft No.1, on the 10th sortie, was: The generalized life label for the aircraft air pressure relief valve, serial number No. 1, on the 180th sortie is: , in turn Calculate the generalized life label for the fault monitoring parameter feature set of an aircraft air pressure reducing valve. For example... Figure 2 Extraction of each aircraft sortie shown This is a characteristic of specific fault monitoring parameters. Each fault monitoring parameter feature corresponds to a generalized lifespan label. The above-mentioned fault monitoring parameter feature set and generalized lifespan label of aircraft air pressure reducing valve are used to construct a unified dataset of aircraft air pressure reducing valve faults. In the embodiment, it is divided into a unified data training set, a unified data reference set, and a unified data test set of aircraft air pressure reducing valve faults in a ratio of 6:1:3.

[0027] Step S2: Construct a trend representation neural network model for predicting the performance degradation of aircraft air pressure relief valves, and calculate the distance metric parameters used to optimize the trend representation neural network model.

[0028] Step S21: Construct a trend representation neural network model; Since the raw data of the fault monitoring parameters of the aircraft air pressure reducing valve are not convenient to directly reflect the performance degradation trend of the aircraft air pressure reducing valve, it is necessary to construct a dedicated trend representation neural network model. Through model training, the automatic mapping from the raw data of the aircraft air pressure reducing valve fault monitoring parameters to the trend representation is achieved. The input to the trend representation neural network model is the feature set of fault monitoring parameters of the aircraft air pressure reducing valve formed in step S12. The output is the corresponding aircraft air pressure reducing valve fault trend representation vector. ,in, The output dimension of the trend representation neural network model is specified. The trend representation neural network model used in this embodiment includes, but is not limited to, LSTM, Transformer, and TCN. The specific neural network model is flexibly selected and optimized according to data characteristics and task requirements. The trend representation neural network model constructed in this invention is as follows: ; in, The proposed model is a trend representation neural network model, specifically a Transformer model in this embodiment of the invention. For the first The first aircraft air pressure relief valve The fault monitoring parameters for each flight include pressure fault monitoring parameters, speed fault monitoring parameters, and temperature fault monitoring parameters. ; For the first The first aircraft air pressure relief valve Failure trend representation vector for each flight ; Input dimensions for the trend representation neural network model; This defines the output dimension of the trend representation neural network model.

[0029] The fault monitoring parameters of the 10th flight of the aircraft air pressure reducing valve, numbered No.1, in the embodiment are as follows: In the output dimension of the trend representation neural network model In this case, After being processed by a trend representation neural network, the failure trend representation vector of the first aircraft air pressure relief valve in the tenth sortie is transformed. Note: This is from arrive The calculations are performed using a trend representation neural network model. The examples above are only for illustrating dimensions; the specific numerical values ​​are not examined. The calculation process is as follows: Figure 3 The diagram illustrates the relationship between the input vector, the trend representation neural network, and the output vector. The input is... * indivual Dimensional input vector, output is * indivual The dimensional output vector, where the trend representation neural network model can be any neural network model such as MLP, LSTM, or Transformer. This invention does not limit the type of neural network model used.

[0030] Step S22: Obtain the unified data training set for aircraft air pressure reducing valve failures from Step S13, and calculate the distance metric parameter of random unified data samples in the trend representation vector of aircraft air pressure reducing valve failures. To optimize the trend representation neural network model, it is necessary to clarify the existing unified data samples in the trend representation space. The relative distribution positions within the vector space. Therefore, in the aircraft air pressure relief valve fault trend characterization vector space. A distance metric method for unified fault data samples is established to quantify the similarity and differences between different unified fault data samples, providing constraints and optimization basis for training the trend representation neural network model. The distance metric parameters obtained from the unified fault data sample distance metric method are mapped one-to-one with the generalized life label of the aircraft air depressurization valve in step S13, ensuring that unified fault data samples with similar aircraft air depressurization valve generalized life labels are represented in the aircraft air depressurization valve fault trend representation vector space. The fault unified data samples, which are closer in size and have more distinct generalized life labels for air depressurization valves of different aircraft, are thus better separated and used in the loss function design of the trend representation neural network model in step S3, improving the discriminative ability and stability of the trend representation neural network model in degradation trend assessment; in the embodiment, the distance metric parameter between the two fault unified data samples is: ; in, For the first unified data sample of faults The first aircraft air pressure relief valve Failure trend representation vector for each flight; For the second fault unified data sample The first aircraft air pressure relief valve Failure trend representation vector for each flight; This is a distance metric parameter between two fault-based unified data samples.

[0031] The first unified fault data sample in the embodiment is the fault trend representation vector of the 10th flight of the 1st aircraft air pressure reducing valve. The second unified fault data sample in the embodiment is the fault trend representation vector of the 9th flight of the 1st aircraft air pressure reducing valve. Then the distance metric between the two samples is obtained as follows: .

[0032] Step S3: Establish a loss function for optimizing the trend representation neural network model to improve the prediction accuracy of aircraft air pressure relief valve performance degradation. The loss function is used to optimize the fault uniform data samples in step S2 in the fault trend representation vector space. The position in the vector space allows fault data samples with similar generalized lifetime labels to be placed in the fault trend representation vector space. Small and medium-sized fault data samples with longer generalized lifespan labels are placed in the fault trend representation vector space. The position in the middle is large. Therefore, based on the unified data samples of the fault to be tested in the fault trend representation vector space... The health status of the aircraft air pressure relief valve corresponding to the unified data sample of the fault under test is inferred from the location and surrounding fault unified data sample.

[0033] Step S31: Based on the results obtained in step S2 * Each fault trend representation vector The set is the sampling data source, and each time three fault trend representation vectors with replacement are randomly selected from the set. Forming a group; "with replacement" means that the same fault trend representation vector can appear repeatedly in different groups. Among the three fault trend representation vectors... In the sample, the first fault trend representation vector is used. Using vector multiplication as the anchor point, the second fault trend characterization vector is calculated sequentially. and the first fault trend representation vector The third fault trend representation vector and the first fault trend representation vector The similarity is calculated by increasing the size of the data. The first, second, and third fault trend representation vectors are located in the fault trend representation vector space. In the middle, respectively use , , express, and The similarity between them is measured by distance parameters. express, and Previous similarity was measured using distance metric parameters. Indicated; the generalized lifetime labels corresponding to the 1st, 2nd, and 3rd fault trend characterization vector samples are respectively represented by... , , This invention uses a positional relationship function to represent the positional relationship between the 1st, 2nd, and 3rd fault unified data samples. The specific output of the positional relationship function is as follows: ; in, Output the positional relationship function between the 1st, 2nd, and 3rd fault trend characterization vector samples; For the first fault trend representation vector sample, the corresponding first... The first aircraft air pressure relief valve A broad lifespan label for each flight; The second fault trend representation vector sample corresponds to the first... The first aircraft air pressure relief valve A broad lifespan label for each flight; The third fault trend representation vector sample corresponds to the first... The first aircraft air pressure relief valve A broad lifespan label for each flight.

[0034] The generalized life label of the first aircraft air depressurization valve in the 10th sortie, corresponding to the first failure trend characterization vector sample in the embodiment. The second failure trend characterization vector sample corresponds to the generalized life label of the first aircraft air pressure reducing valve in the 180th sortie. The third failure trend characterization vector sample corresponds to the generalized life label of the 18th flight of the 1st aircraft air depressurization valve. The output of the positional relationship function is then obtained as follows: .

[0035] The positional relationship function is mainly used to determine the actual position between three fault trend characterization vector samples; when the positional relationship function outputs... =1 indicates that the actual distance between the first and second fault trend representation vector samples is greater than the actual distance between the first and third fault trend representation vector samples. When the positional relationship function outputs... =-1 indicates that the actual distance between the first and second fault trend characterization vector samples is smaller than the actual distance between the first and third fault trend characterization vector samples. Here, distance refers to the distance between generalized lifetime labels, for example... , The distance between the two is The actual distance between the first and second fault trend representation vector samples is... The actual distance between the first and third fault trend representation vector samples is Since 0.94 > 0.04, then .

[0036] like Figure 4 The diagram illustrates the optimization process of the positional relationship function, where the first fault unified data sample is used. The first aircraft air pressure relief valve Failure trend representation vector for each flight The second fault unified data sample The first aircraft air pressure relief valve Failure trend representation vector for each flight And the third fault unified data sample The first aircraft air pressure relief valve Failure trend representation vector for each flight There are three samples, where the first fault trend representation vector sample corresponds to the first... The first aircraft air pressure relief valve Broad lifespan label per flight The first fault trend representation vector sample corresponding to the first fault trend representation vector sample The first aircraft air pressure relief valve Broad lifespan label per flight The absolute value of the difference is less than the first fault trend representation vector sample corresponding to the first fault trend representation vector sample. The first aircraft air pressure relief valve Broad lifespan label per flight The third fault trend representation vector sample corresponds to the first The first aircraft air pressure relief valve Broad lifespan label per flight The absolute value of the difference indicates that the states of sample 1 and sample 2 are closer than those of sample 1 and sample 3. At random initial positions, sample 1 and sample 2 are in the fault trend representation vector space. The distance between Sample 1 and Sample 2 in the fault trend representation vector space is larger than that between Sample 1 and Sample 3, therefore optimization is required. The optimization process involves narrowing the distance between Sample 1 and Sample 2 in the fault trend representation vector space. The positions of samples 1 and 3 in the fault trend representation vector space are pushed further away. The position of sample 1 and sample 2 in the fault trend representation vector space makes them appear in the position of the fault trend representation vector space. The location distance is smaller than that of sample 1 and sample 3.

[0037] Step S32: Establish the trend representation comparison loss function to optimize the trend representation neural network model in step S2. The trend representation comparison loss function has a total of Group, Collect batch parameters for aircraft air pressure reducing valve fault data; calculate the first batch parameters. The trend representation comparison loss function for the aircraft air pressure relief valve failure data is as follows: ; in, For the first Trend representation of aircraft air pressure relief valve failure data compared with loss function; To find the maximum value of the function; Fault trend representation vector and Previous distance metric parameters; Fault trend representation vector and Distance metric between them; This is a parameter for distinguishing trends; This is the batch number for collecting fault data of the aircraft air pressure reducing valve.

[0038] The generalized life label of the first aircraft air depressurization valve in the 10th sortie, corresponding to the first failure trend characterization vector sample in the embodiment. The second failure trend characterization vector sample corresponds to the generalized life label of the first aircraft air pressure reducing valve in the 180th sortie. The third failure trend characterization vector sample corresponds to the generalized life label of the 18th flight of the 1st aircraft air depressurization valve. Fault trend representation vector and The previous distance metric parameters were Fault trend representation vector and The distance metric between them is The trend characterization discrimination parameter is Then we get the first The trend representation comparison loss function for the aircraft air pressure relief valve failure data is as follows: ; When the first Trend representation of aircraft air pressure relief valve failure data compared with loss function When = 0, the fault trend representation vector of the first unified fault data sample The fault trend representation vector of the second unified fault data sample The fault trend representation vector of the third unified fault data sample Its relative position in the trend representation space is correct, so no optimization is needed.

[0039] In the second calculation embodiment, the generalized life label of the 10th sortie of the 1st aircraft air pressure reducing valve corresponds to the first failure trend characterization vector sample. The second failure trend characterization vector sample corresponds to the generalized life label of the first aircraft air pressure reducing valve in the 180th sortie. The third failure trend characterization vector sample corresponds to the generalized life label of the 18th flight of the 1st aircraft air depressurization valve. Fault trend representation vector and The previous distance metric parameters were Fault trend representation vector and The distance metric between them is The trend characterization discrimination parameter is Then we get the first The trend representation comparison loss function for the aircraft air pressure relief valve failure data is as follows: ; At this point, it represents the fault trend representation vector of the first unified fault data sample. The fault trend representation vector of the second unified fault data sample The fault trend representation vector of the third unified fault data sample Its relative position in the trend representation space is incorrect and needs to be optimized.

[0040] The purpose of designing a trend representation comparison loss function for aircraft air pressure reducing valve fault data is to narrow the distance between the target fault trend representation vector sample and the actual similar fault trend representation vector sample, and to widen the distance between the anchor fault trend representation vector sample and the actual far-fetched fault trend representation vector sample. Trend representation discrimination parameter. To ensure sufficient differentiation between trend representations, it is important to avoid misjudgments caused by overly close embedding of faulty trend representation vector samples. Figure 4 As shown, if the three fault trend representation vector samples are in the fault trend representation vector space The relative positions in the text are correct, at this time Returning 0 means that the trend representation neural network will not be optimized.

[0041] Step S33: During the training of the trend representation neural network model, the aircraft air depressurization valve fault data is organized in batches. Each batch undergoes a loss function calculation and backpropagation update. The total loss function output for the aircraft air depressurization valve fault data is as follows: ; in, Output the total loss function results for aircraft air pressure relief valve fault data; For the batch parameters of aircraft air pressure reducing valve fault data collection, 32, 64 and 128 are used in the example.

[0042] The total loss function using the above aircraft air pressure relief valve fault data Using the same data training set as the aircraft air depressurization valve fault in step S1, the trend representation neural network model in step S2 is optimized and trained until the specified number of training iterations is reached, at which point training stops, and the trained trend representation neural network model is obtained.

[0043] Step S4: Input the unified data reference set of aircraft air pressure relief valve failure obtained in step S1 into the trend representation neural network model trained in step S3 to determine the failure state of the aircraft air pressure relief valve.

[0044] After the trend representation neural network model has been trained, it is retained. The samples of the unified data reference set for aircraft air pressure reducing valve failure or the measured data from step S13 are input into the trend representation neural network model to obtain the parameter characteristics, trend representations, and corresponding generalized life labels of all aircraft air pressure reducing valve failure data samples in the generalized life label reference set. This forms the aircraft air pressure reducing valve generalized life label reference set. , where m and n represent the number of aircraft in the reference set and the number of sorties per aircraft, respectively.

[0045] During the testing phase, the unified data test set of aircraft air depressurization valve failure from step S13 is used to test the output effect of the trend characterization neural network model; the test set of aircraft air depressurization valve failure data samples is then used... This indicates that the trend representation neural network model obtained after step S2 yields the fault trend representation vector of the air pressure reducing valve of the aircraft under test. Using the distance metric parameters in step S3, the fault trend representation vector of the air pressure reducing valve of the aircraft under test is calculated sequentially. Reference set of generalized life label for aircraft air pressure reducing valves middle The similarity between them; the distance metric parameters are as follows: ; in, For distance metric parameters; It is the failure trend representation vector of the generalized lifetime label reference set.

[0046] Obtain the set of distance metric parameters ,Pick The smallest in the front A reference set of similar lifespan labels is composed of data samples of the most similar aircraft air depressurization valve failures. , of which Each fault trend representation vector This represents the failure trend representation vector relative to the generalized lifetime label reference set. Most similar One sample. The mean lifespan label of all aircraft air pressure regulator failure data samples in the similarity representative lifespan label reference set is the generalized lifespan label of the aircraft air pressure regulator valve of the sample under test, specifically: ; in, A generalized life label for the aircraft air pressure relief valve of the sample to be tested; The number of similar representative lifetime label reference samples; For the first A generalized lifespan label for an aircraft air depressurization valve.

[0047] Based on the fault data sample of the aircraft's air pressure relief valve, the generalized life label of the aircraft's air pressure relief valve. Determine the health status of the air pressure relief valve of the aircraft under test. , The set health assessment threshold for the aircraft air pressure relief valve is selected based on actual testing; if the threshold is set, the aircraft air pressure relief valve is considered to be in a healthy state. This indicates that the aircraft's air pressure relief valve is about to malfunction, providing an early warning. For example, The algorithm is expected to issue a warning when the aircraft's air depressurization valve has less than 15% of its service life. When 0.7 > 0.15, it indicates that the device is in a healthy state; when When the value is 0.05 < 0.15, it indicates that the equipment is about to malfunction, and a warning is issued.

[0048] like Figure 5 The diagram shows a similar representative life label reference set of fault data samples of the air pressure reducing valve of the aircraft under test. This is based on the fault trend characterization vector space of the test samples. In the middle, select the one that is most similar to it. A set of similar representative life label references is formed by a sample of aircraft air pressure relief valve failure data.

[0049] The second aspect of this invention proposes an aircraft air pressure relief valve fault state assessment system based on a trend representation-based method, which includes: a fault monitoring parameter extraction and processing module, a generalized life tag creation module, a trend representation neural network model construction module, a trend representation neural network model optimization module, and an aircraft air pressure relief valve fault state assessment module.

[0050] The fault monitoring parameter extraction and processing module is used to determine the fault monitoring parameters and their characteristics that affect the aircraft air pressure reducing valve. Based on the design principle of the aircraft air pressure reducing valve and the typical fault phenomena exhibited by the aircraft air pressure reducing valve during field operation, the determined fault monitoring parameters include: pressure fault monitoring parameters, speed fault monitoring parameters, and temperature fault monitoring parameters. The module acquires the full life cycle data of the fault monitoring parameters of the aircraft air pressure reducing valve and calculates the mean characteristics, variance characteristics, waveform factor characteristics, mean absolute deviation characteristics, and peak factor characteristics of the fault monitoring parameter samples for each aircraft air pressure reducing valve, which are then used as the fault monitoring parameter characteristics of the aircraft air pressure reducing valve.

[0051] The generalized lifespan label generation module calculates the generalized lifespan label of the aircraft air pressure reducing valve based on the fault monitoring parameter characteristics. It then constructs a unified dataset of aircraft air pressure reducing valve faults by combining the fault monitoring parameter feature set and the generalized lifespan label. The dataset is divided proportionally into a unified training set, a unified reference set, and a unified test set.

[0052] The trend representation neural network model construction module fuses and nonlinearly maps the statistical characteristics of the fault monitoring parameters to generate a trend representation that reflects the dynamic change law of the aircraft air pressure relief valve. As the input of the trend representation neural network model, the trend representation neural network model adopted in this invention includes, but is not limited to, LSTM, Transformer, TCN and other models.

[0053] The trend representation neural network model optimization module constructs a loss function so that the loss function of each batch of aircraft air pressure relief valve failure data samples is calculated and updated through backpropagation. Using the total loss function and the unified training set of aircraft air pressure relief valve failure data, the trend representation neural network model is optimized and trained until convergence reaches the stopping criterion, thus obtaining the trained trend representation neural network model.

[0054] The aircraft air pressure relief valve fault status assessment module inputs samples from the unified data reference set of aircraft air pressure relief valve faults or measured data into the trained trend representation neural network model to determine the health status of the sample to be tested, judge the health status of the aircraft air pressure relief valve, and issue early warnings.

[0055] The beneficial effects of this invention are as follows: The aircraft air depressurization valve failure state assessment method proposed in this invention directly operates on the trend representation in the trend representation space using a trend representation comparison loss function, solving the problem of chaotic distribution of trend representation in the trend representation space in traditional life prediction methods, so that the trend representation presents a structured distribution in the space according to the life of the aircraft air depressurization valve; through the construction method of the aircraft air depressurization valve generalized life label sample reference set and the aircraft air depressurization valve life prediction method based on the generalized life label sample reference set, the life value of the test sample set is calculated using the reference sample, and the calculation result is more reliable than that of traditional methods; the aircraft air depressurization valve life prediction framework optimizes the trend representation by calculating the distance between samples and comparing their positions, and directly calculates the predicted value of the aircraft air depressurization valve life based on the trend representation.

[0056] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for assessing a fault condition of an aircraft air pressure reducing valve based on trend characterization, the method comprising: It includes: S1: Determine the fault monitoring parameters that affect the aircraft's air pressure relief valve, including pressure fault monitoring parameters, speed fault monitoring parameters, and temperature fault monitoring parameters; Calculate the fault monitoring parameter characteristics of the aircraft air pressure reducing valve, determine the generalized life label of the aircraft air pressure reducing valve, and construct a unified dataset of aircraft air pressure reducing valve faults. S2: Construct a trend characterization neural network model for performance degradation prediction of an aircraft air pressure reducing valve, output a fault trend characterization vector ; Computing distance metric parameters for optimizing trend characterization neural network models to quantify the similarity and difference of each fault unified data in the fault trend characterization vector space ​ S3: Obtain the position relationship function output according to the fault trend characterization vector in step S2 , establish a loss function for optimizing the trend characterization neural network model, and obtain the trend characterization comparison loss function of the first group of aircraft air pressure reducing valve fault data. ； wherein, is a first is a second is a max function; is a position relationship function output; is a failure trend characterization vector and is a distance metric parameter; is a failure trend characterization vector and is a distance metric parameter; is a trend characterization discriminative parameter; is a batch number of aircraft air pressure relief valve failure data collection; is a generalized life label of aircraft air pressure relief valve. Computing a total loss function output result for aircraft air pressure relief valve failure data Total loss function using aircraft air pressure relief valve failure data Optimizing training of the trend characterization neural network model in step S2 using the aircraft air pressure relief valve failure uniform data set in step S1; S4: Input the unified fault data of the aircraft air pressure relief valve obtained in step S1 into the trend representation neural network model trained in step S3 to determine the fault status of the aircraft air pressure relief valve.

2. The method of claim 1, wherein: Step S2 is as follows: S21: constructing a trend characterization neural network model; setting the input as a fault monitoring parameter feature set of an aircraft air pressure reducing valve , including a pressure fault monitoring parameter, a rotating speed fault monitoring parameter and a temperature fault monitoring parameter; The output is the corresponding aircraft air pressure reducing valve fault trend representation vector. ; Input dimensions for the trend representation neural network model; The output dimension of the trend representation neural network model; S22: Obtain a unified training set of aircraft air pressure relief valve fault data, calculate the distance metric parameter of random fault unified data samples in the aircraft air pressure relief valve fault trend representation vector, and quantify the similarity and difference between each fault unified data sample.

3. The method for assessing the fault status of an aircraft air pressure reducing valve based on trend characterization according to claim 2, characterized in that: The distance metric parameter for the fault unified data sample in step S22 is: ； in, For the first unified data sample of faults The first aircraft air pressure relief valve Failure trend representation vector for each flight; For the second fault unified data sample The first aircraft air pressure relief valve Failure trend representation vector for each flight; The distance metric between two fault-compliant data samples; Number the aircraft; This refers to the aircraft sortie number.

4. The method for assessing the fault status of an aircraft air pressure reducing valve based on trend characterization according to claim 1, characterized in that: Step S3 is as follows: S31: Fault trend representation vector obtained in step S2 The set is the sampling data source, and three fault trend representation vectors are randomly selected with replacement each time. The function outputs the positional relationship of the unified fault data samples. ; S32: Establish the trend representation comparison loss function. The trend representation comparison loss function has a total of... Group, calculate the first Trend representation of aircraft air pressure relief valve failure data compared with loss function ; S33: Calculate the total loss function for aircraft air pressure relief valve fault data by summing the data in batches and output the result. .

5. The method for assessing the fault status of an aircraft air pressure reducing valve based on trend characterization according to claim 4, characterized in that: Step S31 is as follows: The first, second, and third fault trend representation vectors in the fault trend representation vector space In the middle, respectively use , , This indicates that the corresponding generalized lifespan labels are respectively used , , The calculated positional relationship function output is as follows: ； in, Output the positional relationship function between the 1st, 2nd, and 3rd fault trend characterization vector samples; For the first fault trend representation vector sample, the corresponding first... The first aircraft air pressure relief valve A broad lifespan label for each flight; The second fault trend representation vector sample corresponds to the first... The first aircraft air pressure relief valve A broad lifespan label for each flight; The third fault trend representation vector sample corresponds to the first... The first aircraft air pressure relief valve A broad lifespan label for each flight.

6. The method for assessing the fault status of an aircraft air pressure reducing valve based on trend characterization according to claim 4, characterized in that: The positional relationship function in step S3 is used to determine the actual position between the three fault trend characterization vector samples; when the positional relationship function outputs... =1 indicates that the actual distance between the first and second fault trend representation vector samples is greater than the actual distance between the first and third fault trend representation vector samples; the positional relationship function output is when =-1 indicates that the actual distance between the first and second fault trend representation vector samples is smaller than the actual distance between the first and third fault trend representation vector samples.

7. The method for assessing the fault status of an aircraft air pressure reducing valve based on trend characterization according to claim 1, characterized in that: Step S4 is as follows: The distance metric parameter set is obtained based on the trend characterization neural network model. ,Pick Center front A sample of aircraft air depressurization valve failure data forms a similar representative lifespan label reference set. ; Calculate the mean life label of the aircraft air pressure reducing valve failure data sample, which is the generalized life label of the aircraft air pressure reducing valve of the test sample. ; Setting the health assessment threshold for aircraft air decompression valves ,like If the aircraft's air decompression valve is in good working order, then the aircraft's air decompression valve is in good working order. This indicates that the aircraft's air pressure relief valve is about to malfunction, serving as a warning.

8. The method for assessing the fault status of an aircraft air depressurization valve based on trend characterization according to claim 1, characterized in that: Step S1 is as follows: S11: Based on the failure modes of the aircraft air pressure relief valve, screen out the fault monitoring parameters that are highly correlated with the health status of the aircraft air pressure relief valve, and determine the fault monitoring parameters that affect the aircraft air pressure relief valve. S12: Calculate the mean, variance, waveform factor, mean absolute deviation, and peak factor characteristics of the fault monitoring parameters of the aircraft air pressure reducing valve, and obtain the following results. Each fault monitoring parameter feature is composed of a fault monitoring parameter feature set. ; S13: Calculate the generalized life label of the aircraft air pressure reducing valve according to the degradation law of the aircraft air pressure reducing valve; construct a unified dataset of aircraft air pressure reducing valve faults based on the fault monitoring parameter feature set and the generalized life label.

9. The method for assessing the fault status of an aircraft air depressurization valve based on trend characterization according to claim 8, characterized in that: The generalized life label for the aircraft air pressure reducing valve in step S1 is: ； in, For the first The first aircraft air pressure relief valve A broad lifespan label for each sortie. and One-to-one correspondence; For the first The first aircraft air pressure relief valve Fault monitoring parameter characteristics for each flight; This refers to the total number of aircraft sorties.

10. An evaluation system for the trend-based assessment method for aircraft air pressure reducing valve failure status as described in any one of claims 1 to 9, characterized in that, It includes: a fault monitoring parameter extraction and processing module, a generalized life tag creation module, a trend characterization neural network model construction module, a trend characterization neural network model optimization module, and an aircraft air pressure reducing valve fault status assessment module; The fault monitoring parameter extraction and processing module is used to determine the fault monitoring parameters and characteristics affecting the aircraft air pressure reducing valve. The fault monitoring parameters include: pressure fault monitoring parameters, speed fault monitoring parameters, and temperature fault monitoring parameters. The fault monitoring parameter characteristics of the aircraft air pressure reducing valve are: the mean, variance, waveform factor, mean absolute deviation, and peak factor of the fault monitoring parameter samples of the aircraft air pressure reducing valve. The generalized life tag generation module calculates the generalized life tag of the aircraft air pressure reducing valve based on the fault monitoring parameter characteristics of the aircraft air pressure reducing valve, and constructs a unified dataset of aircraft air pressure reducing valve faults by combining the fault monitoring parameter feature set and the generalized life tag. The trend representation neural network model building module fuses and nonlinearly maps the statistical characteristics of the fault monitoring parameters to generate a trend representation that reflects the dynamic change law of the aircraft air pressure reducing valve, which serves as the input to the trend representation neural network model. The trend representation neural network model optimization module uses a loss function and a unified training set of aircraft air depressurization valve fault data to optimize and train the trend representation neural network model until convergence reaches the stopping criterion, thus obtaining the trained trend representation neural network model. The aircraft air pressure relief valve fault status assessment module inputs a unified data sample of aircraft air pressure relief valve faults into a trained trend representation neural network model to determine the health status of the aircraft air pressure relief valve.

Images