A method for predicting the life of a sealing structure of an aviation hydraulic pipeline system

Abstract

The present invention discloses a method for predicting the service life of a sealing structure of an aviation hydraulic pipeline system, which relates to the technical field of predicting the service life of a sealing structure and includes the following steps: constructing different service life test data sets and performing clustering analysis according to environmental parameters; respectively pre-training a machine learning model based on each cluster of service life test clustering data sets; collecting environmental parameter data and state parameter data; calculating the distance between the environmental parameter data and the environmental parameter clustering center, and selecting at least two service life prediction models for service life prediction; calculating the weights of each remaining service life prediction result according to the distance, and obtaining the final service life prediction result by weighted summation. The present application improves the accuracy and reliability of the service life prediction of the sealing structure by constructing service life test data sets under different environmental parameter test data, combining clustering analysis, machine learning models, and calculating weights according to the distance.

Description

Technical Field

[0001] This application relates to the technical field of seal structure life prediction, and particularly to a method for predicting the life of a seal structure of an aviation hydraulic pipeline system. Background Art

[0002] The aviation hydraulic pipeline system is a main component of the aviation hydraulic system. This system completes mechanical motion through the flow and pressure transmission of liquid, and is the basis for the normal operation of multiple key aviation systems. The reliability and life of the seal structure of the aviation hydraulic pipeline system directly affect the stability and efficiency of the hydraulic system. By predicting the life of the seal structure of the aviation hydraulic pipeline system, potential hazards that may exist in the seal structure can be detected in time, and hydraulic system failures caused by seal failures can be prevented, thereby ensuring aviation flight safety.

[0003] In the prior art, Chinese Patent CN117034640A discloses a method, system, terminal device and storage medium for predicting the life of a pipeline. The method includes: if a single performance parameter analysis item corresponds to multiple working condition parameters, a multivariate analysis table corresponding to the performance parameter analysis item is generated by combining each working condition parameter; the multivariate analysis table is parsed to obtain the periodic influence trend corresponding to the performance parameter analysis item; a life prediction model corresponding to the target pipeline is established by combining the periodic influence trend and the performance analysis index of the target pipeline; the historical maintenance data of the target pipeline is imported into the life prediction model to generate life prediction data corresponding to the target pipeline; if there are multiple target pipelines, a life prediction distribution map of the pipeline system is generated by combining the life prediction data and associated data between each target pipeline.

[0004] Chinese Patent CN114528662A discloses a method, device and working machine for predicting the remaining life of a structural member. The method includes: inputting the target data of the working machine into the remaining life prediction model so that the remaining life prediction model outputs the remaining life of the structural member to be predicted in the working machine.

[0005] However, the above prior arts do not consider the different non-linear influence effects of different environmental combinations (such as high pressure, high temperature, high vibration, high pressure, low temperature, high vibration, low pressure, low temperature, low vibration, etc.) on the aging of the seal structure. The correlation between environmental parameters and the distribution of training data of the prediction model is poor, and the accuracy and reliability of seal structure life prediction are poor. Summary of the Invention

[0006] This application provides a method for predicting the life of a seal structure of an aviation hydraulic pipeline system to solve the problems that the prior art does not consider the different non-linear influence effects of different environmental combinations on the aging of the seal structure, the correlation between environmental parameters and the distribution of training data of the prediction model is poor, and the accuracy and reliability of seal structure life prediction are poor.

[0007] On the one hand, the present application provides a method for predicting the life of a sealing structure of an aviation hydraulic pipeline system, including the following steps:

[0008] Step 1: Construct a life test data set under different environmental parameter test data, perform cluster analysis according to environmental parameters, and obtain several clusters of life test cluster data sets.

[0009] Step 2: Pre-train a machine learning model based on each cluster of life test cluster data sets respectively to obtain several life prediction models.

[0010] Step 3: Collect environmental parameter data and state parameter data of the sealing structure of the aviation hydraulic pipeline system.

[0011] Step 4: Calculate the distance between the environmental parameter data and the environmental parameter cluster center corresponding to each life prediction model, and select at least two life prediction models in ascending order of distance.

[0012] Step 5: Respectively use each selected life prediction model to predict the life according to the environmental parameter data and the state parameter data to obtain at least two remaining life prediction results.

[0013] Step 6: Calculate the weight of each remaining life prediction result according to the distance between the environmental parameter data and the environmental parameter cluster center of each selected life prediction model, and perform weighted summation to obtain the final life prediction result.

[0014] In a possible implementation manner, Step 1 includes:

[0015] Previously conduct life tests on the sealing structures of the aviation hydraulic pipeline system of the same model under different environmental parameter test data respectively, construct a life test data set under different environmental parameter test data, and each group of data in the life test data set includes environmental parameter test data, state parameter test data at a certain moment, and the corresponding remaining test life.

[0016] Perform cluster analysis on the life test data sets under different environmental parameter test data according to environmental parameters to obtain several clusters of life test cluster data sets.

[0017] In a possible implementation manner, in Step 1, the cluster analysis adopts one of the K-means clustering algorithm, hierarchical clustering algorithm, and spectral clustering algorithm.

[0018] In a possible implementation manner, in Step 2, each life prediction model adopts the same model architecture.

[0019] The machine learning model adopts one of a feedforward neural network model, a convolutional neural network model, and a support vector machine model.

[0020] In a possible implementation, in step two, when the machine learning model adopts a feedforward neural network model or a convolutional neural network model, before pre-training each machine learning model, set the initial weights of each type of data in the corresponding life test clustering dataset.

[0021] In a possible implementation, the setting of the initial weights of each type of data in the corresponding life test clustering dataset adopts one of the entropy weight method and the coefficient of variation method.

[0022] In a possible implementation, in step three, the environmental parameter data includes: pressure data, temperature data, vibration data, flow data, and the state parameter data includes: pressure difference data and contact stress data.

[0023] In a possible implementation, in step four, use one of the Euclidean distance, Mahalanobis distance, cosine similarity, and Manhattan distance to calculate the distance between the environmental parameter data and the environmental parameter clustering center corresponding to each life prediction model.

[0024] In a possible implementation, in step six, use the Gaussian kernel function to calculate the weight of each remaining life prediction result according to the distance between the environmental parameter data and the environmental parameter clustering center of each selected life prediction model.

[0025] In a possible implementation, after step six, it further includes: step seven, obtain several final life prediction results within a preset time period and perform linear fitting to generate a life function line, modify the outliers in the several final life prediction results to the corresponding values in the life function line, and optimize and train the corresponding life prediction model according to the modified final life prediction results and the corresponding environmental parameter data and state parameter data.

[0026] The linear fitting adopts the least squares method.

[0027] A life prediction method for a sealing structure of an aviation hydraulic pipeline system in the present application has the following advantages:

[0028] By constructing a life test dataset under different environmental parameter test data, combining clustering analysis, machine learning models, and calculating weights according to distances, the non-linear influence of different environmental combinations on the aging of the sealing structure is considered, the relevance between environmental parameters and the training data distribution of the prediction model is improved, and the accuracy and reliability of the life prediction of the sealing structure are improved.

[0029] By pre-training a machine learning model respectively based on each cluster of life test clustering datasets to obtain several life prediction models, the training pertinence is improved, and further the reliability of the life prediction model is improved.

[0030] For the initial weights of each type of data in the proposed life test clustering dataset, one of the entropy weight method and the coefficient of variation method is adopted. The initial weights of each cluster of life test clustering datasets are set respectively through the weighting method, which improves the rationality of the initial weights, enables the training process of the life prediction model to converge quickly, and improves the training efficiency.

[0031] The proposed method uses the Gaussian kernel function to calculate the weights of each remaining life prediction result according to the distance between the environmental parameter data and the environmental parameter clustering centers of each selected life prediction model, avoiding the risk of losing valid information by a single prediction model, reducing the fluctuation error of the prediction results, and at the same time realizing the non-linear fusion of the prediction results.

[0032] The proposed method modifies the outliers in several final life prediction results to the corresponding values in the life function straight line, and optimizes and trains the corresponding life prediction model according to the modified final life prediction results and the corresponding environmental parameter data and state parameter data, improving the evolvability and adaptability of the life prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0034] Figure 1 It is a schematic flow chart of a method for predicting the life of a sealing structure of an aviation hydraulic pipeline system provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0035] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0036] As Figure 1 shown, an embodiment of the present application provides a method for predicting the life of a sealing structure of an aviation hydraulic pipeline system, including the following steps:

[0037] Step 1: Construct a life test dataset under different environmental parameter test data, perform clustering analysis according to environmental parameters, and obtain several clusters of life test clustering datasets.

[0038] Step 2: Based on each cluster of life test clustering datasets, pre-train a machine learning model respectively to obtain several life prediction models.

[0039] Step 3: Collect the environmental parameter data and status parameter data of the sealing structure of the aviation hydraulic pipeline system.

[0040] Step 4: Calculate the distances between the environmental parameter data and the environmental parameter clustering centers corresponding to each life prediction model, and select at least two life prediction models in ascending order of the distances.

[0041] Step 5: Use each selected life prediction model to perform life prediction according to the environmental parameter data and the status parameter data respectively to obtain at least two remaining life prediction results.

[0042] Step 6: Calculate the weights of each remaining life prediction result according to the distances between the environmental parameter data and the environmental parameter clustering centers of the selected life prediction models, and perform weighted summation to obtain the final life prediction result.

[0043] Exemplarily, Step 1 includes:

[0044] Pre-conduct life tests on the sealing structures of aviation hydraulic pipeline systems of the same model under different environmental parameter test data respectively, and construct life test datasets under different environmental parameter test data. Each group of data in the life test dataset includes environmental parameter test data, status parameter test data at a certain moment, and the corresponding remaining test life.

[0045] Perform clustering analysis on the life test datasets under different environmental parameter test data according to the environmental parameters to obtain several clusters of life test clustering datasets.

[0046] Exemplarily, in Step 1, the clustering analysis adopts one of the K-means clustering algorithm, hierarchical clustering algorithm, and spectral clustering algorithm.

[0047] Specifically, in this embodiment, a total of 32 life test datasets under different environmental parameter test data are constructed. The K-means clustering algorithm is used to perform clustering analysis on the 32 life test datasets according to the environmental parameters. The specific process is as follows: Select the environmental parameter test data in the 32 life test datasets as clustering features, and perform standardization processing on all the environmental parameter test data; Select the number of clusters for clustering (i.e., the K value). In this embodiment, the elbow method is used to obtain a K value of 4; Apply the K-means clustering algorithm to cluster the life test datasets under different environmental parameter test data according to the K value to obtain 4 clusters of life test clustering datasets.

[0048] Exemplarily, in step two, each life prediction model adopts the same model architecture.

[0049] The machine learning model adopts one of a feedforward neural network model, a convolutional neural network model, and a support vector machine model.

[0050] Specifically, in this embodiment, to match the number of clusters in the life test clustering dataset, 4 life prediction models are set up. Each life prediction model is obtained by pre-training a feedforward neural network model with a cluster of life test clustering dataset. In this embodiment, the model architecture of the feedforward neural network includes an input layer, a hidden layer, and an output layer. The input layer is used to input environmental parameter data and state parameter data (during the training process, environmental parameter test data and state parameter test data are input, and the corresponding remaining life of the test is used as the target prediction value). The hidden layer uses the ReLU activation function, and the output layer is used to output the remaining life prediction result.

[0051] In this embodiment, the process of pre-training each feedforward neural network model includes: normalizing the data in the corresponding life test clustering dataset to ensure the stability and efficiency during the model training process; selecting a loss function (in this embodiment, the mean squared error is selected) to measure the difference between the remaining life prediction result output by the output layer and the corresponding target prediction value; selecting an optimizer (in this embodiment, the Adam optimizer is selected) to update the model weights; dividing the life test clustering dataset into a training set and a validation set, training the feedforward neural network model with the training set until the loss function converges or reaches the preset maximum number of iterations, and using the validation set to evaluate the model performance to prevent overfitting.

[0052] Exemplarily, in step two, when the machine learning model adopts a feedforward neural network model or a convolutional neural network model, before pre-training each machine learning model, the initial weights of each type of data in the corresponding life test clustering dataset are set.

[0053] Exemplarily, the method of setting the initial weights of each type of data in the corresponding life test clustering dataset adopts one of the entropy weight method and the coefficient of variation method.

[0054] Specifically, in this embodiment, the entropy weight method is used to set the initial weights of each type of data in the four clusters of life test clustering datasets respectively. The entropy weight method is an objective weighting method that calculates the weights of various types of data using information entropy. According to the basic principles of information theory, information is a measure of the order degree of a system, while entropy is a measure of the disorder degree of a system. For a certain type of data, the smaller the information entropy value, the greater the degree of dispersion of the data, and the greater the impact (i.e., weight) of this data on the comprehensive evaluation, and vice versa. The specific weight setting process includes: calculating the information entropy of each type of data, which reflects the amount of information and uncertainty of the data; calculating the redundancy of each type of data, that is, 1 minus the information entropy, which reflects the degree of effective information provided by the data; dividing the redundancy of each type of data by the sum of the redundancies of all data to obtain the initial weight of each type of data.

[0055] By setting the initial weights of each type of data in the corresponding life test clustering dataset in this way, the corresponding life prediction model can learn the key features in the corresponding life test clustering dataset faster during training, enabling the model to converge quickly.

[0056] Exemplarily, in step three, the environmental parameter data includes: pressure data, temperature data, vibration data, flow data, and the state parameter data includes: pressure difference data and contact stress data.

[0057] Specifically, in this embodiment, the pressure data uses the average working pressure of the pipeline where the sealing structure is located, the temperature data uses the average working temperature of the pipeline where the sealing structure is located, the vibration data uses the average vibration intensity of the pipeline where the sealing structure is located, and the flow data uses the fluid flow rate of the pipeline where the sealing structure is located. The pressure difference data uses the pressure difference on both sides of the sealing structure, and the contact stress data includes the maximum contact stress, minimum contact stress, and contact stress standard deviation of the contact surface of the sealing structure.

[0058] Exemplarily, in step four, one of the Euclidean distance, Mahalanobis distance, cosine similarity, and Manhattan distance is used to calculate the distance between the environmental parameter data and the environmental parameter clustering center corresponding to each life prediction model.

[0059] Specifically, in this embodiment, the Mahalanobis distance is used to calculate the distance between the environmental parameter data and the environmental parameter clustering center corresponding to each life prediction model. Since the environmental parameter clustering centers corresponding to each cluster of life test clustering datasets have been obtained when performing clustering analysis according to environmental parameters in step one, furthermore, the environmental parameter clustering centers corresponding to each life prediction model are known, and only need to directly substitute the environmental parameter data and the environmental parameter clustering center corresponding to each life prediction model into the existing Mahalanobis distance calculation formula. In other possible embodiments, the Euclidean distance, cosine similarity, Manhattan distance, etc. can also be used.

[0060] In this embodiment, a total of two life prediction models with the closest and the second-closest distances are selected to perform life prediction based on the environmental parameter data and the state parameter data, and two remaining life prediction results are obtained.

[0061] Exemplarily, in step six, a Gaussian kernel function is used to calculate the weight of each remaining life prediction result according to the distance between the environmental parameter data and the environmental parameter clustering center of each selected life prediction model.

[0062] Specifically, in this embodiment, before calculating the weight according to the distance, the distance between the environmental parameter data and the environmental parameter clustering center of each selected environmental parameter is first normalized; then the Gaussian kernel function is used to convert each normalized distance into a corresponding weight; and then the weights are normalized so that the sum of the weights of all remaining life prediction results is equal to 1; the weighted sum of each remaining life prediction result is calculated using the normalized weights to obtain the final life prediction result.

[0063] Exemplarily, after step six, it further includes: step seven, obtaining several final life prediction results within a preset time period and performing linear fitting to generate a life function line, modifying the outliers among the several final life prediction results to the corresponding values in the life function line, and optimizing and training the corresponding life prediction model according to the modified final life prediction results and the corresponding environmental parameter data and state parameter data.

[0064] The linear fitting uses the least squares method.

[0065] Specifically, the least squares method finds the best fitting line by minimizing the sum of the squares of the perpendicular distances from all data points to the fitting line. Let several final life prediction results within a preset time period be represented as (x1,y1),(x2,y2),...,(x n ,y n ), where x represents the time, x n represents the time value of the nth final life prediction result within the preset time period, y represents the predicted remaining life, and y n represents the predicted remaining life value of the nth final life prediction result within the preset time period, n represents the total number of final life prediction results within the preset time period, and the equation of the life function line is set as y = mx + b, where m represents the slope of the line and b represents the intercept of the line. The goal of the least squares method is to find the m value and b value that minimize the error function :

[0066] .

[0067] Among them, x i represents the time value of the ith final life prediction result within the preset time period, yi The predicted remaining life value representing the i-th final life prediction result within a preset time period.

[0068] In this embodiment, taking the preset time period as 24 hours as an example, 25 final life prediction results are obtained at 1-hour intervals and linearly fitted to generate a life function line. If the error between a certain final life prediction result and the corresponding value in the life function line is greater than a preset threshold (when the time values are equal, the error is the difference in predicted remaining life, which is set to 1 hour in this embodiment), then the final life prediction result is modified to the corresponding value in the life function line, and the corresponding life prediction model is optimized and trained based on the modified final life prediction result (the remaining life prediction result before weighted summation can be inversely modified through the modified final life prediction result) and the corresponding environmental parameter data and state parameter data.

[0069] In the embodiment of the present application, by constructing a life test data set under different environmental parameter test data, combining cluster analysis, machine learning models, and calculating weights according to distances, the non-linear influence of different environmental combinations on the aging of the sealing structure is considered, the relevance between environmental parameters and the training data distribution of the prediction model is improved, and the accuracy and reliability of the sealing structure life prediction are enhanced.

[0070] By pre-training a machine learning model based on each cluster of life test clustering data sets respectively, several life prediction models are obtained, which improves the training pertinence and further enhances the reliability of the life prediction models.

[0071] It is proposed that the initial weight of each type of data in the corresponding life test clustering data set is set using one of the entropy weight method and the coefficient of variation method. By setting the initial weights for each cluster of life test clustering data sets through the weighting method, the rationality of the initial weights is improved, enabling the training process of the life prediction model to converge quickly and enhancing the training efficiency.

[0072] It is proposed to calculate the weight of each remaining life prediction result using the Gaussian kernel function based on the distance between the environmental parameter data and the environmental parameter clustering center of each selected life prediction model, avoiding the risk that a single prediction model may lose valid information, reducing the fluctuation error of the prediction results, and at the same time achieving the non-linear fusion of the prediction results.

[0073] It is proposed to modify the outliers in several final life prediction results to the corresponding values in the life function line, and optimize and train the corresponding life prediction models based on the modified final life prediction results and the corresponding environmental parameter data and state parameter data, improving the evolvability and adaptability of the life prediction models.

[0074] Although the preferred embodiments of the present application have been described, additional changes and modifications can be made to these embodiments by those skilled in the art once they learn the basic creative concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present application.

[0075] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these modifications and variations.

Claims

1. A method for predicting the service life of a sealing structure of an aviation hydraulic pipeline system, characterized in that, It includes the following steps: Step 1: Construct a life test data set under different environmental parameter test data, perform clustering analysis according to environmental parameters, and obtain several clusters of life test clustering data sets; Step 2: Pre-train a machine learning model based on each cluster of life test clustering data sets respectively to obtain several life prediction models; Step 3: Collect environmental parameter data and state parameter data of the sealing structure of the aviation hydraulic pipeline system; Step 4: Calculate the distance between the environmental parameter data and the environmental parameter clustering center corresponding to each life prediction model, and select at least two life prediction models in ascending order of distance; Step 5: Use each selected life prediction model to predict the remaining life according to the environmental parameter data and the state parameter data respectively to obtain at least two remaining life prediction results; Step 6: Calculate the weight of each remaining life prediction result according to the distance between the environmental parameter data and the environmental parameter clustering center of each selected life prediction model, and perform weighted summation to obtain the final life prediction result; In Step 2, each life prediction model adopts the same model architecture; The machine learning model adopts one of the feedforward neural network model, convolutional neural network model, and support vector machine model; In Step 2, when the machine learning model adopts the feedforward neural network model or the convolutional neural network model, set the initial weight of each type of data in the corresponding life test clustering data set before pre-training each machine learning model; The method for setting the initial weight of each type of data in the corresponding life test clustering data set adopts one of the entropy weight method and the coefficient of variation method.

2. The life prediction method of a sealing structure for an aviation hydraulic pipeline system according to claim 1, characterized in that, Step 1 includes: Previously conduct life tests on the sealing structures of aviation hydraulic pipeline systems of the same model under different environmental parameter test data respectively to construct a life test data set under different environmental parameter test data. Each group of data in the life test data set includes environmental parameter test data, state parameter test data at a certain moment, and the corresponding remaining test life; Perform clustering analysis on the life test data set under different environmental parameter test data according to environmental parameters to obtain several clusters of life test clustering data sets.

3. A method for predicting the life of a sealing structure of an aviation hydraulic pipeline system according to claim 1, characterized in that, In Step 1, the clustering analysis adopts one of the K-means clustering algorithm, hierarchical clustering algorithm, and spectral clustering algorithm.

4. The life prediction method of a sealing structure for an aviation hydraulic pipeline system according to claim 1, characterized in that, In Step 3, the environmental parameter data includes: pressure data, temperature data, vibration data, flow data, and the state parameter data includes: pressure difference data and contact stress data.

5. The life prediction method of a sealing structure for an aviation hydraulic pipeline system according to claim 1, characterized in that, In Step 4, use one of the Euclidean distance, Mahalanobis distance, cosine similarity, and Manhattan distance to calculate the distance between the environmental parameter data and the environmental parameter clustering center corresponding to each life prediction model.

6. The life prediction method of a sealing structure for an aviation hydraulic pipeline system according to claim 1, characterized in that In Step 6, use the Gaussian kernel function to calculate the weight of each remaining life prediction result according to the distance between the environmental parameter data and the environmental parameter clustering center of each selected life prediction model.

7. A method for predicting the life of a sealing structure of an aviation hydraulic pipeline system according to claim 1, characterized in that After step six, it further includes: step seven, obtaining several final life prediction results within a preset time period and performing linear fitting to generate a life function line, modifying the outliers in the several final life prediction results to the corresponding values in the life function line, and optimizing and training the corresponding life prediction model according to the modified final life prediction results and the corresponding environmental parameter data and state parameter data; The linear fitting uses the least squares method.

Images