Aero-engine measurement parameter selection method based on clustering method

Abstract

The invention discloses an aero-engine measurement parameter selection method based on a clustering method, and belongs to the technical field of engine state monitoring and fault diagnosis, and the method comprises the steps: building an aero-engine model, and generating a gas path fault sample; obtaining a sample and performing normalization processing to enable the sample to only reflect the percentage of the measurement parameter change in the fault mode; hierarchical clustering is executed under different clustering numbers, and a Davies-Bouldin index is calculated; the clustering number corresponding to the minimum Davies-Bouldin and the clustering result of the minimum Davies-Bouldin are selected as the optimal division; and aligning the optimal division result with a real label, and calculating a consistency index. According to the method, on the premise that key fault information is reserved, the number of required measurement parameters is remarkably reduced, the optimized parameter combination still keeps a high fault recognition rate under the condition of limited sensors, and the efficiency and practicability of state monitoring and fault diagnosis are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of engine state monitoring and fault diagnosis, and particularly to an aero-engine measurement parameter selection method based on a clustering method. BACKGROUND

[0002] In early engine state monitoring and fault diagnosis research, researchers often use principal factor model method and multi-state analysis method to carry out fault diagnosis research under limited measurement parameters based on engine influence coefficient matrix (ICM). The principal factor model method is based on all fault types less than the number of measurement parameters for solving, and sorts the solutions based on correlation and rationality criteria to obtain the most reasonable solution that is least likely to be misjudged. The multi-state analysis method compensates for the lack of working points by increasing the number of working points, but its disadvantage is that the selection of working points is highly dependent on expert experience and cannot quantitatively explain the contribution of increased working points to the gas path fault.

[0003] With the progress of computing technology, researchers have carried out related research on measurement parameter selection based on the information contained in different measurement parameter combinations. For example, using sensitivity and correlation analysis methods to analyze the influence coefficient matrix of gas path fault diagnosis, the most suitable combination of fault diagnosis is selected from redundant measurement parameters. However, this type of "selecting less from more" research method cannot explain the characteristic information contained in quantitative measurement parameters, and if some parameters in the optimal solution of measurement parameter selection are difficult to measure or have low measurement accuracy, the application effect of the method will be reduced.

[0004] Other researches take the results of gas path fault diagnosis methods as indicators to reflect the diagnostic ability of measurement parameters, such as whether the root mean square error decreases after adding measurement parameters. Methods such as including measurement parameter subsets, "brute force selection", and neural network modeling are used to quantitatively explain which measurement parameters are beneficial to gas path fault diagnosis. However, this type of method lacks consideration of whether the characteristics of a certain type of parameter are retained after reducing the number of parameters.

[0005] Based on the above research status, it can be seen that current research on measurement parameter selection optimization often faces specific engineering problems and optimizes selection with pertinence, but lacks iterative optimization with performance evaluation needs, i.e., what degree of fault diagnosis can be achieved with a certain number of measurement parameters, and how many measurement parameters are needed to achieve a certain degree of fault diagnosis. When the number of measurement parameters increases, there may be "redundancy compression" or "dimension folding" phenomenon, and the information of the added measurement parameters cannot be effectively applied; when the number of measurement parameters decreases, the main fault characteristics may be retained, and the retention degree of the characteristic information needs to be quantified. SUMMARY

[0006] The purpose of this invention is to provide a clustering-based method for selecting aero-engine measurement parameters. This method employs hierarchical clustering to perform cluster analysis on sample sets with different combinations of measurement parameters and different fault characteristics, extracts the characteristic manifestations of fault modes in the measurement parameter space, and quantifies the ability of measurement parameter combinations to reflect gas path fault characteristics through changes in clustering indices.

[0007] To achieve the above objectives, this invention provides a method for selecting aero-engine measurement parameters based on clustering, comprising the following steps: S1. Establish an aero-engine model and generate air path fault samples through simulation; S2. Obtain samples and perform normalization processing. Through norm normalization, the samples reflect only the percentage of changes in the measured parameters under this fault mode. S3. Perform hierarchical clustering and calculate the Davies-Bouldin index under different numbers of clusters; S4. Select the minimum number of clusters corresponding to Davies–Bouldin and its clustering result as the best partition; S5. Align the best splitting results with the true labels and calculate the consistency index to evaluate the consistency between the clustering and the true classification.

[0008] Preferably, in S1, the gas path fault samples are generated by constructing fault factors and the ratio of component efficiency to flow capacity to quantitatively describe different types of faults; the ratio of component efficiency to flow capacity is determined by the following formula: ; ; in, FM It serves as a fault factor, acting as a quantitative indicator of the degree of fault. η For component efficiency; γ Represents the flow capacity of components; γ : η ratio This represents the ratio of circulation capacity to efficiency. η adjust and γ adjust The adjustment amount represents efficiency and flow capacity; after determining the failure factor of each component and the ratio of component efficiency to flow capacity, the adjustment range of component efficiency and flow capacity is further determined.

[0009] The preferred formula for implementing L1 norm normalization is as follows: ; Where, vector P Represents the relative value of a set of measured parameters with respect to health status. m Represents the sequence number of the measured parameter, ||P || represents a vector P The norm of 1.

[0010] Preferably, in S3, the Davies-Bouldin index measures the combined performance of intra-cluster sample compactness and inter-cluster separation, and is defined as: ; in, S i For clusters i The average distance from the inner sample to the cluster center. S j For clusters j The average distance from the inner sample to the cluster center. M ij For clusters i with cluster j The distance between the center points K Represents the total number of clusters, numerator S i + S j The smaller the value, the more concentrated the two clusters are within each cluster; the smaller the denominator... M ij The larger the value, the farther apart they are. DBI The smaller the value, the higher the clustering quality.

[0011] Preferably, in S5, the Adjusted Land coefficient (ARI) is used to measure the consistency between the clustering results and the true labels at the sample pair level, and the expected value of random consistency is corrected within the interval [-1,1]. The Clustering Accuracy (ACC) is used to directly measure the degree of matching between the clustering labels and the true labels at the sample level. ARI emphasizes the structural consistency at the "sample pair" level and has the ability to correct for random effects. ACC focuses on the direct matching at the "single sample" level. ARI and ACC complement each other to reflect the consistency between the clustering results and the prior labels.

[0012] Preferably, the standard for conformity assessment using ARI is: ; in, n ij They are simultaneously gathered into clusters i And it truly belongs to the class j The number of samples, a i = ∑ j n ij The first one represents the "real label" division. i The number of samples in each cluster b j = ∑i n ij Represents the first division under "clustering results" j The number of samples in each cluster n The total number of samples; an ARI value of 1 indicates complete consistency, 0 indicates no difference from random partitioning, and a negative value indicates worse than random partitioning.

[0013] Preferably, the ACC measure of the degree of matching between cluster labels and true labels at the sample level is specifically as follows: first, the cluster labels are aligned using the optimal one-to-one mapping π, and then the accuracy is calculated. ; in, y k For the first k The true label of each sample c k is the cluster label, and 1[.] is the indicator function.

[0014] Therefore, this invention employs the aforementioned clustering-based method for selecting aero-engine measurement parameters. By optimizing the number of clusters through hierarchical clustering and the Davies-Bouldin index, and combining ARI and ACC indices to evaluate the consistency between clusters and actual fault labels, it can effectively quantify the ability of different combinations of measurement parameters to distinguish gas path fault characteristics. This method significantly reduces the number of required measurement parameters while retaining key fault information. The optimized parameter combinations maintain a high fault identification rate even under limited sensor conditions, improving the efficiency and practicality of condition monitoring and fault diagnosis.

[0015] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0016] Figure 1 This is a flowchart of an embodiment of the aero-engine measurement parameter selection method based on clustering method of the present invention; Figure 2 This is a schematic diagram of the flow path of a small bypass ratio turbofan engine, which is an embodiment of the clustering method for selecting aero-engine measurement parameters according to the present invention. Figure 3 This is an example of the variation law of DBI value with the number of cluster categories in the embodiment of the aero-engine measurement parameter selection method based on clustering method of the present invention; Figure 4 This is a bar chart showing the statistical analysis of clustering results when reducing measurement parameters in an embodiment of the aero-engine measurement parameter selection method based on clustering method of the present invention; Figure 5 This is a heatmap of statistical analysis of clustering results when reducing measurement parameters, according to an embodiment of the aero-engine measurement parameter selection method based on clustering method of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0018] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0019] Example 1 This invention provides a method for selecting aero-engine measurement parameters based on clustering, the process of which is as follows: Figure 1 As shown, it includes the following steps: S1. Establish an aero-engine model and generate airflow fault samples through simulation. The airflow fault samples are generated by constructing fault factors and the ratio of component efficiency to flow capacity to quantitatively describe different types of faults. The ratio of component efficiency to flow capacity is determined by the following formula: ; ; in, FM It serves as a fault factor, acting as a quantitative indicator of the degree of fault. η For component efficiency; γ Represents the flow capacity of components; γ : η ratio This represents the ratio of circulation capacity to efficiency. η adjust and γ adjust The adjustment amount represents efficiency and flow capacity; after determining the failure factor of each component and the ratio of component efficiency to flow capacity, the adjustment range of component efficiency and flow capacity is further determined.

[0020] S2. Obtain samples and perform normalization. L1 normalization ensures the samples only reflect the percentage change in measured parameters under this fault mode. The formula for L1 normalization is as follows: ; Where, vector P Represents the relative value of a set of measured parameters with respect to health status. m Represents the sequence number of the measured parameter, || P || represents a vector P The norm of 1.

[0021] S3. Perform hierarchical clustering at different numbers of clusters and calculate the Davies-Bouldin (DBI) index. The DBI index measures the combined performance of intra-cluster density and inter-cluster separation. A lower DBI value indicates more concentrated intra-cluster samples and more significant inter-cluster differences, resulting in better clustering performance. It is defined as: ; in, S i For clusters i The average distance from the inner sample to the cluster center. S j For clusters j The average distance from the inner sample to the cluster center. M ij For clusters i with cluster j The distance between the center points K Represents the total number of clusters, numerator S i + S j The smaller the value, the more concentrated the two clusters are within each cluster; the smaller the denominator... M ij The larger the value, the farther apart they are. DBI The smaller the value, the higher the clustering quality.

[0022] S4. Select the minimum number of clusters corresponding to Davies–Bouldin and its clustering result as the best partition; S5. Align the best splitting results with the true labels and calculate the consistency index to evaluate the consistency between clustering and the true classification. Specifically, the adjusted Land coefficient (ARI) is used to measure the consistency between the clustering results and the true labels at the sample pair level, and the expected value of random consistency is corrected within the interval [-1,1]. The clustering accuracy (ACC) is used to directly measure the degree of matching between the clustering labels and the true labels at the sample level. ARI emphasizes the structural consistency at the "sample pair" level and has the ability to correct for random effects. ACC focuses on the direct matching at the "single sample" level. ARI and ACC complement each other to reflect the consistency between the clustering results and the prior labels.

[0023] The standard for conformity assessment using ARI is as follows: ; in,n ij They are simultaneously gathered into clusters i And it truly belongs to the class j The number of samples, a i = ∑ j n ij The first one represents the "real label" division i The number of samples in each cluster b j = ∑ i n ij Represents the first division under "clustering results" j The number of samples in each cluster n The total number of samples; an ARI value of 1 indicates complete consistency, 0 indicates no difference from random partitioning, and a negative value indicates worse than random partitioning.

[0024] Accuracy Analytical Profile (ACC) measures the degree of match between cluster labels and true labels at the sample level. Specifically, it first aligns the cluster labels using the optimal one-to-one mapping π, and then calculates the accuracy. ; in, y k For the first k The true label of each sample c k is the cluster label, and 1[.] is the indicator function.

[0025] To verify the actual effectiveness of this method, a system was established as follows: Figure 2 The figure shows a small bypass ratio turbofan engine with a flow path diagram and the baseline measurement parameters related to this model are selected, as shown in Table 1.

[0026] Table 1. Definition of Baseline Measurement Parameters

[0027] This embodiment selects 15 typical air circuit fault combinations of the engine, including single faults of components such as fan, compressor, high-pressure turbine and low-pressure turbine, as well as combined faults involving 2 to 4 components. γ:η The ratio is set to 1.5 in the compression component and -0.75 in the turbine component. The selected operating condition is the ground design point. The fault severity range is set to 0.5% to 7%, with sampling in 0.5% increments. Under a given fault mode library, cluster analysis is performed on the "baseline measurement parameter scheme." Based on the scheme's ability to differentiate within the current fault library, research is conducted to reduce the impact of measurement parameters on the clustering effect; that is, while maintaining clustering performance, the minimum number and combination of required measurement parameters are explored.

[0028] Under a single operating condition and 15 fault modes, the variation of DBI with the number of clusters is as follows: Figure 3 As shown in Table 2, the clustering results corresponding to the minimum DBI, i.e. the optimal clustering, are shown in Table 2. The predicted clusters are aligned with the true categories through the optimal mapping, and the accuracy of the hierarchical clustering prediction is also given.

[0029] from Figure 3 It can be seen that, under a single operating condition and 15 fault modes, the optimal classification result of the baseline measurement parameters corresponds to 16 clusters. Analysis of the predicted clustering results shows that in the real fault database, the 7th type of fault is divided into two subclusters; however, all samples within each subcluster are correctly classified. Therefore, the baseline measurement parameters can effectively distinguish all fault types. When the number of clusters is set to 15, the 7th type of fault, which was originally divided into two subclusters when the number of clusters was 16, is merged. Therefore, it can be further concluded that, in the clustering results, although a real category may be subdivided into two subclusters, if these two subclusters completely correspond to the same real category (and their respective accuracies are both 1), then this subdivision reflects not misclassification, but measurable differences within that category. Thus, the measurement parameters can not only effectively distinguish different fault types, but also reveal potential subclass characteristics within the same fault category.

[0030] Table 2. Clustering Results Analysis Corresponding to the Minimum DBI ;

[0031] Depend on Figure 3 As shown in Table 2, the five baseline measurement parameters exhibit strong gas path fault discrimination capabilities under the existing fault database. Considering the characteristics of the fault database, these parameters still have potential for simplification. This embodiment further investigates maintaining fault discrimination capabilities while reducing the number of measurement parameters. The results are shown in the table below: Table 3 Clustering results when reducing measurement parameters

[0032] The results show that, starting from the baseline measurement parameters, the fault differentiation capability gradually decreases as the number of parameters decreases. When one parameter is removed (leaving only 4 measurement parameters), the parameter set can no longer effectively distinguish between fault type 7 and fault type 14, which correspond to two-component faults and three-component faults, respectively. When two parameters are removed (leaving only 3 measurement parameters), further confusion arises between fault types 8, 10, and 12, as well as between fault types 4 and 10. When only two measurement parameters are retained, only the combination of "1 and 5" can still accurately distinguish all fault types in the current fault database.

[0033] Statistical analysis was performed on all combinations that achieved ACC=1. Figure 4 The results show the frequency with which each feature was retained when all ACC=1. Feature 5 appeared most frequently (9 times), followed by feature 1 (7 times), features 2 and 3 each appeared 5 times, and feature 4 appeared the least (4 times). This indicates that, in achieving perfect consistency, feature 5 is the most critical, followed by feature 1, features 2 and 3 are of moderate importance, while feature 4 is relatively weak.

[0034] Figure 5 A heatmap of ACC=1 success coverage stratified by "number of retained features" is presented. The horizontal axis represents the feature number, and the vertical axis represents the actual number of features retained in each analysis (e.g., 4, 3, or 2). Each cell represents the proportion (or frequency) of that feature appearing in all ACC=1 success cases under the corresponding retention constraint. Horizontal comparison can be used to assess the relative robustness of features at the same retention level; vertical comparison is used to examine whether a feature still maintains a high proportion when the number of retained features decreases, thus reflecting its persistence and necessity. The lighter the color (the higher the value), the more frequently the feature appears in success cases under the corresponding constraint. The results show that when 4 features are retained, features 1-4 all appear in 75% of all cases, while feature 5 appears in 100%. When 3 features are retained, feature 5 still maintains 100%, feature 1 is 75%, features 2 and 3 are 50%, and feature 4 is only 25%. When only 2 features are retained, ACC=1 is only achieved when features 1 and 5 are included simultaneously; the coverage of all other features at this level is 0%.

[0035] In summary, regardless of the number of retained features, feature 5 remains central. Under conditions of highly compressed feature counts, retaining both features 1 and 5 simultaneously is particularly important. Features 2 and 3 exhibit moderate importance, while feature 4 is most prone to failure when the number of features decreases. For scenarios with a limited number of sensors / features, priority should be given to including feature 5, and it should be retained along with feature 1 whenever possible to maintain consistency in clustering results.

[0036] Based on the above analysis, it can be concluded that the baseline measurement parameter scheme demonstrates high effectiveness in distinguishing the 15 typical faults. When the number of measurement parameters is limited, the low-pressure rotor physical speed should be retained first to maximize the distinguishability of fault modes. Furthermore, it is recommended to retain at least three measurement parameters where conditions permit to ensure that as many fault types as possible can be distinguished even under information-limited circumstances.

[0037] Therefore, this invention employs the aforementioned clustering-based method for selecting aero-engine measurement parameters. By optimizing the number of clusters through hierarchical clustering and the Davies-Bouldin index, and combining ARI and ACC indices to evaluate the consistency between clusters and actual fault labels, it can effectively quantify the ability of different combinations of measurement parameters to distinguish gas path fault characteristics. This method significantly reduces the number of required measurement parameters while retaining key fault information. The optimized parameter combinations maintain a high fault identification rate even under limited sensor conditions, improving the efficiency and practicality of condition monitoring and fault diagnosis.

[0038] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for selecting aero-engine measurement parameters based on clustering, characterized in that: Includes the following steps: S1. Establish an aero-engine model and generate air path fault samples through simulation; S2. Obtain samples and perform normalization processing. Through norm normalization, the samples reflect only the percentage of changes in the measured parameters under this fault mode. S3. Perform hierarchical clustering and calculate the Davies-Bouldin index under different numbers of clusters; S4. Select the minimum number of clusters corresponding to Davies–Bouldin and its clustering result as the best partition; S5. Align the best splitting results with the true labels and calculate the consistency index to evaluate the consistency between the clustering and the true classification.

2. The method for selecting aero-engine measurement parameters based on clustering as described in claim 1, characterized in that: In S1, the gas path fault samples are generated by constructing fault factors and the ratio of component efficiency to flow capacity to quantitatively describe different types of faults; the ratio of component efficiency to flow capacity is determined by the following formula: ； ； in, FM It serves as a fault factor, acting as a quantitative indicator of the degree of fault. η For component efficiency; γ Represents the flow capacity of components; γ : η ratio Represents the ratio of circulation capacity to efficiency; η adjust and γ adjust The adjustment amount represents efficiency and flow capacity; after determining the failure factor of each component and the ratio of component efficiency to flow capacity, the adjustment range of component efficiency and flow capacity is further determined.

3. The method for selecting aero-engine measurement parameters based on clustering as described in claim 2, characterized in that: In S2, the formula for implementing the first norm normalization is as follows: ； Where, vector P Represents the relative value of a set of measured parameters with respect to health status. m Represents the sequence number of the measured parameter, || P || represents a vector P The norm of 1.

4. The method for selecting aero-engine measurement parameters based on clustering as described in claim 3, characterized in that: In S3, the Davies-Bouldin index measures the combined performance of intra-cluster sample compactness and inter-cluster separation, and is defined as: ； in, S i For clusters i The average distance from the inner sample to the cluster center. S j For clusters j The average distance from the inner sample to the cluster center. M ij For clusters i with cluster j The distance between the center points K Represents the total number of clusters, numerator S i + S j The smaller the value, the more concentrated the two clusters are within each cluster; the smaller the denominator... M ij The larger the value, the farther apart they are. DBI The smaller the value, the higher the clustering quality.

5. The method for selecting aero-engine measurement parameters based on clustering as described in claim 4, characterized in that: In S5, the Adjusted Land coefficient (ARI) is used to measure the consistency between the clustering results and the true labels at the sample pair level, and the expected value of random consistency is corrected within the interval [-1,1]. The Clustering Accuracy (ACC) is used to directly measure the degree of matching between the clustering labels and the true labels at the sample level. ARI emphasizes the structural consistency at the "sample pair" level and has the ability to correct for random effects. ACC focuses on the direct matching at the "single sample" level. ARI and ACC complement each other to reflect the consistency between the clustering results and the prior labels.

6. The method for selecting aero-engine measurement parameters based on clustering as described in claim 5, characterized in that: The standard for conformity assessment using ARI is as follows: ； in, n ij They are simultaneously gathered into clusters i And it truly belongs to the class j The number of samples, a i = ∑ j n ij The first one represents the "real label" division. i The number of samples in each cluster b j = ∑ i n ij Represents the first division under "clustering results" j The number of samples in each cluster n The total number of samples; an ARI value of 1 indicates complete consistency, 0 indicates no difference from random partitioning, and a negative value indicates worse than random partitioning.

7. The method for selecting aero-engine measurement parameters based on clustering as described in claim 5, characterized in that: Accuracy Analytical Profile (ACC) measures the degree of match between cluster labels and true labels at the sample level. Specifically, it first aligns the cluster labels using the optimal one-to-one mapping π, and then calculates the accuracy. ； in, y k For the first k The true label of each sample c k is the cluster label, and 1[.] is the indicator function.

Images