Cluster analysis method of turboshaft engine load spectrum based on three-dimensional K-Means

Abstract

The present invention discloses a cluster analysis method for load spectrum of turboshaft engine based on three-dimensional K-Means, comprising the following steps: step (1), pre-processing the original speed spectrum and height spectrum of n sections; step (2), characteristic analysis of parameter spectrum of civil turboshaft engine; step (3), establishment of three-dimensional K-Means model based on typical mission segment and division criteria of typical sections. The present invention provides an important basis for life determination and life extension of aeroengine, and provides an effective pre-processing method for strength analysis of aeroengine and compilation of comprehensive mission spectrum.

Description

Technical Field

[0001] The present invention relates to a method for cluster analysis of aero-engine load spectrum, and in particular to a method for cluster analysis of firefighting load spectrum of transport helicopter aero-engine. It is extremely important to carry out research on aero-engine load spectrum, which is a combination of important load elements such as flight altitude, flight Mach number, flight speed, normal overload, throttle lever angle, turbine after-gas temperature, etc. The recording and analysis of these important flight parameter data can provide important basis for determining and extending the life of aero-engines. Background Art

[0002] The flight mission profile corresponding to the aircraft engine load spectrum contains the flight information of the engine. It is the specific data generated by the aircraft during actual flight and reflects the actual use of the engine. However, due to the large number and disorder of flight mission profiles, it is difficult to find the regularity of similar profiles. In view of this situation, the concept of clustering needs to be mentioned.

[0003] Birds of a feather flock together. As the name implies, profile clustering refers to grouping together profiles that are not very different to form multiple categories of profiles. Clustering was originally a statistical concept, but now it belongs to the category of unsupervised learning in machine learning. It is mostly used in the fields of data mining and data analysis. It can be simply summarized in one word - birds of a feather flock together. By definition, clustering is to study classification methods for a large amount of data or samples based on the characteristics of the data itself, and to reasonably classify the data according to this classification method, and finally group similar data into one group, that is, the same kind of data is the same, and different kinds of data are different. The difference between clustering and classification is that the division criteria of clustering are multiple, and the algorithm must be used to judge the similarity between data. Before the conclusion of clustering comes out, scholars have no idea what the characteristics of each category are. The essence of cluster analysis of aero-engines is the automatic classification of measured engine data. In the past, measured data were processed by manual classification. As more and more measurable engine parameters and more and more data are available, aero-engine scholars are more inclined to use clustering methods to classify the data first and then perform feature analysis of the data when processing data. This method greatly improves the efficiency and accuracy of scholars in processing data, thereby laying a solid foundation for subsequent damage analysis and spectrum compilation.

[0004] There are various forms of clustering algorithms. As for which clustering algorithm to use in the end, it needs to be decided according to the purpose of clustering and the type of data. If cluster analysis is used as a description and exploration tool, multiple algorithms can be tried on the same type of data to reveal that different clusters of data produce different results. However, although there are many different clustering algorithms, they can be roughly divided into the following categories: K-Means clustering algorithm, hierarchical clustering algorithm, FCM clustering algorithm, DBSCAN clustering algorithm, GMM clustering algorithm, SOM clustering algorithm.

[0005] For aircraft engines, in the prior art:

[0006] Ren Jibin (Ren Jibin; Study on Typical Flight Mission Profile of Aircraft Fatigue Load Spectrum Based on Flight Parameter Sensitivity; 2008) used principal component analysis to select pitch angular velocity, normal overload of center of gravity, flight altitude, aircraft weight and flight speed, which are more sensitive to normal overload, as the initial reference parameters for the classification of a certain aircraft flight mission profile. Then, a dendrogram was formed by the center of gravity method in the hierarchical clustering method for corresponding cluster analysis.

[0007] Song Yingdong (Song Yingdong; Principal Component Clustering Method for Engine Flight Mission Profile; 2002) listed the method of classifying the flight mission profile of aircraft engines through principal component analysis, and conducted corresponding clustering analysis on 18 flight mission profiles of a fighter engine. The flight altitude, flight Mach number, engine speed and engine center of gravity normal overload were found as the main parameters for classification. The center of gravity method was used for quantitative clustering.

[0008] Ren Jibin and Song Yingdong both used the principal component clustering method in the hierarchical clustering method. That is, the principal component analysis is performed first, and then the cluster analysis. The so-called cluster analysis method is to find samples with a high degree of fit between samples according to several observation indicators of the given samples, classify them, and then classify the remaining samples with high fit into one category. The principal component analysis method is a way to find comprehensive indicators and analyze multi-indicator data with correlation. Therefore, to carry out the merger of flight mission profiles through the principal component clustering method, we must first use the principal component analysis method to compress the multi-indicators in the flight mission profile as much as possible, and then use the cluster analysis method to carry out the cluster analysis of the flight mission profile according to the new indicators obtained by the principal component analysis method. But in the end, quantitative clustering is carried out through the centroid method of hierarchical clustering, which is cumbersome in programming and not conducive to batch processing.

[0009] Du Yufei (Du Yufei; Research on the Cluster Analysis Method and Compilation of Aero Engine Load Spectrum; 2016) selected four parameters based on the principal component analysis method, and used the height as the primary partition parameter, and the Mach number parameter as the auxiliary. Then, the flight altitude and Mach number of 31 flight profiles of a fighter engine were averaged using the K-means clustering idea to form a cluster scatter plot, and a preliminary division was made based on the distance of the scattered points. Finally, the densely populated areas were further divided based on the profile characteristics. Although the military aircraft profile was divided, the division process was too cumbersome, requiring frequent manual operations and an overall understanding of the numerous data, and the task training outline was not obtained. The load spectrum cluster analysis method was studied with the profile as the basic unit, which was not accurate enough.

[0010] With the increase of aviation engine load parameter data, data processing methods are becoming more and more diversified. Cluster analysis can intelligently classify data according to a certain goal when processing data. This patent is based on the typical mission segments of helicopters. On the basis of the most classic K-Means algorithm, a three-dimensional K-Means clustering algorithm is developed. According to the mission characteristics of firefighting helicopters, two types of typical mission segments, execution missions and cruise missions, are extracted. According to the spectral characteristics and correlation of the turboshaft engine speed spectrum and altitude spectrum, based on 6 types of typical mission segments, the mission profile is divided into 3D K-Means clustering, and 4 types of typical mission profiles are divided, which lays the foundation for subsequent spectrum compilation. Summary of the invention

[0011] In order to solve the problems in the prior art, the present invention provides a three-dimensional K-Means based turboshaft engine load spectrum clustering analysis method, which is an analysis method for clustering load spectra of civil turboshaft engines. The method is relatively intuitive and concise, and solves the problems of cumbersome programming and rough division in cluster analysis, and provides a basis for the strength analysis of aircraft engines and the compilation of comprehensive mission spectra, and provides an important basis for the life determination and life extension of aircraft engines.

[0012] In order to achieve the above object, the present invention adopts the following technical solutions:

[0013] A three-dimensional K-Means turboshaft engine load spectrum clustering analysis method, characterized by comprising the following steps:

[0014] Step (1), preprocessing the original speed spectra and height spectra of n sections;

[0015] Step (2), characteristic analysis of civil turboshaft engine parameter spectrum;

[0016] Step (3): Establishment of a three-dimensional K-Means model based on typical task segments and division criteria of typical sections.

[0017] Furthermore, the specific steps of step (1) are:

[0018] Step (11) determines the division criteria of the take-off segment and the landing segment, such as Figure 1 As shown,

[0019] Step (12) divides the n speed profiles into take-off and landing sections according to the take-off and landing section division criteria of step (11).

[0020] Step (13) filters the n speed intermediate task segments,

[0021] Furthermore, the division criteria of the take-off segment and the landing segment in step (11) are:

[0022] Criteria for dividing the takeoff segment and the intermediate flight segment: the time point t1 when the engine speed changes from 0 to the maximum value for the first time and then drops from the maximum value to 5% of the maximum speed, and the time point t2 when the engine altitude changes from 0 to the maximum value for the first time. The smaller time value of t1 and t2 is taken as the separation point of the takeoff segment and the intermediate flight segment;

[0023] The criteria for dividing the intermediate flight segment and the landing segment are as follows: the time point t3 when the engine speed starts to drop from the maximum value and never reaches the maximum value again during the descent; the time point t4 when the engine altitude starts to drop from a certain value and there is no obvious increase in altitude during the process; the larger time value between t3 and t4 is taken as the separation point between the intermediate flight segment and the landing segment.

[0024] Furthermore, the specific steps of step (13) are:

[0025] Step (131) determines the filtering thresholds of n intermediate task segments. The general filtering method is to remove noise from the signal. After collecting the peak and valley values ​​of the load parameters, filtering operations are performed in the following two aspects: (1) small range filtering; (2) small peak filtering, thereby removing small cycles that have little or no effect on fatigue damage;

[0026] First, the peak and valley values ​​of the intermediate task segment are extracted, and the calculation formula of the filter threshold Δ% is:

[0027] Δ%=G max *ω% (1)

[0028] Where: G max is the maximum amplitude of each cycle in the load history; ω is an empirical constant. For aircraft engines, the value range of the empirical constant ω is 8.0 to 12.5. For n firefighting mission profiles, the value range of the empirical constant ω is 4.5 to 8.5.

[0029] Step (132) calculates the filtering thresholds of n speed intermediate task segments respectively based on step (131), such as Figure 2(a)-2(i) As shown;

[0030] Step (133) writes the filtering thresholds of the filtering and n speed intermediate task segments into the rain flow count, and filters the n speed intermediate task segments.

[0031] Furthermore, the specific steps of step (2) are:

[0032] The flight action of a helicopter can be decomposed into the basic flight actions of level flight, inclined climb, vertical climb, hovering (with or without ground effect) and maneuvering flight. The flight mission of a helicopter is a combination of basic flight actions. According to the changing characteristics of the helicopter flight mission, it can be divided into the take-off segment, the intermediate mission segment and the landing segment. Among them, the characteristics of the take-off segment and the landing segment are fixed, and they are deterministic mission segments; the intermediate mission segment is directly related to the flight mission performed by the helicopter. When the helicopter performs different flight missions, the take-off segment and the landing segment are consistent, and the intermediate mission segment will be significantly different, that is, the intermediate mission segment of the mission profile reflects the mission characteristics of the helicopter. The changes in the intermediate mission segment spectrum of the helicopter speed spectrum and altitude spectrum correspond one-to-one to the actions performed by the helicopter, and the changes in the speed spectrum and altitude spectrum also correspond one-to-one. When the speed spectrum climbs from C to segment D and then descends to C, the speed spectrum segment is consistent with the changes in the altitude spectrum from altitude a to altitude b and finally descending to altitude c, and they correspond one-to-one. The changes in the altitude spectrum and the speed spectrum have the same rules, and the changes in the speed spectrum and the altitude spectrum are repeated in a typical segment as a unit. It can be found that when the speed spectrum climbs from C to D and then drops to C, Figure 3 The figure shows a fragment of the speed spectrum, which is basically consistent with the change in the altitude spectrum from altitude a to altitude b and finally to altitude c, and corresponds one to one, as shown in the figure. Figure 4 Shown with Figure 3 The corresponding height spectrum fragment. Figure 5 The figure shows the segments of the rotation speed spectrum and the height spectrum. The changes of the height spectrum and the rotation speed spectrum of segments 1 to 4 have the same rules, and the changes of the rotation speed spectrum and the height spectrum are repeated in a typical segment. The rotation speed spectrum and the height spectrum of segment 1 are similar to those of segment 2, with the same duration and height. The rotation speed spectrum and the height spectrum of segment 3 are similar to those of segment 4 in terms of spectrum type, duration and height. The change rules of segments 1 to 2 and segments 3 to 4 in the rotation speed spectrum and height spectrum are basically the same, and the difference lies in the duration and the maximum value of the height spectrum.

[0033] Furthermore, the specific steps of step (3) are:

[0034] Based on the cluster analysis of military aircraft engine load spectra, a clustering algorithm for civil turboshaft engine load spectra was developed. Through the above analysis, it was found that the speed spectrum and altitude spectrum have the same change law and the altitude spectrum is simpler and easier to identify. For the clustering of the measured spectrum of civil aircraft turboshaft engines, the K-Means clustering algorithm is used:

[0035] In this example, the parameter height is used as the main parameter, and the segments of each of the n sections are classified through intelligent recognition, where: H is the maximum height of the segment, in m, and H1, H2, and H3 are the three types of segments divided according to the height of each section;

[0036]

[0037] Based on the duration of each segment, the three segments of H1, H2, and H3 of each section are further divided through intelligent recognition, where: Δt represents the duration of each segment, in seconds, and t1, t2, t3, and t4 represent the four types of segments divided according to the duration of each segment;

[0038]

[0039] Finally, all the segments of the intermediate mission sections of all profiles are divided into two categories of typical mission sections, namely, typical mission sections of execution missions and typical mission sections of cruise missions.

[0040] Furthermore, the characteristic of the typical task segment of the task execution is that the duration of the typical task segment of the task execution is no more than 1400s, such as Figure 6(a)-6(c) As shown, the speed spectra of the typical task segments of the execution task are the same, but the duration is different. The Δt1 of the typical task segment 1 is 300-500, the Δt2 of the typical task segment 2 is 500-800, and the Δt3 of the typical task segment 3 is 800-1400.

[0041] The duration of typical mission segments of cruise missions is greater than 1400s. Figure 7(a)-7(c) The typical segments of the typical mission segments of the cruise mission are shown, with H1 = 300-700 for the typical mission segment 4, H2 = 700-1000 for the typical mission segment 5, and H3 > 1000 for the typical mission segment 6;

[0042] Furthermore, according to the six typical mission segments, the n mission profiles are intelligently identified, the number of execution mission segments and cruise mission segments in each profile is recorded, and the largest number of typical mission segments are extracted. The number of cruise segments-the number of execution segments-the number of typical mission segments extracted are used as the three-dimensional coordinate axis, and the characteristics of the n mission profiles of the turboshaft engine are represented by a three-dimensional graph, as shown in FIG. Figure 8 As shown in FIG. 1 , according to the distribution of data in the three-dimensional diagram, the speed spectrum is divided into three parts: typical section a, typical section b and typical section c, wherein: typical section a is a combination of typical task segment 1 and typical task segment 2; typical section b is a combination of 6 types of typical task segments mainly based on typical task segment 3 or typical task segment 4 or typical task segment 5 or typical task segment 6; typical section c is a combination of 6 types of typical task segments mainly based on typical task segment 1 or typical task segment 2, as shown in FIG. Figure 9(a)-9(c) Three typical profiles are shown.

[0043] Furthermore, by specifically analyzing the n firefighting profiles, based on the typical characteristics of helicopters performing firefighting missions, it can be found that most mission profiles are composed of typical mission segment 1 and typical mission segment 2, namely typical profile a. Typical profile c is mostly composed of typical mission segment 6 and typical mission segment 1 or typical mission segment 2. There are two types of typical profiles b, namely, the cruise profile and the cruise profile with typical mission segment 3. The typical profile b is further subdivided, as shown in Figure 1. Fig.10 As shown in Figure 2, the profiles with only cruise missions are divided into a category named d, such as Fig.11 As shown, the typical section d is obtained by re-dividing the typical section b.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] (1) Simple and intuitive, with precise description;

[0046] The typical task segment is used as the clustering unit, and the load spectrum is divided by analyzing the correlation of the typical task segment of the parameter spectrum. The operation is simple, the division is more accurate, and the versatility is stronger.

[0047] (2) It has extensive engineering application value;

[0048] The load spectrum division method proposed by the present invention is simple to operate and has strong versatility. Therefore, under existing technical conditions, the load profile of a turboshaft engine can be reasonably analyzed and divided by using the present invention, which has a wide range of engineering application value.

[0049] (3) Develop a cluster analysis method for military aircraft load spectra;

[0050] The clustering analysis method of turboshaft engine load spectrum is to divide the helicopter load spectrum profile as reasonably and accurately as possible, mainly based on the actual use mission of the helicopter. This method is also applicable to the clustering division of military aircraft load profiles.

[0051] The present invention provides an important basis for determining the life span and extending the life span of an aircraft engine, and provides an effective preprocessing method for strength analysis and compilation of a comprehensive mission spectrum of an aircraft engine. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a combination of the speed spectrum and the altitude spectrum, including: ① take-off part, ② intermediate flight part, ③ landing part;

[0053] Figure 2(a) is the filtered spectrum after filtering the middle task segment of the speed spectrum, and its section is 0108. Figure 2(b) is the filtered spectrum after filtering the middle task segment of the speed spectrum, and its section is 0318. Figure 2(c) is the filtered spectrum after filtering the middle task segment of the speed spectrum, and its section is 0813. Figure 2(d) is the filtered spectrum after filtering the middle task segment of the speed spectrum, and its section is 0826. Figure 2(e) is the filtered spectrum after filtering the middle task segment of the speed spectrum, and its section is 0827. Figure 2(f) is the filtered spectrum after filtering the middle task segment of the speed spectrum, and its section is 0828. Figure 2(g) is the filtered spectrum after filtering the middle task segment of the speed spectrum, and its section is 0904. Figure 2(h) is the filtered spectrum after filtering the middle task segment of the speed spectrum, and its section is 1028. Figure 2(i) is the filtered spectrum after filtering the middle task segment of the speed spectrum, and its section is 1029.

[0054] Figure 3 is a segment of the rotation speed spectrum, where: C is hovering for water collection, D is cruising, E is descending, F is hovering for firefighting, and C is hovering for water collection;

[0055] Figure 4 is a height spectrum fragment, where: a is the water intake area, b is the fire fighting area, and c is the water intake area;

[0056] Figure 5 This is the rotation speed spectrum and altitude spectrum fragment of profile 0108;

[0057] Figure 6(a) shows a typical task segment 1 of the execution task, Figure 6(b) shows a typical task segment 2 of the execution task, and Figure 6(c) shows a typical task segment 3 of the execution task;

[0058] FIG7(a) is a typical mission segment 4 of a cruise mission, FIG7(b) is a typical mission segment 5 of a cruise mission, and FIG7(c) is a typical mission segment 6 of a cruise mission;

[0059] Figure 8 It is a preliminary division of typical sections;

[0060] Figure 9(a) is a typical cross section a, Figure 9(b) is a typical cross section b, and Figure 9(c) is a typical cross section c;

[0061] Fig.10 There are four typical profiles;

[0062] Fig.11 It is a typical section d. DETAILED DESCRIPTION

[0063] The present invention will be further described below in conjunction with embodiments.

[0064] Example 1

[0065] A three-dimensional K-Means turboshaft engine load spectrum clustering analysis method, characterized by comprising the following steps:

[0066] Step (1), preprocessing the original speed spectra and height spectra of n sections;

[0067] The specific steps of step (1) are:

[0068] Step (11) determines the division criteria of the take-off segment and the landing segment, such as Figure 1 As shown,

[0069] The division criteria of the take-off segment and the landing segment in step (11) are:

[0070] Criteria for dividing the takeoff segment and the intermediate flight segment: the time point t1 when the engine speed changes from 0 to the maximum value for the first time and then drops from the maximum value to 5% of the maximum speed, and the time point t2 when the engine altitude changes from 0 to the maximum value for the first time. The smaller time value of t1 and t2 is taken as the separation point of the takeoff segment and the intermediate flight segment;

[0071] The criteria for dividing the intermediate flight segment and the landing segment are as follows: the time point t3 when the engine speed starts to decrease from the maximum value and never reaches the maximum value again during the descent process, and the time point t4 when the engine altitude starts to decrease from a certain value and there is no obvious altitude increase during the process. The larger time value between t3 and t4 is taken as the separation point between the intermediate flight segment and the landing segment.

[0072] Step (12) divides the n speed profiles into take-off and landing sections according to the take-off and landing section division criteria of step (11).

[0073] Step (13) filters the n speed intermediate task segments,

[0074] The specific steps of step (13) are:

[0075] Step (131) determines the filtering thresholds of n intermediate task segments. The general filtering method is to remove noise from the signal. After collecting the peak and valley values ​​of the load parameters, filtering operations are performed in the following two aspects: (1) small range filtering; (2) small peak filtering, thereby removing small cycles that have little or no effect on fatigue damage;

[0076] First, the peak and valley values ​​of the intermediate task segment are extracted, and the calculation formula of the filter threshold Δ% is:

[0077] Δ%=G max *ω% (1)

[0078] Where: G max is the maximum amplitude of each cycle in the load history; ω is an empirical constant. For aircraft engines, the value range of the empirical constant ω is 8.0 to 12.5. For n firefighting mission profiles, the value range of the empirical constant ω is 4.5 to 8.5.

[0079] Step (132) calculates the filtering thresholds of n speed intermediate task segments respectively based on step (131), such as Figure 2(a)-2(i) As shown;

[0080] Step (133) writes the filtering thresholds of the filtering and n speed intermediate task segments into the rain flow count, and filters the n speed intermediate task segments;

[0081] Step (2), characteristic analysis of civil turboshaft engine parameter spectrum;

[0082] The specific steps of step (2) are:

[0083] The mission characteristics of a helicopter can be considered as the specific differences in the mission profile when the helicopter performs different missions. Different flight missions have different mission characteristics.

[0084] The flight action of a helicopter can be decomposed into the basic flight actions of level flight, inclined climb, vertical climb, hovering (with or without ground effect) and maneuvering flight. The flight mission of a helicopter is a combination of basic flight actions. According to the changing characteristics of the helicopter flight mission, it can be divided into the take-off segment, the intermediate mission segment and the landing segment. Among them, the characteristics of the take-off segment and the landing segment are fixed, and they are deterministic mission segments; the intermediate mission segment is directly related to the flight mission performed by the helicopter. When the helicopter performs different flight missions, the take-off segment and the landing segment are consistent, and the intermediate mission segment will be significantly different, that is, the intermediate mission segment of the mission profile reflects the mission characteristics of the helicopter. The changes in the intermediate mission segment spectrum of the helicopter speed spectrum and altitude spectrum correspond one-to-one to the actions performed by the helicopter, and the changes in the speed spectrum and altitude spectrum also correspond one-to-one. When the speed spectrum climbs from C to segment D and then descends to C, the speed spectrum segment is consistent with the changes in the altitude spectrum from altitude a to altitude b and finally descending to altitude c, and they correspond one-to-one. The changes in the altitude spectrum and the speed spectrum have the same rules, and the changes in the speed spectrum and the altitude spectrum are repeated in a typical segment as a unit. It can be found that when the speed spectrum climbs from C to D and then drops to C, Figure 3 The figure shows a fragment of the speed spectrum, which is basically consistent with the change in the altitude spectrum from altitude a to altitude b and finally to altitude c, and corresponds one to one, as shown in the figure. Figure 4 Shown with Figure 3 The corresponding height spectrum fragment. Figure 5The figure shows the segments of the rotation speed spectrum and the height spectrum. The changes of the height spectrum and the rotation speed spectrum of segments 1 to 4 have the same rules, and the changes of the rotation speed spectrum and the height spectrum are repeated with a typical segment as a unit. The rotation speed spectrum and the height spectrum of segment 1 are similar to the spectrum type of segment 2, the same duration, and the height are similar. The rotation speed spectrum and the height spectrum of segment 3 are similar to the spectrum type, duration, and height of segment 4. The change rules of segments 1 to 2 and segments 3 to 4 in the rotation speed spectrum and height spectrum are basically the same, the difference is the duration and the maximum value of the height spectrum;

[0085] Step (3): Establishment of a three-dimensional K-Means model based on typical task segments and division criteria of typical sections.

[0086] The specific steps of step (3) are:

[0087] Based on the cluster analysis of military aircraft engine load spectra, a clustering algorithm for civil turboshaft engine load spectra was developed. Through the above analysis, it was found that the speed spectrum and altitude spectrum have the same change law and the altitude spectrum is simpler and easier to identify. For the clustering of the measured spectrum of civil aircraft turboshaft engines, the K-Means clustering algorithm is used:

[0088] In this example, the parameter height is used as the main parameter, and the segments of each of the n sections are classified through intelligent recognition, where: H is the maximum height of the segment, in m, and H1, H2, and H3 are the three types of segments divided according to the height of each section;

[0089]

[0090] Based on the duration of each segment, the three segments of H1, H2, and H3 of each section are further divided through intelligent recognition, where: Δt represents the duration of each segment, in seconds, and t1, t2, t3, and t4 represent the four types of segments divided according to the duration of each segment;

[0091]

[0092] Finally, all the segments of the intermediate mission sections of all profiles are divided into two categories of typical mission sections, namely, typical mission sections of execution missions and typical mission sections of cruise missions.

[0093] The characteristic of the typical task segment of the task execution is that the duration of the typical task segment of the task execution is no more than 1400s, such as Figure 6(a)-6(c) As shown, the speed spectra of the typical task segments of the execution task are the same, but the duration is different. The Δt1 of the typical task segment 1 is 300-500, the Δt2 of the typical task segment 2 is 500-800, and the Δt3 of the typical task segment 3 is 800-1400.

[0094] The duration of typical mission segments of cruise missions is greater than 1400s. Figure 7(a)-7(c) The typical segments of the typical mission segments of the cruise mission are shown, with H1 = 300-700 for the typical mission segment 4, H2 = 700-1000 for the typical mission segment 5, and H3 > 1000 for the typical mission segment 6;

[0095] According to the six typical mission segments, the n mission profiles are intelligently identified, the number of execution mission segments and cruise mission segments in each profile is recorded, and the typical mission segments with the largest number are extracted. The number of cruise segments-the number of execution segments-the number of typical mission segments with the largest number are used as the three-dimensional coordinate axis, and the characteristics of the n mission profiles of the turboshaft engine are expressed in a three-dimensional graph, as shown in the following figure: Figure 8 As shown in FIG. 1 , according to the distribution of data in the three-dimensional diagram, the speed spectrum is divided into three parts: typical section a, typical section b and typical section c, wherein: typical section a is a combination of typical task segment 1 and typical task segment 2; typical section b is a combination of 6 types of typical task segments mainly based on typical task segment 3 or typical task segment 4 or typical task segment 5 or typical task segment 6; typical section c is a combination of 6 types of typical task segments mainly based on typical task segment 1 or typical task segment 2, as shown in FIG. Figure 9(a)-Figure 9(c) Three typical profiles are shown.

[0096] Example 2

[0097] A method for analyzing the clustering of load spectra of civil turboshaft engines is presented. Taking n speed spectrum profiles of a certain type of helicopter performing n firefighting missions as an example, the analysis process of the clustering of load spectra of civil turboshaft engines is described in detail. The specific implementation steps are as follows:

[0098] Step (1) preprocessing the original speed spectra and height spectra of n sections;

[0099] Step (11) determining the division criteria of the take-off segment and the landing segment;

[0100] The division criteria for the take-off and landing phases are as follows: Figure 1 As shown;

[0101] Criteria for dividing the takeoff segment and the intermediate flight segment: the time point t1 when the engine speed changes from 0 to the maximum value for the first time and then decreases from the maximum value by a certain amplitude (5% of the maximum speed), and the time point t2 when the engine altitude changes from 0 to the maximum value for the first time. The smaller time value between t1 and t2 is taken as the separation point of the takeoff segment and the intermediate flight segment;

[0102] The criteria for dividing the intermediate flight segment and the landing segment are as follows: the time point t3 when the engine speed starts to decrease from the maximum value and never reaches the maximum value again during the descent, and the time point t4 when the engine altitude starts to decrease from a certain value and there is no obvious altitude increase during the process. The larger time value between t3 and t4 is taken as the separation point between the intermediate flight segment and the landing segment.

[0103] Step (12) divides the n speed profiles into take-off and landing sections according to the take-off and landing section division criterion of step (11);

[0104] Step (13) filters the n speed intermediate task segments,

[0105] The specific steps of step (13) are:

[0106] Step (131) determines the filtering thresholds of n intermediate task segments. The general filtering method is to remove noise from the signal. After collecting the peak and valley values ​​of the load parameters, filtering operations are performed in the following two aspects: (1) small range filtering; (2) small peak filtering, thereby removing small cycles that have little or no effect on fatigue damage;

[0107] First, the peak and valley values ​​of the intermediate task segment are extracted, and the calculation formula of the filter threshold Δ% is:

[0108] The general filtering method aims to remove noise from the signal. After collecting the peak and valley values ​​of the load parameters, filtering operations are performed through the following two aspects: (1) small range filtering; (2) small peak filtering, thereby removing small cycles that have little or no effect on fatigue damage.

[0109] First, the peak and valley values ​​of the intermediate task segment are extracted, and the filtering threshold is:

[0110] Δ%=G max *ω% (4)

[0111] Among them, Δ% is the filtering threshold; G max is the maximum amplitude of each cycle in the load history; ω is an empirical constant. For aircraft engines, the recommended value range is 8.0 to 12.5. For the n firefighting mission profiles in this patent, the empirical constant range is 4.5 to 8.5.

[0112] Step (3): Establishment of a three-dimensional K-Means model based on typical task segments and division criteria of typical sections.

[0113] K-Means is a particularly common clustering algorithm. Due to its high efficiency, it is often used in cluster analysis of large-scale data. At present, many algorithms are improved and extended based on this algorithm. Dividing objects into k clusters increases the differentiation between clusters and reduces the differentiation within clusters.

[0114] The process of K-Means algorithm is as follows: pre-divided into k clusters, k objects are randomly selected, and the initial location of each object represents the center value of the cluster. For all the remaining objects, the distance between the object and the center value of each cluster is determined, and the object is assigned to the nearest cluster; then the mean of all clusters is expanded and calculated again, and this step is repeated until the criterion function converges; the square error criterion is usually used, such as formula (5):

[0115]

[0116] Where: E represents the sum of squared errors of all objects in the database, i represents one of the k clusters, and i = 1, 2, ..., k; p represents the object in space, c i represents the i-th cluster, m i Represents cluster c i The objective function is to make the derived clusters as compact and independent as possible. The distance metric is the Euclidean distance, although other distance metrics are also applicable.

[0117] The specific steps of step (3) are: developing a clustering algorithm for the load spectrum of a civil turboshaft engine based on the clustering analysis of the load spectrum of a military aircraft engine. Through the above analysis, it is found that the speed spectrum and the altitude spectrum have the same variation law and the altitude spectrum type is simpler and easier to identify. For the clustering of the measured spectrum of a civil aircraft turboshaft engine, the following algorithm is used:

[0118] Based on the parameter height, the segments of each of the n profiles are classified through intelligent recognition, where H is the maximum height of the segment, in meters, and H1, H2, and H3 are the three types of segments divided according to the height of each profile. Based on the duration of each segment, the three types of H1, H2, and H3 of each profile are further divided through intelligent recognition, where Δt represents the duration of each segment, in seconds, and t1, t2, t3, and t4 represent the four types of segments divided according to the duration of each segment.

[0119] Finally, all the segments of the intermediate mission sections of all profiles are divided into two categories of typical mission sections, namely, typical mission sections of execution missions and typical mission sections of cruise missions.

[0120] The characteristic of a typical task segment is that the duration of the task segment is no more than 1400s, such as Figure 6(a)-6(c)Shown are typical fragments of typical task segments of the execution task. The execution segments have the same speed spectrum type but different durations. Δt1 of typical task segment 1 is 300-500, Δt2 of typical task segment 2 is 500-800, and Δt3 of typical task segment 3 is 800-1400.

[0121] like Figure 3 The typical mission segment is a fragment of a helicopter performing a firefighting mission, and is the smallest unit of a helicopter performing a firefighting mission. It cruises from destination B to destination C to perform mission a, then climbs to destination D, cruises to destination E, and descends to destination F to start mission b, and returns to destination C after completion (C(a)→D→E→F (b)→C). This process is repeated continuously with this mission segment as the standard until the helicopter completes the mission.

[0122] The duration of typical mission segments of cruise missions is greater than 1400s. Figure 7(a)-7(c) The figure shows typical segments of a typical mission segment of a cruise mission, with H1 = 300-700 for typical mission segment 4, H2 = 700-1000 for typical mission segment 5, and H3 > 1000 for typical mission segment 6. The cruise mission segment is a task performed by a helicopter to check the firefighting situation when performing a firefighting mission.

[0123] According to the six typical mission segments, the n mission profiles are intelligently identified, the number of execution mission segments and cruise mission segments in each profile is recorded, and the typical mission segments with the largest number are extracted. The number of cruise segments-the number of execution segments-the number of typical mission segments with the largest number of extracted segments 1 or 2 or 3 or 4 or 5 or 6 are used as the three-dimensional coordinate axes, and the characteristics of the n mission profiles of the turboshaft engine are expressed by the K-Means three-dimensional graph of the speed spectrum, as shown in the figure: Figure 8 As shown. First, according to the distribution of data in the three-dimensional graph, the speed spectrum is divided into three parts. Typical profile a is a combination of typical task segment 1 and typical task segment 2; typical profile b is a combination of 6 types of typical task segments mainly based on typical task segment 3 or typical task segment 4 or typical task segment 5 or typical task segment 6; typical profile c is a combination of 6 types of typical task segments mainly based on typical task segment 1 or typical task segment 2. Figure 9(a)-9(c) As shown, there are three typical profiles.

[0124] Based on the typical characteristics of helicopters performing firefighting missions, it can be found that most mission profiles are composed of typical mission segments 1 and 2, namely typical profile a. Typical profile c is mostly composed of typical mission segment 6 and typical mission segment 1 or typical mission segment 2. Typical profile b consists of two types of profiles, namely, the cruise profile and the cruise profile with typical mission segment 3. The typical profile b is further subdivided, as shown in Figure 1. Fig.10 As shown in Figure 2, the profiles with only cruise missions are divided into a category named d, such as Fig.11 As shown, the typical section d is obtained by re-dividing the typical section b.

[0125] Typical profiles a and d are typical firefighting profiles, where the helicopter is in a state of continuously fetching water and then spraying it on the fire scene, and in a state of continuously cruising at the fire scene to check the fire situation. The task of typical profile b is to continuously cruise to check the fire situation, and spray water to put out the fire in a serious situation. The task of typical profile c is to continuously perform firefighting tasks, i.e., fetch water and spray water, until the fire situation in area a is reduced, then continuously cruise to check the fire situation, and go to area b to put out the fire.

[0126] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A three-dimensional K-Means turboshaft engine load spectrum clustering analysis method, characterized in that: The following steps are involved: Step (1), preprocessing the original speed spectra and height spectra of n sections; Step (2), characteristic analysis of civil turboshaft engine parameter spectrum; Step (3), establishment of a three-dimensional K-Means model based on typical task segments and division criteria of typical sections; The specific steps of step (1) are: Step (11) determines the division criteria of the take-off segment and the landing segment. Step (12) divides the n speed profiles into take-off and landing sections according to the take-off and landing section division criteria of step (11). Step (13) filtering the n speed intermediate task segments; The specific steps of step (2) are: The flight action of a helicopter can be decomposed into the basic flight actions of level flight, inclined climb, vertical climb, hovering and maneuvering flight. The flight mission of a helicopter is a combination of the basic flight actions. According to the changing characteristics of the helicopter flight mission, it can be divided into the take-off stage, the intermediate mission stage and the landing stage. Among them, the characteristics of the take-off stage and the landing stage are fixed, and they are deterministic mission stages; the intermediate mission stage is directly related to the flight mission performed by the helicopter. When the helicopter performs different flight missions, the take-off stage and the landing stage are consistent, and the intermediate mission stage will be significantly different, that is, the intermediate mission stage of the mission profile reflects the mission characteristics of the helicopter. The changes in the intermediate mission stage spectrum of the helicopter speed spectrum and the altitude spectrum correspond one-to-one to the actions performed by the helicopter, and the changes in the speed spectrum and the altitude spectrum also correspond one-to-one; The specific steps of step (3) are: Based on the parameter height, the segments of each of the n profiles are classified through intelligent recognition, where: H is the maximum height of the segment, in meters, and H1, H2, and H3 are the three types of segments divided according to the height of each profile; Based on the duration of each segment, the three segments of H1, H2 and H3 of each section are further divided through intelligent recognition, where: △t represents the duration of each segment, the unit is s, t1, t2, t3, t4 represent the four types of segments divided according to the duration of each segment; Finally, all the segments of the intermediate mission sections of all profiles are divided into two categories of typical mission sections, namely, typical mission sections of execution missions and typical mission sections of cruise missions.

2. The method for clustering analysis of turboshaft engine load spectrum based on three-dimensional K-Means according to claim 1, characterized in that: The division criteria of the take-off segment and the landing segment in step (11) are: Criteria for dividing the takeoff segment and the intermediate flight segment: the time point t1 when the engine speed changes from 0 to the maximum value for the first time and then drops from the maximum value to 5% of the maximum speed, and the time point t2 when the engine altitude changes from 0 to the maximum value for the first time. The smaller time value of t1 and t2 is taken as the separation point of the takeoff segment and the intermediate flight segment; The criteria for dividing the intermediate flight segment and the landing segment are as follows: the time point t3 when the engine speed starts to drop from the maximum value and never reaches the maximum value again during the descent; the time point t4 when the engine altitude starts to drop from a certain value and there is no obvious increase in altitude during the process; the larger time value between t3 and t4 is taken as the separation point between the intermediate flight segment and the landing segment.

3. The method for clustering analysis of turboshaft engine load spectrum based on three-dimensional K-Means according to claim 1, characterized in that: The specific steps of step (13) are: Step (131) determines the filtering thresholds of n intermediate task segments. First, the peak and valley values ​​of the intermediate task segments are extracted. The calculation formula of the filtering threshold △% is: △%=G max *ω% (1) Where: G max is the maximum amplitude of each cycle in the load history; ω is an empirical constant, and the value range of the empirical constant ω is 8.0~12.5; Step (132) calculates the filtering thresholds of n speed intermediate task segments respectively based on step (131); Step (133) writes the filtering thresholds of the filtering and n speed intermediate task segments into the rain flow count, and filters the n speed intermediate task segments.

4. The method for clustering analysis of turboshaft engine load spectrum based on three-dimensional K-Means according to claim 1, characterized in that: The characteristics of the typical task segment of the task are that the duration of the typical task segment is no more than 1400s, the speed spectrum of the typical task segment is the same, and the duration is different, the △t1 of the typical task segment 1 is 300-500, the △t2 of the typical task segment 2 is 500-800, and the △t3 of the typical task segment 3 is 800-1400; The duration of typical mission segments of cruise missions is greater than 1400s, H1 of typical mission segment 4 is 300-700s, H2 of typical mission segment 5 is 700-1000s, and H3 of typical mission segment 6 is >1000s.

5. The method for clustering analysis of turboshaft engine load spectrum based on three-dimensional K-Means according to claim 4 is characterized in that: According to the 6 types of typical mission segments, n mission profiles are intelligently identified, the number of execution mission segments and cruise mission segments in each profile is recorded respectively, and the typical mission segments with the largest number are extracted. The number of cruise segments-the number of execution segments-the number of typical mission segments with the largest number extracted are used as the three-dimensional coordinate axis, and the characteristics of the n mission profiles of the turboshaft engine are represented by a three-dimensional graph. According to the distribution of data in the three-dimensional graph, the speed spectrum is divided into three parts: typical profile a, typical profile b and typical profile c, among which: typical profile a is a combination of typical mission segment 1 and typical mission segment 2; typical profile b is a combination of 6 types of typical mission segments mainly composed of typical mission segment 3 or typical mission segment 4 or typical mission segment 5 or typical mission segment 6; typical profile c is a combination of 6 types of typical mission segments mainly composed of typical mission segment 1 or typical mission segment 2.

Images