Aero-engine double-characteristic program spectrum compilation method

Abstract

The invention discloses an aero-engine double-characteristic program spectrum compilation method, and relates to the technical field of aero-engine structural part load spectrum compilation, and the aero-engine double-characteristic program spectrum compilation method comprises the steps: obtaining and preprocessing an aero-engine load time history, and obtaining an aero-engine load cycle; performing clustering analysis to obtain a typical characteristic cycle, and constructing a typical characteristic load program spectrum block; establishing a multi-objective optimization model, solving by using a particle swarm to obtain an optimal load grade frequency, and determining the number of program spectrum block loads of each grade of typical characteristic circulation; and determining the sequence of program spectrum blocks according to the position of a typical characteristic load hysteresis loop. According to the method, blindness of a traditional equal-interval or fixed-proportion division method is effectively overcome, scientificity and representativeness of load grading are remarkably improved, program spectrum compiling work is remarkably simplified, and standardization, repeatability and engineering implementation efficiency of the program spectrum compiling method are improved.

Description

Technical Field

[0001] This invention relates to the field of load spectrum compilation technology for aero-engine structural components, and in particular to a method for compiling dual-feature program spectra for aero-engines. Background Technology

[0002] With the continuous advancement of aero-engine technology, the fatigue performance of key engine structural components (such as compressor disks, turbine blades, and turbine disks) has become a critical factor affecting overall engine safety and the accuracy of life prediction. To obtain reliable fatigue life prediction results, it is essential to realistically reproduce the complex flight load environment in ground tests. Programmed load spectra not only structurally simplify the original continuous load history but also effectively retain the main load characteristics of the measured load spectrum, thus they are widely used for structural life assessment and accelerated fatigue testing.

[0003] The core of program spectrum compilation is to approximate the actual flight history as closely as possible through methods such as damage equivalence. Huang Weiqing reconstructed various cycles into discrete program spectra according to the principles of time and damage equivalence, achieving equivalent compilation in terms of cumulative damage and working condition distribution. However, his compilation process ignored the load sequence effect, making it difficult to truly reflect the loading situation of the measured load. Moreover, his load classification was too coarse, dividing the amplitude into three categories and converting Class III cycles to Class II, which may dilute the damage characteristics of typical high-amplitude cycles and make it impossible to correspond one-to-one with the typical characteristic cycles of actual working conditions. Xue Hai (a non-equal interval stress spectrum compilation method based on cluster analysis and support vector machine) considered the attribute characteristics of each stress cycle, used fatigue damage as the criterion, and used cluster analysis to classify all stress cycles, proposing a non-equal interval adaptive ratio coefficient determination method, which can reflect the actual load characteristics to a certain extent. However, this method uses linear cumulative damage as the analysis factor, lacks consideration of the load order, and ignores the inherent characteristic attributes of the load.

[0004] Currently, most program spectra only consider statistical characteristics, such as mean, amplitude, and frequency, while the load order is mostly randomized, ignoring the impact of load sequences on fatigue damage. This results in deficiencies in the reliability, standardization, and testing efficiency of aero-engine fatigue test spectra. Furthermore, aero-engine program spectra typically employ load level division methods with equal intervals or fixed proportions, failing to adaptively extract non-stationary and operationally relevant characteristics of aero-engines under different flight conditions, making it difficult to obtain typical cycles that truly represent actual service loads. The allocation of cycle counts lacks multi-objective evaluation; traditional area methods or empirical formulas only allocate cycle counts for a single dimension (such as amplitude), lacking multi-objective optimization evaluation indicators that simultaneously minimize the errors in mean and amplitude distributions. This leads to insufficient matching between the program spectra and the measured load spectra in two-dimensional distribution. Summary of the Invention

[0005] In view of the problems existing in the current method for compiling dual-feature program spectra of aero-engines, this invention is proposed. Therefore, the problem to be solved by this invention is how to provide a method for compiling dual-feature program spectra of aero-engines.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for compiling a dual-feature program spectrum of an aero-engine, which includes acquiring the aero-engine load time history and performing preprocessing to obtain the aero-engine load cycle.

[0008] Cluster analysis was performed on the load cycles of aero-engines to obtain typical characteristic cycles, and typical characteristic load program blocks were constructed.

[0009] A multi-objective optimization model was established and particle swarm optimization was used to obtain the optimal load level frequency, and the number of program spectral block loads for typical characteristic cycles of each level was determined.

[0010] The sequence of program blocks is determined based on the location of typical characteristic load hysteresis loops, and a dual-characteristic program spectrum for aero-engines is compiled.

[0011] As a preferred embodiment of the dual-feature program spectrum compilation method for aero-engines described in this invention, the preprocessing includes extracting and filtering peak and valley values ​​from the aero-engine load time history, counting rainflows, extracting the average amplitude information of load cycles, dividing the extracted load cycles into primary and secondary cycles and front and rear cycles, using the maximum load time point in the flight profile as the dividing boundary, dividing the aero-engine load time history into front and rear cycles, and performing cluster analysis on the front and rear cycles to determine typical feature cycles.

[0012] As a preferred embodiment of the dual-feature program spectrum compilation method for aero-engines described in this invention, the step of obtaining typical feature cycles through cluster analysis of aero-engine load cycles includes:

[0013] Cluster analysis of aero-engine load cycles was performed using the CFSFDP clustering method. For a given two-dimensional sample set, the local density and density distance of each sample point were calculated before identifying the cluster centers.

[0014] Calculate the Euclidean distance for all sample points;

[0015] Calculate the local density and density distance based on the Euclidean distance of the sample points;

[0016] Decision variables are constructed by combining local density and density distance. A decision graph is drawn based on the decision variables in descending order. The cluster centers in the decision graph are selected as typical feature cycles.

[0017] In a preferred embodiment of the dual-feature program spectrum compilation method for aero-engines described in this invention, the Euclidean distance is calculated using the following formula:

[0018]

[0019] Where: d(x) i ,x i Let x be any two points in the dataset. i and x j Euclidean distance, (m) i ,a i ) represents the sample point x i The average amplitude of engine load obtained from the rainflow counting method is statistically analyzed (m). j ,a j ) represents the sample point x j The average amplitude of engine load obtained from the rainflow counting method is statistically analyzed.

[0020] The formula for calculating the local density is:

[0021]

[0022] Where: ρ i For sample point x i Local density, KNN(X) i ) represents x i The set of k nearest neighbor data points of the center, here k=10;

[0023] All local densities are higher than the sample point x i Among the points, find the sample point with the smallest distance, which is the density distance, denoted as:

[0024]

[0025] Where: δ i Let j represent the density distance, and j denote the index of all data points.

[0026] The decision variables are represented as follows:

[0027] γ i =ρ i ·δ i

[0028] Wherein: γ i These are decision variables.

[0029] As a preferred embodiment of the dual-feature program spectrum compilation method for aero-engines described in this invention, the establishment of the multi-objective optimization model includes:

[0030] With the optimization objectives of minimizing the relative absolute error of the mean and the relative absolute error of the amplitude, and with the constraints of keeping the number of load cycles unchanged before and after spectrum compilation and ensuring that the frequency of each load level in the program spectrum is not less than the number of main cycles, the following multi-objective optimization model is established, expressed as:

[0031]

[0032] in: and These are the cumulative frequency curves of the one-dimensional marginal distribution of the mean of the measured load spectrum and the programmed spectrum, respectively. and These are the cumulative frequency curves of the one-dimensional marginal distribution of the measured load spectrum and the programmed spectrum amplitude, respectively. X represents the frequency of each load level, RIAE_m(X) is the mean relative comprehensive absolute error, RIAE_a(X) is the amplitude relative comprehensive absolute error, N is the number of main loops, and n... spec represents the total number of cycles extracted from the original load spectrum, i represents the i-th load level, represents the total number of load levels, and Δ represents the step size in the discretization process.

[0033] As a preferred embodiment of the dual-feature program spectrum compilation method for aero-engines described in this invention, the step of obtaining the optimal load level frequency using particle swarm optimization includes:

[0034] The optimization objectives of minimizing the relative absolute error of the mean and the relative absolute error of the magnitude are transformed into a single-objective optimization problem through weighted summation, expressed as:

[0035] min F(x)=ω1f1(x)+ω2f2(x)

[0036] Where: F(x) is the weighted sum objective function, ω1, ω2 are weight coefficients, satisfying ω1+ω2=1, f1(x) is the magnitude objective function, and f2(x) is the mean objective function;

[0037] After the weighted sum objective function is determined, particle swarm optimization is used to solve for the global optimal solution, and the number of program spectral block loads for typical characteristic cycles at each level is determined.

[0038] As a preferred embodiment of the dual-feature program spectrum compilation method for aero-engines described in this invention, the step of determining the order of program spectrum blocks based on the location of typical characteristic load hysteresis loops includes:

[0039] Based on the hysteresis loop theory, the parent-child relationship, load level and load direction of fatigue load are extracted by rainflow counting method to obtain the longitudinal attributes of fatigue load. Each load level contains a different number of load cycles.

[0040] The load spectrum block sorting strategy is determined based on the back-interpolation rule. Each transverse load level is treated as a single load, and back-interpolated into the reference cycle level by level according to the longitudinal attributes from low to high. The sorting strategy is as follows:

[0041] The reference cycle is considered as a first-order longitudinal load cycle and serves as the base load for all typical cycle back-interpolation.

[0042] All secondary longitudinal load cycles are sub-cycles of the reference load cycle. Their relative positions in the primary longitudinal load cycle are determined by the load magnitude, and the shape of the inserted load spectrum is determined by the load cycle direction. The relative position refers to the order of multiple sub-cycles in the parent cycle. In the load increasing half-cycle, the maximum value of the typical characteristic cycle is used as the criterion, from small to large. In the load decreasing half-cycle, the minimum value of the typical characteristic cycle is used as the criterion, from large to small.

[0043] Determine whether a secondary parent cycle exists for a third-level typical characteristic load based on the load size. Assume the size of a certain third-level typical load is (σ'). max ,σ' min If σ exists min ≥σ' min And σ max ≤σ' max If the second-level typical characteristic loads meet the conditions, then the second-level typical loads are regarded as the nominal parent cycle of the third-level typical loads currently being analyzed. If a third-level typical load has multiple nominal parent cycles, then the one with the smallest Euclidean distance from the real parent cycle is selected as the final selection.

[0044] For Level 3 typical characteristic loads that do not have a nominal parent cycle, the longitudinal load level is upgraded to Level 2 and then back-inserted; for Level 3 typical loads that have a nominal parent cycle, they are inserted into Level 2 typical characteristic loads according to the cycle direction.

[0045] Based on the determined typical feature cycle frequency, the corresponding number of times the typical feature cycle is repeated under different levels constitutes a program block spectrum. In the load increase half-cycle, the lower-level typical loads are all inserted back to 1 / 2 of the frequency of the previous level typical load. In the load decrease half-cycle, the lower-level typical loads are inserted back to 1 / 2 of the frequency of the previous level typical load. If there are multiple loads of the same level, the average value of the spectrum block is used as the judgment standard. The one with the smallest average value of the spectrum block is inserted back to the end of the previous level typical load, and the rest are inserted back to 1 / 2 of the frequency of the previous level typical load.

[0046] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of a method for compiling a dual-feature program spectrum for an aero-engine.

[0047] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements the steps of a method for compiling a dual-feature program spectrum for an aero-engine.

[0048] The beneficial effects of this invention are as follows: This invention introduces the CFSFDP density peak clustering algorithm, which can adaptively and automatically extract the most representative typical feature cycles from the preceding / following sub-cycles without pre-setting the number of load levels. This effectively overcomes the blindness of traditional equal-interval or fixed-ratio division methods, significantly improving the scientific rigor and representativeness of load classification. Regarding load frequency determination, a multi-objective optimization model is constructed with the relative integral absolute error of the load average value and amplitude as dual objectives. Under the premise of maintaining a constant total number of cycles and satisfying the minimum frequency constraint for each load level, the optimal cycle number allocation scheme is obtained, thereby achieving a high degree of approximation of the original spectrum in terms of statistical distribution of the program spectrum.

[0049] This invention fully considers the sequence effect of fatigue loads. By identifying the parent-child relationship and loading direction characteristics between sub-cycles, and based on the loading direction + relative position principle, each typical characteristic cycle is inserted back into the main cycle in hierarchical order, effectively reproducing the sequence dependency behavior in non-stationary operating loads.

[0050] This invention constructs an integrated compilation process that organically integrates load preprocessing, typical feature cyclic identification, multi-target frequency optimization, and hierarchical sequence back-insertion, significantly simplifying the program spectrum compilation work and improving the standardization, repeatability, and engineering implementation efficiency of the method. Attached Figure Description

[0051] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a flowchart of the method.

[0053] Figure 2 This is a schematic diagram of the preprocessed load spectrum.

[0054] Figure 3 For the measured flight data rainflow matrix histogram'

[0055] Figure 4 This is a schematic diagram of the sub-cyclic load division.

[0056] Figure 5 This is a decision graph based on CFSFDP clustering.

[0057] Figure 6 This represents the forward cyclic load based on CFSFDP.

[0058] Figure 7 This is a schematic diagram of a typical feature loop based on CFSFDP.

[0059] Figure 8 This is a schematic diagram of the characteristic program spectrum load level classification.

[0060] Figure 9 This is a diagram showing the longitudinal load level.

[0061] Figure 10 A schematic diagram for determining the load frequency of the previous cycle.

[0062] Figure 11 A schematic diagram for determining longitudinal loads.

[0063] Figure 12 This is a schematic diagram of the cyclic insertion rule.

[0064] Figure 13 This is a schematic diagram of two-dimensional information on fatigue load.

[0065] Figure 14 A schematic diagram for constructing an equivalent program block spectrum.

[0066] Figure 15 This is a schematic diagram of the equivalent program block spectrum. Detailed Implementation

[0067] To make the above-mentioned objects, features, and advantages of the present invention more readily understood, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0068] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0069] Secondly, the term "one embodiment" or "example" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the invention. An embodiment appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment that selectively excludes other embodiments.

[0070] Example 1

[0071] Reference Figure 1 This is the first embodiment of the present invention, which provides a method for compiling a dual-feature program spectrum for an aero-engine, including:

[0072] S1: Obtain the time history of the aero-engine load and perform preprocessing to obtain the aero-engine load cycle;

[0073] Specifically, the measured load spectrum preprocessing involves preprocessing the aero-engine load time history by extracting peak and valley values, filtering, and counting rainflows, and extracting the average amplitude information of the load cycles.

[0074] Primary and secondary loop partitioning: For the extracted loops, assuming there are N original flight profiles, the 0-max-0 loop of each flight profile is used as the primary loop, and the remaining loops are considered secondary loops. The average amplitude of the primary loop is determined as follows: There are N in total.

[0075] Pre- and post-cycle division: Using the time point of maximum load in the flight profile as the dividing line, the load history is divided into pre- and post-cycles. Cluster analysis is performed on the pre- and post-cycles respectively to determine the typical characteristic cycles.

[0076] S2: Cluster analysis is performed on the load cycles of aero-engines to obtain typical characteristic cycles, and typical characteristic load program blocks are constructed.

[0077] Specifically, S2.1: Clustering Algorithm Selection: The CFSFDP clustering method used is a fast search clustering method for spatial data based on density peaks. The clustering process is insensitive to initial values ​​and can find non-spherical clusters. For a given two-dimensional sample set X = {x1, x2, ..., x...} n |x i =(m i ,a i )}, where (m i ,a i The mean amplitude of the engine load obtained from the rainflow counting method is the statistical cycle result. Before identifying the cluster centers, two types of indices are calculated for each sample point: local density ρ. i and density distance δ i .

[0078] S2.2: Calculate the clustering matrix, i.e., calculate the Euclidean distance for all sample points. The calculation formula is:

[0079]

[0080] Where: d(x) i ,x i Let x be any two points in the dataset. i and x j Euclidean distance, (m)i ,a i Let x be a point. i The average amplitude of engine load obtained from the rainflow counting method is statistically analyzed (m). j ,a j Let x be a point. j The average amplitude of engine load obtained from the rainflow counting method is statistically analyzed.

[0081] S2.3: Calculate the local density for each sample point x. i The local density is defined as the inverse of the k-nearest neighbor distance, KNN(x) i The calculation formula is:

[0082]

[0083] Where: ρ i For local density, KNN(X) i ) represents x i The set of k nearest neighbor data points of the center, here k=10;

[0084] S2.4: Calculate the density distance for each sample point x. i Among all points with higher density, find the one with the smallest distance; this smallest distance is the density distance, denoted as:

[0085]

[0086] Where: δ i Let j represent the density distance, and j denote the index of all data points.

[0087] S2.5: Construct decision variables and a decision graph. Decision variables comprehensively measure the frequency (density) of cycle occurrence and their independence (density distance) in the load spectrum distribution, and are expressed as follows:

[0088] γ i =ρ i ·δ i

[0089] Wherein: γ i For decision variables;

[0090] The decision variables calculated are arranged in descending order to form a decision graph. Based on the CFSFDP principle, the cluster centers in the decision variable decision graph are selected as typical feature cycles. The typical feature cycles represent the load levels that are most representative and different in terms of amplitude-mean characteristics in the original fatigue load.

[0091] S2.6: Construct the typical characteristic load program spectrum block. The typical characteristic load in the first loop has t levels, denoted as F1…F1. t The subsequent loop has p stages, denoted as B1…B1.p Based on the load characteristics of aero engines, F t and B p The number of load cycles included should be greater than the total number of flight profiles N.

[0092] S3: Establish a multi-objective optimization model and use particle swarm optimization to obtain the optimal load level frequency, and determine the number of program spectral block loads for typical characteristic cycles of each level.

[0093] Specifically, S3.1: Establish a multi-objective optimization model. The higher the similarity between the cumulative frequency curves of the programmed spectrum and the measured load spectrum, the more reasonable the frequency determination. The relative comprehensive absolute error (RIAE) is used to quantitatively characterize the closeness of the load spectra before and after compilation. The smaller the RIAE, the closer the two lines are, and the better the compilation effect. A reasonable... The value should be chosen to minimize RIAE; therefore, determining the load level frequency is transformed into using RIAE as the optimization objective. This is an optimization problem involving design variables. Furthermore, the load cycle is a two-dimensional variable (m...). i ,a i Since both the measured load spectrum and the programmed spectrum have two marginal distributions—mean and amplitude—the optimization objective is transformed into a multi-objective optimization problem that simultaneously minimizes RIAE_m(X) and RIAE_a(X). With the constraints of keeping the number of load cycles unchanged before and after spectrum compilation and ensuring that the frequency of each load level in the programmed spectrum is not less than the number of main cycles, the following optimization model is established:

[0094]

[0095] in: and These are the cumulative frequency curves of the one-dimensional marginal distribution of the mean of the measured load spectrum and the programmed spectrum, respectively. and These are the cumulative frequency curves of the one-dimensional marginal distribution of the measured load spectrum and the programmed spectrum amplitudes, respectively, and the optimal solution. This refers to the frequency of each load level, RIAE_m(X) is the mean relative comprehensive absolute error, RIAE_a(X) is the amplitude relative comprehensive absolute error, N is the number of main cycles, and n spec represents the total number of cycles extracted from the original load spectrum, i represents the i-th load level, represents the total number of load levels, and Δ represents the step size in the discretization process.

[0096] S3.2: Solving the multi-objective optimization model. Although the model originally has two optimization objectives, directly handling a bi-objective optimization problem is complex and prone to low algorithm efficiency. Therefore, the bi-objective problem is transformed into a single-objective optimization problem through weighted summation, expressed as:

[0097] min F(x)=ω1f1(x)+ω2f2(x)

[0098] Where: F(x) is the weighted sum objective function, ω1, ω2 are weight coefficients, satisfying ω1+ω2=1, f1(x) is the magnitude objective function, and f2(x) is the mean objective function.

[0099] After the weighted sum objective function F(x) is determined, the global optimal solution is obtained by using particle swarm optimization, thereby determining the number of program spectral block loads for each level of typical characteristic cycle.

[0100] S4: Determine the sequence of program spectrum blocks based on the location of typical characteristic load hysteresis loops, and compile dual-characteristic program spectra for aero-engines.

[0101] Specifically, S4.1: Determining longitudinal load information based on hysteresis loop theory: The longitudinal attributes of fatigue loads are obtained by extracting the parent-child relationship, load level, and load direction of fatigue loads using the rainflow counting method based on hysteresis loops. Each load level contains a different number of load cycles, where the longitudinal level, parent-child relationship, and cycle direction are inherent longitudinal attributes of each load cycle and are unique.

[0102] S4.2: Determining the load spectral block sorting strategy based on the back-interpolation rule:

[0103] The core idea of ​​spectral block sorting is to replace the actual load time history with a load time history that is arranged in the true order of all typical characteristic loads, since both cause the same level of damage. In this way, the program load spectrum basically maintains the original load order characteristics.

[0104] The hysteresis loop order block sorting strategy treats each transverse load level as a single load, and then backinserts it into the baseline loop level by level from low to high according to its longitudinal properties. The specific sorting strategy is as follows:

[0105] (a) The reference cycle (0-max-0) is regarded as a first-order longitudinal load cycle and is used as the base load for all typical cycle back-interpolation;

[0106] (b) All secondary longitudinal load cycles must belong to the sub-cycles of the reference load cycle. Then, the relative position in the primary longitudinal load cycle is determined according to the load size, and the shape of the load spectrum after insertion is determined according to the cycle direction.

[0107] Here, relative position refers to the order of multiple sub-loops within the parent loop, and in the load-increasing half-loop, it is based on the maximum value (σ) of the typical characteristic loop. max =σ m +σ a The criterion is to judge from small to large values; while in the load reduction half-cycle, the minimum value of the typical characteristic cycle (σ) is used as the criterion. max =σ m -σ aUsing the largest value as the criterion, proceed from largest to smallest. The rules for back-insertion in different cycle directions are as follows: Figure 12 As shown;

[0108] (c) Determine whether a secondary parent cycle exists for a typical tertiary load based on the load size: Assume that the size of a typical tertiary load is (σ' max ,σ' min If σ exists min ≥σ' min And σ max ≤σ' max For the second-level typical characteristic loads, those that meet the conditions are considered as the nominal parent cycle of the currently analyzed third-level typical loads. If a third-level typical load has multiple nominal parent cycles, the one with the smallest Euclidean distance from the real parent cycle is selected as the final choice.

[0109] (d) For a Level 3 typical characteristic load that does not have a nominal parent cycle, its longitudinal load level is upgraded to Level 2, and then it is back-inserted according to the principle in (b). The upgraded Level 3 typical characteristic load may change its original cycle direction in order to meet the insertion rules of the Level 2 load cycle;

[0110] (e) For a typical three-level load with a nominal parent cycle, the load is determined according to its cycle direction. Figure 12 The back-insertion rule is applied to the secondary typical characteristic loads. If multiple tertiary typical loads belong to the same nominal parent load, the order of the loads is determined according to the rule in (b).

[0111] (f) Analyze typical loads of level four, level five, etc., and process them in the same way as the above methods.

[0112] S4.3: Single load history spectrum blockization. Based on the determined typical feature cycle frequency, the typical feature cycle repeats for different levels a corresponding number of times to form a program block spectrum. In the load increasing half-cycle, the lower-level typical loads are all interpolated back to 1 / 2 of the frequency of the previous level typical load; in the load decreasing half-cycle, the lower-level typical loads are interpolated back to 1 / 2 of the frequency of the previous level typical load. If there are multiple loads of the same level, the average value of the spectrum block is used as the judgment standard. The one with the smallest average value is interpolated back to the end of the previous level typical load, and the rest are interpolated back to 1 / 2 of the frequency of the previous level typical load.

[0113] Example 2

[0114] Reference Figures 2-15 This embodiment comprehensively considers the structural and fatigue characteristics of the load spectrum, and mainly includes three parts: load level classification, load frequency determination, and load spectrum block sorting.

[0115] Step 1: Preprocessing of the measured load spectrum;

[0116] The original load spectrum, compiled using 37 measured profiles of the compressor disk of the X-type third-generation aero-engine as the program spectrum, is normalized, peak-valley value detected, and small loads removed. The random load-time history is as follows: Figure 2 As shown.

[0117] Step 2: Divide the main loop and the secondary loop;

[0118] Figure 2 The load-time history shown yielded 1291 load cycles through rainflow counting, including 37 main cycles and 1254 secondary cycles. The rainflow matrix histogram is shown below. Figure 3 As shown. The measured flight profile t... max As a dividing line, a total of 492 pre-loops and 762 post-loops were defined, such as... Figure 4 As shown.

[0119] Step 3: Classify load levels based on clustering;

[0120] Taking 492 previous cycles as an example, cluster analysis is performed, with each loading cycle representing a two-dimensional sample point. The Euclidean distance matrix D of the previous cycles is calculated as follows:

[0121]

[0122] Calculate the local density vector of each sample cycle when k=10. and density distance vector for:

[0123]

[0124] Depend on and The decision diagram is constructed, and then the decision variables are further calculated.

[0125]

[0126] For decision variables arranged in descending order, γ gradually flattens out below the threshold γ′ = 164. Based on the CFSFDP principle, the four cluster centers preceding γ′ are selected as the typical feature loops of the preceding loop, and... Figure 6 The values ​​are marked in red. The analysis process is the same, yielding typical load cycles for subsequent cycles. Cluster analysis reveals the number and magnitude of typical characteristic loads, as shown below. Figure 7 As shown, there are a total of eight typical load cycles in this cycle.

[0127] Step 4: Determine the frequency of load levels through multi-objective optimization;

[0128] Taking the previous loop as an example, the process of determining the frequency of typical characteristic loads based on the multi-objective optimization method proposed earlier is as follows: Figure 10As shown. Figure 10 (a) is the typical feature load information determined by cluster analysis; Figure 10 (b) is the optimal solution when RIAE_a and RIAE_m simultaneously reach their minimum values. At this point, the program spectrum, whether in terms of amplitude marginal distribution or mean marginal distribution, is closest to the shape of the original load spectrum. The black solid line represents the discrete marginal distribution of the measured load spectrum at amplitude and mean values, respectively. The blue and red dashed lines represent the discretized program load spectrum. Each horizontal segment represents the frequency of the load at that level. The blue and red shaded areas represent RIAE_a and RIAE_m, and the values ​​in parentheses are the specific frequency optimal solutions for each typical characteristic load. Figure 10 (c) is a load spectrum block composed of typical characteristic loads of the pre-cycle and the cycle frequency.

[0129] Step 5: Determine the longitudinal load level;

[0130] Taking the typical characteristic load F2 of the previous cycle as an example, F2 is in the 0-max process of the reference cycle, but it is not directly included in the reference cycle (0-max-0). Instead, it is included in the 'load reduction' process of a positive load in the 0-max process of the reference cycle. The longitudinal load history is as follows: Figure 9 As shown in (b). Therefore, its parent cycle is a second-order longitudinal load, F2 is determined to be a third-order longitudinal load, and its cycle direction is determined to be a negative load based on the rainflow counting principle, denoted by -; the longitudinal load levels of the eight typical characteristic loads obtained from cluster analysis are as follows: Figure 9 As shown, where Figure 9 (a) to (d) represent the longitudinal load level information for the previous cycle. Figure 9 (e)~(h) represent the longitudinal load level information for the subsequent cycle.

[0131] Load spectrum blocks and related lateral and longitudinal load information of 37 measured flight profiles of aero-engines are as follows: Figure 13 As shown, the lateral and longitudinal load information of the main cycle and eight typical characteristic cycles is summarized in the table below:

[0132] Table 1. Lateral and longitudinal loads of the main cycle and eight typical characteristic cycles.

[0133]

[0134] Step 6: Generate program block spectrum.

[0135] Taking into account both lateral and longitudinal load information of typical characteristic loads, a program spectrum is generated in the order of previous loops followed by subsequent loops. In the previous loop, the longitudinal load levels of F1 and F3 are both level two, and since there is no need to consider load escalation or de-escalation, they should be prioritized, with the load with the smaller maximum value taking precedence. The maximum load value of F1 is... U3 maximum load satisfy Therefore, in the program spectrum, F1 is generated first, followed by F3. In the remaining two pre-loops, the third-level load U2 is generated first, followed by the fourth-level load F4. In the post-loop, there are simultaneously second-level (B2), third-level (B4), and fourth-level (B1, B3) longitudinal load levels. In the post-loop of the program spectrum, B2 is generated first, followed by B4, and finally B1 and B3. Among them, the third-level load B4 satisfies the inclusion condition of being the nominal parent loop of both B1 and B3. Since the average value of B3 load is less than that of B1, according to the program spectrum block sorting strategy, B1 is inserted back at the end of the B4 block, and B3 is inserted back in the middle of the B4 block.

[0136] Taking F4 as an example, because the third-level load F2 does not meet the inclusion condition for the fourth-level load F4, it cannot be used as the nominal parent cycle of F4. Therefore, F4 is upgraded to a third-level longitudinal load. Further analysis shows that the second-level load F3 meets the inclusion condition for being the nominal parent cycle of F4. Therefore, F4 is used as a third-level longitudinal load, maintaining its original cycle direction. Figure 14 (b) Insert back into the drop load segment of F3 to generate as follows Figure 14 The load sequence is shown in (k). Place F4 in... Figure 14 (i) The second spectral block is looped 160 times to generate a program spectrum containing 3 load levels and 4 spectral blocks, such as... Figure 14 As shown in (l), and the spectral blocks also satisfy the same condition. Figure 12 The insertion rule shown in (b) is as follows.

[0137] The sequential generation process of the next cyclic spectral block is as follows: Figure 14 As shown in (a) to (x). Finally, in Figure 14 (x) Add all reference cycles at the point of maximum load to form the final equivalent program block spectrum, such as Figure 15 As shown.

[0138] This embodiment also provides a computer device applicable to a dual-feature program spectrum compilation method for an aero-engine, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement all or part of the steps of the method described in the above embodiments of the present invention.

[0139] This embodiment also provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, it performs any of the optional implementations of the above embodiments, wherein: the method is described above. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0140] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0141] In summary, this invention introduces the CFSFDP density peak clustering algorithm, which can adaptively and automatically extract the most representative typical feature cycles from the preceding / following sub-cycles without pre-setting the number of load levels. This effectively overcomes the blindness of traditional equal-interval or fixed-ratio division methods and significantly improves the scientificity and representativeness of load grading. Regarding load frequency determination, a multi-objective optimization model is constructed with the relative integral absolute error of the load average value and amplitude as dual objectives. Under the premise of keeping the total number of cycles constant and satisfying the minimum frequency constraint for each load level, the optimal cycle number allocation scheme is obtained, thereby achieving a high degree of approximation of the original spectrum in terms of statistical distribution of the program spectrum. The sequence effect of fatigue loads is fully considered. By identifying the parent-child relationship and loading direction characteristics between sub-cycles, and based on the loading direction + relative position principle, each typical feature cycle is inserted back into the main cycle in hierarchical order, effectively reproducing the sequence dependency behavior in non-stationary operating loads. An integrated compilation process was constructed, which organically integrates load preprocessing, typical feature cyclic identification, multi-target frequency optimization and hierarchical sequence back-insertion, significantly simplifying the program spectrum compilation work and improving the standardization, repeatability and engineering implementation efficiency of the method.

[0142] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. 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 be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for compiling a dual-feature program spectrum for an aero-engine, characterized in that: include, The time history of the aero-engine load is obtained and preprocessed to obtain the aero-engine load cycle. Cluster analysis was performed on the load cycles of aero-engines to obtain typical characteristic cycles, and typical characteristic load program blocks were constructed. A multi-objective optimization model was established and particle swarm optimization was used to obtain the optimal load level frequency, and the number of program spectral block loads for typical characteristic cycles of each level was determined. The sequence of program blocks is determined based on the location of typical characteristic load hysteresis loops, and a dual-characteristic program spectrum for aero-engines is compiled.

2. The method for compiling a dual-feature program spectrum for an aero-engine as described in claim 1, characterized in that: The preprocessing includes extracting and filtering peak and valley values ​​from the aero-engine load time history, counting rainflows, and extracting the average amplitude information of the load cycles. The extracted load cycles are divided into primary and secondary cycles and front and back cycles. The maximum load time point in the flight profile is used as the dividing line to divide the aero-engine load time history into front and back cycles. Cluster analysis is performed on the front and back cycles to determine typical characteristic cycles.

3. The method for compiling a dual-feature program spectrum for an aero-engine as described in claim 2, characterized in that: The typical characteristic cycles obtained by cluster analysis of aero-engine load cycles include: Cluster analysis of aero-engine load cycles was performed using the CFSFDP clustering method. For a given two-dimensional sample set, the local density and density distance of each sample point were calculated before identifying the cluster centers. Calculate the Euclidean distance for all sample points; Calculate the local density and density distance based on the Euclidean distance of the sample points; Decision variables are constructed by combining local density and density distance. A decision graph is drawn based on the decision variables in descending order. The cluster centers in the decision graph are selected as typical feature cycles.

4. The method for compiling a dual-feature program spectrum for an aero-engine as described in claim 3, characterized in that: The formula for calculating the Euclidean distance is: Where: d(x) i ,x i Let x be any two points in the dataset. i and x j Euclidean distance, (m) i ,a i ) represents the sample point x i The average amplitude of engine load obtained from the rainflow counting method is statistically analyzed (m). j ,a j ) represents the sample point x j The average amplitude of engine load obtained from the rainflow counting method is statistically analyzed. The formula for calculating the local density is: Where: ρ i For sample point x i Local density, KNN(X) i ) represents x i The set of k nearest neighbor data points centered at the center; All local densities are higher than the sample point x i Among the points, find the sample point with the smallest distance, which is the density distance, denoted as: Where: δ i Let j represent the density distance, and j denote the index of all data points. The decision variables are represented as follows: c i =ρ i ·d i Wherein: γ i These are decision variables.

5. The method for compiling a dual-feature program spectrum for an aero-engine as described in claim 4, characterized in that: The establishment of the multi-objective optimization model includes: With the optimization objectives of minimizing the relative absolute error of the mean and the relative absolute error of the amplitude, and with the constraints of keeping the number of load cycles unchanged before and after spectrum compilation and ensuring that the frequency of each load level in the program spectrum is not less than the number of main cycles, the following multi-objective optimization model is established, expressed as: in: and These are the cumulative frequency curves of the one-dimensional marginal distribution of the mean of the measured load spectrum and the programmed spectrum, respectively. and These are the cumulative frequency curves of the one-dimensional marginal distribution of the measured load spectrum and the programmed spectrum amplitude, respectively. X represents the frequency of each load level, RIAE_m(X) is the mean relative comprehensive absolute error, RIAE_a(X) is the amplitude relative comprehensive absolute error, N is the number of main loops, and n... spec represents the total number of cycles extracted from the original load spectrum, i represents the i-th load level, represents the total number of load levels, and Δ represents the step size in the discretization process.

6. The method for compiling a dual-feature program spectrum for an aero-engine as described in claim 5, characterized in that: The optimal load level frequency obtained by using particle swarm optimization includes: The optimization objectives of minimizing the relative absolute error of the mean and the relative absolute error of the magnitude are transformed into a single-objective optimization problem through weighted summation, expressed as: min F(x)=ω1f1(x)+ω2f2(x) Where: F(x) is the weighted sum objective function, ω1, ω2 are weight coefficients, satisfying ω1+ω2=1, f1(x) is the magnitude objective function, and f2(x) is the mean objective function; After the weighted sum objective function is determined, particle swarm optimization is used to solve for the global optimal solution, and the number of program spectral block loads for typical characteristic cycles at each level is determined.

7. The method for compiling a dual-feature program spectrum for an aero-engine as described in claim 6, characterized in that: The process of determining the order of program spectral blocks based on the location of typical characteristic load hysteresis loops includes: Based on the hysteresis loop theory, the parent-child relationship, load level and load direction of fatigue load are extracted by rainflow counting method to obtain the longitudinal attributes of fatigue load. Each load level contains a different number of load cycles. The load spectrum block sorting strategy is determined based on the back-interpolation rule. Each transverse load level is treated as a single load, and back-interpolated into the reference cycle level by level according to the longitudinal attributes from low to high. The sorting strategy is as follows: The reference cycle is considered as a first-order longitudinal load cycle and serves as the base load for all typical cycle back-interpolation. All secondary longitudinal load cycles are sub-cycles of the reference load cycle. Their relative positions in the primary longitudinal load cycle are determined by the load magnitude, and the shape of the inserted load spectrum is determined by the load cycle direction. The relative position refers to the order of multiple sub-cycles in the parent cycle. In the load increasing half-cycle, the maximum value of the typical characteristic cycle is used as the criterion, from small to large. In the load decreasing half-cycle, the minimum value of the typical characteristic cycle is used as the criterion, from large to small. Determine whether a secondary parent cycle exists for a third-level typical characteristic load based on the load size. Assume the size of a certain third-level typical load is (σ'). max ,σ' min If σ exists min ≥σ' min And σ max ≤σ' max If the second-level typical characteristic loads meet the conditions, then the second-level typical loads are regarded as the nominal parent cycle of the third-level typical loads currently being analyzed. If a third-level typical load has multiple nominal parent cycles, then the one with the smallest Euclidean distance from the real parent cycle is selected as the final selection. For Level 3 typical characteristic loads that do not have a nominal parent cycle, the longitudinal load level is upgraded to Level 2 and then back-inserted; for Level 3 typical loads that have a nominal parent cycle, they are inserted into Level 2 typical characteristic loads according to the cycle direction. Based on the determined typical feature cycle frequency, the corresponding number of times the typical feature cycle is repeated under different levels constitutes a program block spectrum. In the load increase half-cycle, the lower-level typical loads are all inserted back to 1 / 2 of the frequency of the previous level typical load. In the load decrease half-cycle, the lower-level typical loads are inserted back to 1 / 2 of the frequency of the previous level typical load. If there are multiple loads of the same level, the average value of the spectrum block is used as the judgment standard. The one with the smallest average value of the spectrum block is inserted back to the end of the previous level typical load, and the rest are inserted back to 1 / 2 of the frequency of the previous level typical load.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the method for compiling a dual-feature program spectrum for an aero-engine as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the method for compiling a dual-feature program spectrum for an aero-engine as described in any one of claims 1 to 7.

Images