A confidence analysis method for flutter prediction results based on small sample test data

By generating test samples with identical amplitude-frequency characteristics and consistent second-order statistical moments, and combining the ARMA model and Jury stability criterion, the confidence analysis problem under finite sample conditions in aircraft flutter tests was solved, improving the accuracy and efficiency of flutter prediction.

CN112001027BActive Publication Date: 2024-11-22NANJING UNIV OF AERONAUTICS & ASTRONAUTICS
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202010685354.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-07-16
Publication Date
2024-11-22
Estimated Expiration
2040-07-16

AI Technical Summary

Technical Problem

In existing technologies for aircraft flutter testing, the confidence analysis of flutter prediction results based on limited test samples is insufficient, making it difficult to guarantee the accuracy and validity of the test results.

Method used

By generating test samples with the same amplitude-frequency characteristics and consistent second-order statistical moments as the original signal, flutter boundary prediction is performed using the ARMA model and Jury stability criterion. The probability density distribution of the prediction results is estimated using a parameterization method, and a confidence analysis is given.

Benefits of technology

It improves the confidence level of flutter prediction results based on limited test samples, enhances the efficiency of the test and the reliability of the results, and is applicable to model wind tunnel and flight flutter tests.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112001027B_ABST
    Figure CN112001027B_ABST
Patent Text Reader

Abstract

The present invention discloses a confidence analysis method for flutter prediction results under small sample test data, comprising the following steps: (1) using existing flutter test data samples, generating K groups of test samples {S1, S2, ..., S K}; (2) adopting the flutter analysis method based on test data, flutter boundary prediction is performed on K groups of test samples respectively; (3) adopting the parameterization method to estimate the probability density distribution of the flutter prediction result #imgabs0#, and giving the confidence of the prediction result. The present invention is applicable to model wind tunnel flutter test and flight flutter test, and can give the confidence of flutter prediction result based on limited test samples, thereby improving the efficiency of single test and enhancing the credibility of flutter prediction result.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of aircraft flutter testing, and in particular to a flutter prediction result confidence analysis method under small sample test data. Background Art

[0002] Flutter is an aeroelastic instability phenomenon with catastrophic consequences caused by the coupling of aerodynamic, elastic and inertial forces in the aircraft structure. Flutter is not allowed to occur in the entire flight envelope. Since theoretical analysis and numerical simulation are not enough to simulate the actual flight state, flight flutter testing is still an essential part of the aircraft design process for new aircraft or aircraft with major modifications to ensure that there is no flutter phenomenon in the entire flight envelope.

[0003] There are high risks and uncertainties in the flutter test process. To ensure the safety of the test, the test is usually carried out in a "subcritical" state, and the purpose of test verification is achieved by applying external excitation to artificially induce flutter. Therefore, how to predict the real flutter boundary based on the structural response in the "subcritical" state is a key part of the flight flutter test, and its prediction accuracy is directly related to the accuracy and effectiveness of the test results. At present, domestic and foreign scholars have proposed a variety of flutter prediction methods, including velocity damping method based on test data, flutter margin method, etc., as well as robust flutter analysis method based on test models.

[0004] In view of the complexity of flight flutter test conditions and the interference of a large number of random factors during the test, confidence analysis of flutter prediction results is particularly critical, but related research has rarely been carried out. In addition, due to the particularity of flight flutter tests, a single test requires a lot of manpower and material resources, and the test samples obtained are also very limited. However, the confidence of flutter prediction results given by traditional statistical analysis methods requires the support of a large number of test samples.

[0005] Therefore, a new confidence analysis method suitable for flutter tests is needed, which can effectively give the confidence of flutter prediction results based on limited test samples, thereby improving the efficiency of a single test and enhancing the credibility of flutter prediction results. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a confidence analysis method for flutter prediction results under small sample test data, which can effectively give the confidence of flutter prediction results based on limited flutter test samples.

[0007] In order to solve the above technical problems, the present invention provides a method for analyzing the confidence of flutter prediction results under small sample test data, comprising the following steps:

[0008] (1) Using the existing flutter test data samples, generate K groups of test samples {S1, S2, …, S K};

[0009] (2) Using the flutter analysis method based on test data, flutter boundaries are predicted for each of the K groups of test samples;

[0010] (3) Estimating flutter prediction results using parametric methods The probability density distribution of gives the confidence of the prediction result.

[0011] Preferably, in step (1), the existing flutter test data samples are used to generate K groups of test samples {S1, S2, ..., S K}Specific steps include:

[0012] (11) The original signal is rank(x n ) is x n In ascending order, Performing discrete Fourier transform gives:

[0013] in

[0014] (12) Generate a set of random numbers that follow a Gaussian distribution For g n By rank(x n ) are rearranged to obtain r n =g(rank(x n )),n=1,2,…,N;

[0015] (13) Yes Perform discrete Fourier transform, add random phase and perform inverse Fourier transform, and we get

[0016] (14) The original signal is converted into Rearrange the order of

[0017] (15) Yes Performing discrete Fourier transform, we get:

[0018]

[0019] (16)Use y k replace And keep the phase Doing the inverse Fourier transform gives:

[0020]

[0021] (17) x n according to Rearrange the order of

[0022] (18) Repeat steps (15) to (17) until It has the same amplitude distribution as the original signal, that is, Where ε is the specified precision, then A set of data samples to meet the requirements;

[0023] (19) Repeat the above steps until K groups of test samples {S1, S2, …, S K}.

[0024] Preferably, in step (2), using a flutter analysis method based on test data to perform flutter boundary prediction on K groups of test samples respectively specifically comprises the following steps:

[0025] (21) For the kth group of test samples S k , construct ARMA model A(z -1 )s(k)=B(z -1 )e(k);

[0026] (22) The AR coefficient a(n) is estimated using the least squares method;

[0027] (23) According to the Jury stability criterion, the system stability test parameter is obtained: F - (n-1) = det(X n-1 -Y n-1 );

[0028] in

[0029] (24) For the p-order modal system, construct the flutter prediction index F Z,p , as a stability parameter indicating the critical point of chatter, when the system is stable, F Z,p >0, as the test state approaches the flutter critical point, F Z,p Gradually decays until it becomes 0, where

[0030] (25) Linear fitting and extrapolation F Z,p Regarding the relationship curve of flight status, we get F Z,p The state of decaying to 0 is the current predicted flutter boundary p k ;

[0031] (26) Repeat the above steps to obtain the flutter prediction results for all test samples.

[0032] Preferably, in step (3), a parameterized method is used to estimate the flutter prediction result The probability density distribution of , giving the confidence of the prediction result specifically includes the following steps:

[0033] (31) Based on the maximum likelihood estimation method, the average value of the prediction results is calculated and variance σ 2 :

[0034]

[0035] (32) For a given confidence level α, the confidence interval of the flutter prediction result is calculated:

[0036]

[0037] where z α / 2 is the probability distribution of the standard normal distribution, which can be obtained by looking up the table.

[0038] The beneficial effects of the present invention are as follows: the present invention is applicable to model wind tunnel flutter tests and flight flutter tests, and can provide the confidence level of flutter prediction results based on limited test samples, thereby improving the efficiency of a single test and enhancing the credibility of flutter prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a CAD model of the wing of the present invention and a schematic diagram of the sensor layout.

[0040] Figure 2 It is a schematic diagram of the time history of the original wingtip acceleration response (solid line) and the generated data sample (dashed line) of the present invention.

[0041] Figure 3 Schematic diagram of the Fourier spectrum of the original wingtip acceleration response (solid line) and the generated data sample (dashed line) of the present invention.

[0042] Figure 4 It is a schematic diagram of the single flutter prediction result of the present invention.

[0043] Figure 5 Schematic diagram of probability density distribution of the flutter prediction results of the present invention. DETAILED DESCRIPTION

[0044] In this embodiment, a low-speed wind tunnel flutter test of a certain elastic wing is taken as an example. Figure 1As shown, two acceleration sensors are arranged at the wing tip and wing root of the wing respectively. During the test, the wind speed gradually increases until the critical point of flutter. At each wind speed measurement point, after the wing response tends to be stable, the value of the acceleration sensor is recorded. For this flutter test data, the technical method proposed in the present invention is used to perform confidence analysis on the flutter prediction results of the experimental data recorded by sensor No. 1. The detailed implementation process is as follows:

[0045] Step 1: Using the existing flutter test data, generate K groups of test samples {S1, S2, …, S K}, in this example K = 1000, the specific steps are as follows:

[0046] Step 1.1) Record the acceleration response recorded by sensor 1 at a certain wind speed as rank(x n ) is x n In ascending order, Performing discrete Fourier transform gives:

[0047] in

[0048] Step 1.2) Generate a set of random numbers that follow a Gaussian distribution For g n By rank(x n ) are rearranged to obtain r n =g(rank(x n )),n=1,2,…,N;

[0049] Step 1.3) Perform discrete Fourier transform, add random phase and perform inverse Fourier transform, and we get

[0050] Step 1.4) Convert the original signal into Rearrange the order of

[0051] Step 1.5) Performing discrete Fourier transform, we get:

[0052]

[0053] Step 1.6) Use y k replace And keep the phase Doing the inverse Fourier transform gives:

[0054]

[0055] Step 1.7) For x n according to Rearrange the order of

[0056] Step 1.8) Repeat steps 1.5) to 1.7) until It has the same amplitude distribution as the original signal, that is, Where ε is the specified precision, in this example ε = 10 -4 ,but A set of data samples to meet the requirements;

[0057] Step 1.9) Figure 2 and Figure 3 The figure shows a set of data samples obtained by the above steps at a certain wind speed. The above steps are repeated at different wind speed measurement points until K groups of test samples {S1, S2, …, S K}.

[0058] Step 2: Use the flutter analysis method based on test data to predict the flutter boundary of K groups of test samples, which specifically includes the following steps:

[0059] Step 2.1) For the kth group of test samples S k , based on the experimental data at a certain wind speed, construct the ARMA model A(z -1 )s(k)=B(z -1 )e(k), where the AR term is represented by A(z -1 )=1+a1z -1 +...+a n z -n , the MA term is expressed as B(z -1 )=1+b1z -1 +...+b m z -m , e(k) is a random sequence with mean 0;

[0060] Step 2.2) Use the least squares method to estimate the AR coefficient a(n), and then get the characteristic equation of the system G(z) = z n +a1z n-1 +...+a n ;

[0061] Step 2.3) According to the Jury stability criterion, the system stability test parameter is obtained: F - (n-1) = det(X n-1 -Y n-1 );

[0062] in

[0063] Step 2.4) For the p-order modal system, construct the flutter prediction index F Z,p , as a stability parameter indicating the critical point of chatter, when the system is stable, F Z,p >0, as the test state approaches the flutter critical point, F Z,p Gradually decays until it becomes 0, where

[0064] Step 2.5) Linear fitting and extrapolation of F Z,p Regarding the relationship curve of dynamic pressure, we can get F Z,p The state of decaying to 0 is the current predicted flutter boundary p k ; Figure 4 The flutter prediction results for a single test sample are shown. The horizontal axis in the figure is the normalized dynamic pressure, i.e. where v i is the wind speed at the current measuring point, v f is the test flutter speed, and the ordinate is the flutter prediction index F Z,p , linear fitting and extrapolation of the relationship curve between the two yields a predicted flutter point of 0.976;

[0065] Step 2.6) Repeat the above steps to obtain the flutter prediction results for all test samples

[0066] Step 3: Estimate the flutter prediction results using a parametric approach The probability density distribution of , gives the confidence of the prediction result, which specifically includes the following steps:

[0067] Step 3.1) Based on the maximum likelihood estimation method, calculate the average value of the prediction results and variance σ 2 :

[0068]

[0069] Step 3.2) For a given confidence level α, calculate the confidence interval of the flutter prediction result:

[0070]

[0071] where z α / 2 is the probability distribution of the standard normal distribution, which can be obtained by looking up the table. Figure 5 The results of the flutter prediction are shown below. The probability density distribution of , at a confidence level of α = 5%, the confidence boundary of the current predicted chatter result is p∈(0.942,1.039).

[0072] In this embodiment, the confidence analysis of the flutter prediction result is performed only on the test data recorded by sensor No. 1. For multi-channel test data, the technical method proposed by the present invention can also be used to process the data of each channel separately and then average them to obtain a more accurate analysis result.

[0073] The present invention discloses a confidence analysis method for flutter prediction results under small sample test data, which is applicable to model wind tunnel flutter tests and flight flutter tests. The confidence of flutter prediction results can be given based on limited test samples, thereby improving the efficiency of a single test and enhancing the credibility of flutter prediction results.

Claims

1. A confidence analysis method for flutter prediction results under small sample test data, characterized in that: The steps include: (1) Using the existing flutter test data samples, generate K groups of test samples {S1, S2, …, S K }; specifically includes the following steps: (11) The original signal is For x n In ascending order, Performing discrete Fourier transform gives: in (12) Generate a set of random numbers that follow a Gaussian distribution For g n By rank(x n ) are rearranged to obtain r n =g(rank(x n )),n=1,2,…,N; (13) Yes Perform discrete Fourier transform, add random phase and perform inverse Fourier transform, and we get (14) The original signal is converted into Rearrange the order of (15) Yes Perform discrete Fourier transform, where i = 0, 1, ..., and we get: (16)Use y k replace And keep the phase Doing the inverse Fourier transform gives: (17) x n according to Rearrange the order of (18) Repeat steps (15) to (17) until It has the same amplitude distribution as the original signal, that is, Where ε is the specified precision, then A set of data samples to meet the requirements; (19) Repeat the above steps until K groups of test samples {S1, S2, …, S K }; (2) Using the flutter analysis method based on test data, the flutter boundaries of K groups of test samples are predicted respectively; (3) Estimating flutter prediction results using parametric methods The probability density distribution of is used to give the confidence of the prediction result; the specific steps include: (31) Based on the maximum likelihood estimation method, the average value of the prediction results is calculated and variance σ 2 : (32) For a given confidence level α, the confidence interval of the flutter prediction result is calculated: where z α2 is the probability distribution of the standard normal distribution, which can be obtained by looking up the table.

2. The method for analyzing the confidence of flutter prediction results under small sample test data according to claim 1, characterized in that: In step (2), a flutter analysis method based on test data is used to predict the flutter boundaries of K groups of test samples respectively, which specifically includes the following steps: (21) For the kth group of test samples S k , construct ARMA model A(z -1 )s(k)=B(z -1 )e(k); (22) The AR coefficient a(n) is estimated using the least squares method; (23) According to the Jury stability criterion, the system stability test parameter is obtained: F - (n-1) = det(X n-1 -Y n-1 ); in (24) For the p-order modal system, construct the flutter prediction index F Z,p , as a stability parameter indicating the critical point of chatter, when the system is stable, F Z,p >0, as the test state approaches the flutter critical point, F Z,p Gradually decays until it becomes 0, where (25) Linear fitting and extrapolation F Z,p Regarding the relationship curve of flight status, we get F Z,p The state of decaying to 0 is the current predicted flutter boundary p k ; (26) Repeat the above steps to obtain the flutter prediction results for all test samples.