CN114459644B - Landing gear drop shock load identification method based on optical fiber strain response and Gaussian process - Google Patents

Abstract

The invention relates to an undercarriage drop load identification method based on optical fiber strain response and Gaussian process, which is characterized in that S FBG sensors distributed along the direction of a bearing structure at preset positions on a target undercarriage supporting arm are used for constructing drop load prediction data which respectively comprise noise at each time point in the sinking duration of a test working condition and correspond to a target undercarriage according to Gaussian process regression models f-GP (mu (-) and k (-) on the basis of S FBG sensors

Training under training conditions and testing conditions in each constructed sample by combining with the corresponding overflow conditions compared with the actual measured values to obtain a falling load identification model corresponding to the target undercarriage; and then in practical application, the model can be identified through the falling shock load, the falling shock load of the target undercarriage can be identified in real time, and the working efficiency of health monitoring of the target undercarriage can be effectively improved.

Description

Landing gear drop shock load identification method based on optical fiber strain response and Gaussian process

technical field

The invention relates to a landing gear drop shock load identification method based on optical fiber strain response and Gaussian process, and belongs to the technical field of landing gear drop shock load identification.

Background technique

Studies have shown that flight accidents often occur during take-off and landing, and more than 50% of safety accidents occur during take-off and landing. The fundamental reason is that only the static strength of the structure is considered in the design of the landing gear. Sexual considerations are less. Therefore, real-time monitoring of landing gear shock loads has become an important part of aircraft health management and can provide technical support for future multifunctional/integrated landing gear designs.

Contents of the invention

The technical problem to be solved by the present invention is to provide a landing gear drop-shock load identification method based on optical fiber strain response and Gaussian process, which can realize efficient and fast drop-shock load monitoring for landing gear by using the feature characterization method.

In order to solve the above-mentioned technical problems, the present invention adopts the following technical solutions: the present invention designs a landing gear drop-shock load identification method based on optical fiber strain response and Gaussian process, based on the preset position on the target landing gear support arm distributed along the load-bearing structure direction S FBG sensors are used to identify the drop-shock load of the target landing gear; according to step A to step H, the drop-shock load identification model corresponding to the target landing gear is obtained; then step i is executed in real time to realize the drop-shock load identification of the target landing gear real-time identification of

Step A. Based on the preset number of M working conditions formed by the preset different launch weights carried on the support arm of the target landing gear and the corresponding preset different sinking speeds of the target landing gear, each time in the sinking duration of each working condition is obtained point, the strain response data set corresponding to each FBG sensor on the support arm of the target landing gear, and the measurement data of each time point in the sinking duration of various working conditions, the drop shock load measurement data of the target landing gear, and then enter step B;

Step B. Based on m≤M/2, randomly select m kinds of working conditions among the M working conditions to form each training working condition, and at the same time select m working conditions randomly selected from the remaining working conditions among the M working conditions to form Each test working condition, and construct a random one-to-one relationship between each training working condition and each testing working condition to form m samples, and then enter step C;

Step C. randomly select s FBG sensors from S FBG sensors as each target FBG sensor, and then enter step D;

Step D. Based on the input of the strain response data set corresponding to each FBG sensor on the support arm of the target landing gear at each time point within the sinking time of the working condition, the corresponding noise contained at each time point within the sinking time of the working condition and the target landing gear The drop shock load prediction data is the output Gaussian process regression model f, combined with f～GP(μ(·), k(·)), the target landing gear support arm at each time point within the sinking duration of the training condition in the sample The strain response data group X _i corresponding to each FBG sensor above is measured at each time point within the sinking time of the training condition in this sample, and the target landing gear drop shock load measurement data F _{i is measured,} and the sinking time of the test condition in this sample For each time point and the strain response data set Y _j corresponding to each FBG sensor on the target landing gear support arm, the following model is constructed:

Among them, 1≤i≤I, I represents the number of each time point in the sinking time of the training condition in the sample; J represents the number of each time point in the sinking time of the test condition in the corresponding sample, F _{j, pre} -represents the Each time point within the sinking time of the test condition in the sample, the target landing gear corresponds to the noise-free drop-shock load prediction data; μ( ) represents the mean function, and μ( ) includes μ _i and μ _j ; σ ² represents the starting The noise variance corresponding to the load history response of the landing gear, I′ represents the identity matrix, K(X _i , _Xi ) is the covariance matrix of the strain response characteristics of the landing gear shock history corresponding to the sample training conditions, K(X _i , Y _j ) is the strain response characteristics of the landing gear drop history corresponding to the covariance matrix between the training conditions and the test conditions in the sample, K(Y _j , Y _j ) is the strain response characteristics of the landing gear drop history corresponding to the sample The covariance matrix of the test condition, μ _i and μ _j represent the mean function of the strain response characteristics of the landing gear drop process corresponding to the training condition and the test condition in the sample; k(·) represents the covariance function; N represents the joint Normal distribution symbol; then go to step E;

Step E. Calculate the value of each hyperparameter involved in the covariance function k( ), and combine the model constructed in step D and the Bayesian principle to obtain the values of each hyperparameter in the sample during the sinking time of the test condition. The time point, the target landing gear correspond to the noise-free drop shock load prediction data F _{j, the pre} -posterior distribution, and then enter step F;

然后进入步骤G；Step F. Obtain the sinking duration of the test condition in the sample according to the predicted posterior distribution of each time point within the sinking duration of the test condition, the target landing gear corresponding to the noise-free drop-shock load prediction data Fj _{, and the predicted} posterior distribution Each time point in the target landing gear corresponds to the drop shock load prediction data including noise

Then go to step G;

以及该样本中测试工况下沉时长内各时间点、目标起落架落震载荷测量数据F_j，测，以两者之间差值与预设差值阈值的比较构成溢出条件，然后进入步骤H；Step G. According to the drop-shock load prediction data containing noise corresponding to the target landing gear at each time point within the sinking duration of the test condition in the sample

As well as the measurement data F _{j of} the drop shock load of the target landing gear at each time point within the sinking time of the test condition in the sample, the overflow condition is formed by comparing the difference between the two with the preset difference threshold, and then enter the step H;

Step H. Based on each sample, train the Gaussian process regression model f according to the above step D to step G, and obtain the trained Gaussian process regression model, which is the drop shock load identification model corresponding to the target landing gear;

Step i. Obtain the strain response data set corresponding to each FBG sensor on the support arm of the target landing gear at each time point within the sinking duration of the real-time working condition of the target landing gear, and apply the drop-shock load identification model corresponding to the target landing gear to obtain the target The landing gear corresponds to each time point within the sinking duration of the working condition, and the target landing gear corresponds to the drop-shock load prediction data including noise, so as to realize real-time identification of the target landing gear drop-shock load.

As a preferred technical solution of the present invention: the μ _i =0, μ _j =0, the covariance function k(·) is as follows:

k(a, b) = λ ₁ k _RBF (a, b) + λ ₂ k _PKF (a, b) + λ ₃ k _Noise (a, b)

in:

k _PKF (a, b) = (a b+1) ^q

k _Noise (a, b) = σ _N ² δ ^*

In the formula, λ ₁ , λ ₂ , and λ ₃ represent preset weight factors, and satisfy 0<λ ₁ ≤0.5, 0<λ ₂ ≤0.5, 0<λ ₃ ≤0.5, λ ₁ +λ ₂ +λ ₃ =1 ; σ _f , σ _l , σ _N represent the various hyperparameters involved in the covariance function k(·); q represents the polynomial degree; δ ^* represents the Kronecker function.

As a preferred technical solution of the present invention: in the step E, calculate and obtain the value of each hyperparameter involved in the covariance function k ( ) as follows;

Firstly, the maximum likelihood function is constructed by using the strain response characteristics of the shock history of the target landing gear to correspond to the test conditions in the sample, and the unknown hyperparameters involved in the covariance function k( ), and according to the Bayesian principle, according to the likelihood The integral of P(f _j |f _i ,X _i ,θ) and prior P(f _i |X _i ,θ), that is, the marginal likelihood P(f _j |X _i ,θ) is as follows:

P(f _j |X _i , θ)=∫P(f _j |f _i ,X _i ,θ)P(f _i |X _i ,θ)df _i =N(f _i |0,K(X _i , X _i )+σ ² I′)

Among them, θ represents a vector composed of unknown hyperparameters involved in the covariance function k( );

Then, for the marginal likelihood P(f _j |X _i , θ), take the logarithm to obtain the logarithmic marginal likelihood as follows:

Among them, N′ represents the number of time points in the sinking time of the working condition, that is, the sampling frequency;

Finally, according to the principle of matrix calculus, for each unknown hyperparameter θ _i involved in the covariance function k(·), the following formula is used:

Obtain the value of θ _i corresponding to the maximum probability by derivation, that is, obtain the value of the hyperparameter involved in the covariance function k( ), and then obtain the values of the unknown hyperparameters involved in the covariance function k( ); Wherein, Tr represents the trace of the matrix, α=[K(X _i ,X _i )+σ ² I′] ^-1 F _{i, measure} , and T represents the transpose.

As a preferred technical solution of the present invention: in the step F, according to the drop shock load prediction data F _{j corresponding to the target landing gear at each time point within the sinking duration of the test condition in the sample and without noise, the pre} -prediction distribution, according to the following formula:

cov(F _{j, pre} )=K(Y _j , Y _j )-K(Y _j ,X _i )[K(X _i ,X _i )+σ ² I′] ^-1 K(X _i ,Y _j )

其中，cov(F_i，预)表示目标起落架落震载荷历程响应预测值的置信水平。Obtain the drop-shock load prediction data corresponding to the target landing gear including noise at each time point within the sinking time of the test condition in this sample

Among them, cov(F _{i, pre} ) represents the confidence level of the predicted value of the target landing gear drop shock load history response.

The landing gear drop shock load identification method based on optical fiber strain response and Gaussian process described in the present invention adopts the above technical scheme compared with the prior art, and has the following technical effects:

并结合与实际测量值之间比较所对应的溢出条件，在所构建各样本中训练工况、测试工况下进行训练，获得目标起落架所对应的落震载荷辨识模型；进而在实际应用中，即可通过该落震载荷辨识模型，对目标起落架的落震载荷实现实时辨识，能够有效提高目标起落架健康监测的工作效率。(1) The landing gear shock load identification method based on the optical fiber strain response and the Gaussian process designed by the present invention is based on the S FBG sensors distributed along the load-bearing structure direction at the preset position on the target landing gear support arm, according to the Gaussian process regression Model f～GP(μ(·), k(·)), construct the drop-shock load prediction data corresponding to the target landing gear including noise at each time point within the sinking duration of the test condition

Combined with the overflow conditions corresponding to the comparison with the actual measured values, training is carried out under the training conditions and test conditions in the constructed samples to obtain the drop shock load identification model corresponding to the target landing gear; and then in practical applications , the drop shock load identification model can be used to realize real-time identification of the drop shock load of the target landing gear, which can effectively improve the work efficiency of the health monitoring of the target landing gear.

Description of drawings

Fig. 1 is a schematic diagram of the application structure of the optical fiber identification of landing gear shock load in the design of the present invention;

Fig. 2 is a flow chart of optical fiber identification for landing gear shock load based on Gaussian process regression.

Detailed ways

The specific implementation manners of the present invention will be described in further detail below in conjunction with the accompanying drawings.

The landing gear drop shock load identification method based on optical fiber strain response and Gaussian process designed by the present invention is based on S FBG sensors distributed along the direction of the bearing structure at the preset position on the support arm of the target landing gear to realize the landing gear for the target landing gear. Seismic load identification.

Regarding the design of S FBG sensors arranged in the preset position distribution, in practical applications, as shown in Figure 1, such as selecting a monitoring area along the bearing structure direction on the outer surface of the target landing gear support arm, and arranging 4 sensors in the monitoring area FBG sensors, respectively denoted as FBG1, FBG2, FBG3,..., FBGk, where 1≤S≤10; the location of FBG sensors should be based on the structure of the support arm and the characteristics of the load, when the strain amplitude is large and the strain gradient The number of FBG sensors can be appropriately increased for larger key parts of the measurement; in areas with small strain gradients, the number of FBG sensors can be reduced; S FBG sensors are connected in series using wavelength division multiplexing technology, These FBG sensors are then pasted on the outer surface of the support arm structure to form a FBG sensor monitoring network, as shown in Figure 1, which is used to collect the strain response signals of the support arm during the shock of the target landing gear.

The reason for choosing the FBG sensor is that the fiber optic Bragg grating sensor (FBG) uses optical signals as the carrier, and has unique advantages such as corrosion resistance, light weight, and anti-electromagnetic interference. It can also realize multi-parameter monitoring and distributed sensing layout. For aviation The large-scale and complex structure of the spacecraft can better reflect its unique advantages.

Based on the distribution of S FBG sensors on the preset positions of the target landing gear support arm along the direction of the bearing structure, in practical applications, as shown in Figure 2, according to steps A to H, the drop shock load corresponding to the target landing gear is obtained Identify the model.

Step A. Based on the preset number of M working conditions formed by the preset different launch weights carried on the support arm of the target landing gear and the corresponding preset different sinking speeds of the target landing gear, each time in the sinking duration of each working condition is obtained point, the strain response data set corresponding to each FBG sensor on the target landing gear support arm, and the measurement data of each time point within the sinking duration of various working conditions and the target landing gear drop shock load measurement data, and then enter step B.

Step B. Based on m≤M/2, randomly select m kinds of working conditions in the M working conditions to form each training working condition, and at the same time select m working conditions randomly selected from the remaining working conditions in the M working conditions to form Each test condition, and construct a random one-to-one relationship between each training condition and each test condition to form m samples, and then enter step C.

Step C. Randomly select s FBG sensors from S FBG sensors as each target FBG sensor, and then enter step D.

Step D. Based on the input of the strain response data set corresponding to each FBG sensor on the support arm of the target landing gear at each time point within the sinking time of the working condition, the corresponding noise contained at each time point within the sinking time of the working condition and the target landing gear The drop shock load prediction data is the output Gaussian process regression model f, combined with f～GP(μ(·), k(·)), the target landing gear support arm at each time point within the sinking duration of the training condition in the sample The strain response data group X _i corresponding to each FBG sensor above is measured at each time point within the sinking time of the training condition in this sample, and the target landing gear drop shock load measurement data F _{i is measured,} and the sinking time of the test condition in this sample For each time point and the strain response data set Y _j corresponding to each FBG sensor on the target landing gear support arm, the following model is constructed:

Among them, 1≤i≤I, I represents the number of each time point in the sinking time of the training condition in the sample; J represents the number of each time point in the sinking time of the test condition in the corresponding sample, F _{j, pre} -represents the Each time point within the sinking duration of the test condition in the sample, the target landing gear corresponds to the noise-free drop-shock load prediction data; μ(·) represents the mean value function, and μ(·) includes μi and _μj ; ^σ2 represents the landing gear The noise variance corresponding to the drop shock load history response, I′ represents the identity matrix, K(X _i , X _i ) is the covariance matrix of the strain response characteristics of the landing gear drop shock history corresponding to the sample training conditions, K(X _i , Y _j ) is the strain response characteristics of the landing gear drop process corresponding to the covariance matrix between the training conditions and the test conditions in the sample, K(Y _j , Y _j ) is the strain response characteristics of the landing gear drop process corresponding to the sample test The covariance matrix of the working conditions, μ _i and μ _j represent the mean function of the strain response characteristics of the landing gear shock history corresponding to the training and testing conditions in the sample; k( ) represents the covariance function; N represents the joint normal state distribution symbol; then go to Step E.

In specific calculation applications, based on μ _i =0, μ _j =0, the above covariance function k(·) is as follows:

k(a, b) = λ ₁ k _RBF (a, b) + λ ₂ k _PKF (a, b) + λ ₃ k _Noise (a, b)

in:

k _PKF (a, b) = (a b+1) ^q

k _Noise (a, b) = σ _N ² δ ^*

In the formula, λ ₁ , λ ₂ , and λ ₃ represent preset weight factors, and satisfy 0<λ ₁ ≤0.5, 0<λ ₂ ≤0.5, 0<λ ₃ ≤0.5, λ ₁ +λ ₂ +λ ₃ =1 ; σ _f , σ _l , σ _N represent the various hyperparameters involved in the covariance function k(·); q represents the polynomial degree; δ ^* represents the Kronecker function.

In practical application, according to the Bayesian principle, from the model constructed in step D, the drop-shock load prediction corresponding to the target landing gear without noise at each time point in the test condition of the sample can be obtained Data Fj, pre-distributed as follows:

Step E. Calculate the value of each hyperparameter involved in the covariance function k( ), and combine the model constructed in step D and the Bayesian principle to obtain the values of each hyperparameter in the sample during the sinking time of the test condition. The time point and the target landing gear correspond to the noise-free drop shock load prediction data F _{j, the pre} -posterior distribution, and then enter step F.

In the above step E, the values of the various hyperparameters involved in the covariance function k(·) are calculated and obtained specifically as follows.

Firstly, the maximum likelihood function is constructed by using the strain response characteristics of the shock history of the target landing gear to correspond to the test conditions in the sample, and the unknown hyperparameters involved in the covariance function k( ), and according to the Bayesian principle, according to the likelihood The integral of P(f _j |f _i ,X _i ,θ) and prior P(f _i |X _i ,θ), that is, the marginal likelihood P(f _j |X _i ,θ) is as follows:

P(f _j |X _i , θ)=∫P(f _j |f _i ,X _i ,θ)P(f _i |X _i ,θ)df _i =N(f _i |0,K(X _i , X _i )+σ ² I′)

Among them, θ represents a vector composed of unknown hyperparameters involved in the covariance function k(·).

Then, for the marginal likelihood P(f _j |X _i , θ), take the logarithm to obtain the logarithmic marginal likelihood as follows:

Among them, N' represents the number of time points in the sinking time of the working condition, that is, the sampling frequency.

Finally, according to the principle of matrix calculus, for each unknown hyperparameter θ _i involved in the covariance function k(·), the following formula is used:

Obtain the value of θ _i corresponding to the maximum probability by derivation, that is, obtain the value of the hyperparameter involved in the covariance function k( ), and then obtain the values of the unknown hyperparameters involved in the covariance function k( ); Among them, Tr represents the trace of the matrix, α=「K(X _i ,X _i )+σ ² I′] ^-1 F _{i, measured} , and T represents the transpose.

After obtaining the values of the hyperparameters involved in the covariance function k( ), determine the covariance function k( ) based on the values of the hyperparameters, and combine the model constructed in step D and the Bayesian principle , to obtain the predicted posterior distribution of the noise-free drop-shock load prediction data F _j corresponding to the target landing gear at each time point within the sinking duration of the test condition in this sample, and then enter step F.

Step F. According to each time point in the sinking time of the test condition in the sample, the target landing gear corresponds to the drop shock load prediction data Fj without noise _{, and the predicted} posterior distribution is according to the following formula:

cov(F _{j, pre} )=K(Y _j , Y _j )-K(Y _j ,X _i )[K(X _i ,X _i )+σ ² I′] ^-1 K(X _i ,Y _j )

其中，cov(F_j，预)表示目标起落架落震载荷历程响应预测值的置信水平，然后进入步骤G。Obtain the drop-shock load prediction data corresponding to the target landing gear including noise at each time point within the sinking time of the test condition in this sample

Among them, cov(F _{j, pre} ) represents the confidence level of the predicted value of the target landing gear drop shock load history response, and then enter step G.

以及该样本中测试工况下沉时长内各时间点、目标起落架落震载荷测量数据F_j，测，以两者之间差值与预设差值阈值的比较构成溢出条件，然后进入步骤H。Step G. According to the drop-shock load prediction data containing noise corresponding to the target landing gear at each time point within the sinking duration of the test condition in the sample

As well as the measurement data F _{j of} the drop shock load of the target landing gear at each time point within the sinking time of the test condition in the sample, the overflow condition is formed by comparing the difference between the two with the preset difference threshold, and then enter the step H.

Step H. Based on each sample, train the Gaussian process regression model f according to the above steps D to G, and obtain the trained Gaussian process regression model, which is the shock load identification model corresponding to the target landing gear.

Based on the acquisition of the above-mentioned drop-shock load identification model corresponding to the target landing gear, and further based on this model, step i is executed in real time to realize real-time identification of the drop-shock load of the target landing gear.

Step i. Obtain the strain response data set corresponding to each FBG sensor on the support arm of the target landing gear at each time point within the sinking duration of the real-time working condition of the target landing gear, and apply the drop-shock load identification model corresponding to the target landing gear to obtain the target The landing gear corresponds to each time point within the sinking duration of the working condition, and the target landing gear corresponds to the drop-shock load prediction data including noise, so as to realize real-time identification of the target landing gear drop-shock load.

并结合与实际测量值之间比较所对应的溢出条件，在所构建各样本中训练工况、测试工况下进行训练，获得目标起落架所对应的落震载荷辨识模型；进而在实际应用中，即可通过该落震载荷辨识模型，对目标起落架的落震载荷实现实时辨识，能够有效提高目标起落架健康监测的工作效率。The landing gear shock load identification method based on the optical fiber strain response and the Gaussian process designed in the above technical scheme is based on S FBG sensors distributed along the direction of the bearing structure at the preset position on the target landing gear support arm, and according to the Gaussian process regression model f ～GP(μ(·), k(·)), construct the drop-shock load prediction data corresponding to the target landing gear including noise at each time point within the sinking duration of the test condition

Combined with the overflow conditions corresponding to the comparison with the actual measured values, training is carried out under the training conditions and test conditions in the constructed samples to obtain the drop shock load identification model corresponding to the target landing gear; and then in practical applications , the drop shock load identification model can be used to realize real-time identification of the drop shock load of the target landing gear, which can effectively improve the work efficiency of the health monitoring of the target landing gear.

The embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments, and can also be made without departing from the gist of the present invention within the scope of knowledge possessed by those of ordinary skill in the art. Variations.

Claims (4)

1. The landing gear drop-shock load identification method based on optical fiber strain response and Gaussian process, based on S FBG sensors distributed along the direction of the bearing structure at the preset position on the support arm of the target landing gear, realizes the drop-shock load of the target landing gear identification; it is characterized in that: according to step A to step H, obtain the drop-shock load identification model corresponding to the target landing gear; then execute step i in real time to realize the real-time identification of the target landing gear drop-shock load; Step A. Based on the preset number of M working conditions formed by the preset different launch weights carried on the support arm of the target landing gear and the corresponding preset different sinking speeds of the target landing gear, each time within the duration of the sinking of each working condition is obtained. point, the strain response data set corresponding to each FBG sensor on the support arm of the target landing gear, and the measurement data of each time point within the sinking duration of various working conditions, and the drop shock load measurement data of the target landing gear, and then enter step B; Step B. Based on m≤M/2, randomly select m kinds of working conditions among the M working conditions to form each training working condition, and at the same time select m working conditions randomly selected from the remaining working conditions among the M working conditions to form Each test condition, and construct a random one-to-one relationship between each training condition and each test condition to form m samples, and then enter step C; Step C. randomly select s FBG sensors from S FBG sensors as each target FBG sensor, and then enter step D; Step D. Based on the input of the strain response data set corresponding to each FBG sensor on the support arm of the target landing gear at each time point within the sinking time of the working condition, the corresponding noise contained at each time point within the sinking time of the working condition and the target landing gear The drop shock load prediction data is the output Gaussian process regression model f, combined with f～GP(μ(·),k(·)), the target landing gear support arm at each time point within the sinking duration of the training condition in the sample The strain response data set X _i corresponding to each FBG sensor above is measured at each time point within the sinking time of the training condition in this sample, and the target landing gear drop shock load measurement data F _{i, is measured} , and the sinking time of the test condition in this sample is For each time point and the strain response data set Y _j corresponding to each FBG sensor on the target landing gear support arm, the following model is constructed:

Among them, 1≤i≤I, I represents the number of each time point in the sinking time of the training condition in the sample; J represents the number of each time point in the sinking time of the test condition in the corresponding sample, F _{j, pre} -represents the Each time point within the sinking time of the test condition in the sample, the target landing gear corresponds to the noise-free drop-shock load prediction data; μ( ) represents the mean function, and μ( ) includes μ _i and μ _j ; σ ² represents the starting The noise variance corresponding to the load history response of the landing gear drop, I' represents the identity matrix, K(X _i ,X _i ) is the covariance matrix of the strain response characteristics of the landing gear drop shock history corresponding to the sample training conditions, K(X _i , Y _j ) is the strain response characteristics of the landing gear drop history corresponding to the covariance matrix between the training conditions and the test conditions in the sample, K(Y _j , Y _j ) is the strain response characteristics of the landing gear drop history corresponding to the sample The covariance matrix of the test condition, μ _i and μ _j represent the mean function of the strain response characteristics of the landing gear drop process corresponding to the training condition and the test condition in the sample; k(·) represents the covariance function; N represents the joint Normal distribution symbol; then go to step E; Step E. Calculate the value of each hyperparameter involved in the covariance function k( ), and combine the model constructed in step D and the Bayesian principle to obtain the values of each hyperparameter in the sample during the sinking time of the test condition. The time point, the target landing gear corresponding to the noise-free drop shock load prediction data Fj _{, and the predicted} posterior distribution, and then enter step F; Step F. According to _{the predicted posterior distribution of each time point in the sinking time of the test condition in the sample and the target landing gear corresponding to the noise-free drop-shock load prediction data Fj,} the sinking time of the test condition in the sample is obtained Each time point in the target landing gear corresponds to the drop shock load prediction data including noise

Then go to step G; Step G. According to the drop-shock load prediction data containing noise corresponding to the target landing gear at each time point within the sinking duration of the test condition in the sample

And each time point in the sinking time of the test condition in the sample, the target landing gear drop shock load measurement data F _{j, are measured} , and the overflow condition is formed by comparing the difference between the two with the preset difference threshold, and then enter the step H; Step H. Based on each sample, train the Gaussian process regression model f according to the above step D to step G, and obtain the trained Gaussian process regression model, which is the drop shock load identification model corresponding to the target landing gear; Step i. Obtain the strain response data set corresponding to each FBG sensor on the support arm of the target landing gear at each time point within the sinking duration of the real-time working condition of the target landing gear, and apply the drop-shock load identification model corresponding to the target landing gear to obtain the target The landing gear corresponds to each time point within the sinking duration of the working condition, and the target landing gear corresponds to the drop-shock load prediction data including noise, so as to realize real-time identification of the target landing gear drop-shock load. 2. The landing gear drop shock load identification method based on optical fiber strain response and Gaussian process according to claim 1, characterized in that: said μ _i =0, μ _j =0, and said covariance function k(·) is as follows : k(a,b)＝λ ₁ k _RBF (a,b)+λ ₂ k _PKF (a,b)+λ ₃ k _Noise (a,b) in:

k _PKF (a,b)=(a b+1) ^q k _Noise (a,b)＝σ _N ² δ ^* In the formula, λ ₁ , λ ₂ , and λ ₃ represent preset weight factors, and satisfy 0<λ ₁ ≤0.5, 0<λ ₂ ≤0.5, 0<λ ₃ ≤0.5, λ ₁ +λ ₂ +λ ₃ =1 ; σ _f , σ _l , σ _N represent the various hyperparameters involved in the covariance function k(·); q represents the polynomial degree; δ ^* represents the Kronecker function. 3. according to claim 2, based on the optical fiber strain response and the landing gear drop shock load identification method of Gaussian process, it is characterized in that: in the described step E, as follows, calculate and obtain involved in the covariance function k ( ) The value of each hyperparameter of the target landing gear; firstly, the maximum likelihood function is constructed by using the strain response characteristics of the shock history of the target landing gear corresponding to the test conditions in the sample, and the unknown hyperparameters involved in the covariance function k( ), and according to the Yeesian principle, according to the integral of likelihood P(f _j |f _i ,X _i ,θ) and prior P(f _i |X _i ,θ), that is, marginal likelihood P(f _j |X _i ,θ) as follows: P(f _j |X _i ,θ)＝∫P(f _j |f _i ,X _i ,θ)P(f _i |X _i ,θ)df _i ＝N(f _i |0,K(X _i , X _i )+σ ² I') Among them, θ represents a vector composed of unknown hyperparameters involved in the covariance function k( ); Then for the marginal likelihood P(f _j |X _i ,θ), take the logarithm to obtain the logarithmic marginal likelihood as follows:

Among them, N' represents the number of time points in the sinking time of the working condition, that is, the sampling frequency; Finally, according to the principle of matrix calculus, for each unknown hyperparameter θ _i involved in the covariance function k(·), the following formula is used:

Obtain the value of θ _i corresponding to the maximum probability by derivation, that is, obtain the value of the hyperparameter involved in the covariance function k( ), and then obtain the values of the unknown hyperparameters involved in the covariance function k( ); Wherein, Tr represents the trace of the matrix, α=[K(X _i ,X _i )+σ ² I'] ^-1 F _{i, measure} , and T represents the transpose. 4. The landing gear drop-shock load identification method based on optical fiber strain response and Gaussian process according to claim 1, characterized in that: in the step F, according to the test conditions in the sample, each time point, target The landing gear corresponds to the noise-free drop-shock load prediction data F _{j, the predicted} posterior distribution, according to the following formula:

cov(F _j,pre )＝K(Y _j ,Y _j )-K(Y _j ,X _i )[K(X _i ,X _i )+σ ² I'] ^-1 K(X _i ,Y _j ) Obtain the drop-shock load prediction data corresponding to the target landing gear including noise at each time point within the sinking time of the test condition in this sample

Among them, cov(F _j,pre ) represents the confidence level of the predicted value of the target landing gear drop shock load history response.

Images