Wing skin damage identification method based on support vector machine and fiber optic sensing technology

Abstract

The present invention discloses a wing skin damage identification method based on support vector machine and optical fiber sensing technology, the method comprising the following steps: establishing a three-dimensional model of the wing skin, and performing modal analysis on the three-dimensional model of the wing skin to obtain a modal vibration matrix; calculating effective independent coefficients for the modal vibration matrix, applying a decreasing function to the effective independent coefficients, and determining the fiber grating sensor point location through a modal confidence factor; obtaining fiber grating sensing signals of the wing skin in normal conditions and when different positions are impacted to different degrees, and performing signal fusion of multiple fiber grating sensing signals using a support matrix method; extracting fusion signal features and constructing a training set; establishing a support vector machine model, and performing wing skin damage identification after the model training is completed. The wing skin damage identification method of the present invention has high efficiency, low cost, anti-electromagnetic interference, rapid response, and high accuracy, and has obvious advantages over traditional manual detection methods.

Description

Technical Field

[0001] The present invention belongs to the technical field of optical fiber sensing and artificial intelligence interaction, and specifically relates to a wing skin damage identification method based on support vector machine and optical fiber sensing technology. Background Art

[0002] The wing skin is an important component of the aircraft structure. Flaps, ailerons and other devices are installed in the front and rear positions to improve the dynamic aerodynamics of the wing and ensure the stability, maneuverability and controllability of the aircraft. The landing gear, engine and other devices are installed on the lower surface of the wing skin to ensure the normal flight and take-off and landing of the wing, and external components are mounted. The wing structure is usually composed of skin, transverse components, longitudinal components and joints.

[0003] The skin is used to form and maintain the shape of the wing skin, and bear local aerodynamic forces as well as torque and bending moments. During the actual flight of the aircraft, the wing skin is easily impacted by various foreign objects, such as birds, hail, etc. Tiny objects can have a huge impact on the wing skin, such as direct impacts such as grooves, holes, cracks, pits, scratches, deformation, shedding, performance changes, and more likely to cause malignant indirect impacts such as casualties. Therefore, being able to detect the type and intensity of impact damage in a timely manner can provide the pilot with a good aircraft health level and make corresponding response strategies in a timely manner when necessary. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention provides a wing skin damage identification method based on support vector machine and optical fiber sensing technology, which has high efficiency, low cost, high degree of automation, and can reflect the location and degree of wing skin damage.

[0005] In order to solve the above technical problems, the present invention provides a wing skin damage identification method based on support vector machine and optical fiber sensing technology, comprising the following steps:

[0006] S1. Conduct three-dimensional modeling of the wing skin to be monitored, perform static simulation using ANSYS Workbench, and obtain the modal vibration matrix Φ of the model;

[0007] S2, calculating the effective independent coefficient E0 for the first six modal vibration shape matrices obtained in step S1, and roughly selecting the sensor layout locations by applying a decreasing function to the effective independent coefficients;

[0008] S3, calculating the modal confidence factor for the rough placement of the sensor obtained in step S2, and optimizing the accurate placement of the fiber Bragg grating by the modal confidence factor;

[0009] S4, based on step S3 and in combination with the actual monitoring model or structure, the fiber Bragg grating is applied to its surface, and a certain amount of pre-stretching effect is generated, and the strain data of the multi-grating is obtained through the demodulator and the host computer;

[0010] S5. Measure the grating strain data under normal conditions and at different locations under different degrees of impact, label them accordingly and pre-process the signals, including signal denoising, differentiation and setting the number of sampling points to 2000. Divide the data set into a training set and a validation set in a ratio of 3:1.

[0011] S6. Perform signal fusion based on the support matrix algorithm on the multi-sensor signal group. After reducing the signal from a multi-dimensional sequence to a single-dimensional sequence, extract the features of the fused signal that reflect the signal intensity, including the root mean square value, peak-to-peak value, etc., as well as feature extraction by other methods such as singular spectrum entropy, power spectrum entropy, LCD time-frequency space entropy, etc.

[0012] S7. Perform PCA feature dimensionality reduction on the extracted feature signals.

[0013] S8. Build a GA-SVM machine learning model based on Python to obtain the optimal C and gamma parameters, and set the loss function to the joint function of the soft cross entropy function and Diceloss. Input the classified features and labels into the support vector machine.

[0014] Further, in step S2, the effective independent coefficient E0 is obtained by the modal vibration matrix Φ through the following formula:

[0015] E0=diag(Φ*inv(Φ'*Φ)*Φ')

[0016] [AS,POS]=sort(E0)

[0017] Among them, diag represents singular value decomposition, Φ represents the obtained modal vibration matrix, inv represents the inverse of the matrix; sort represents the arrangement of E0 from small to large, AS represents the size of the effective independent coefficient at the corresponding position, and POS represents the corresponding position of AS.

[0018] The calculation method of the modal confidence factor used in step S2 is implemented by the following formula:

[0019]

[0020] Among them, Φ1 represents the modal vibration matrix after the decreasing function selection is applied to E0, Φ 1i Φ 1j represents the i-th or j-th order modal vector, and k represents the k-th row of the Φ1 matrix.

[0021] The signal preprocessing method of step S4 is implemented by the following formula:

[0022] x d =|x(i+1)-x(i)|(i∈[1,N-1])

[0023] S∈[index(x d >threshold)-100,index(x d >threshold)+1899]

[0024] In the formula, N represents the number of sampling points before preprocessing, S represents the range of sampling points after preprocessing, x(i) represents the signal of the i-th sampling point, x d Represents the absolute value of the difference between the previous and next signals, index(x d >threshold) indicates the first x in the signal d The index where the value is greater than the set impact threshold threshold.

[0025] The dual loss function model of Python-based GA-SVM in step S6 is implemented by the following formula:

[0026]

[0027] Focal_loss=-a t (1-p t ) γ log(p t )

[0028]

[0029]

[0030] Among them, GA_func is the genetic algorithm adaptive function, M represents the number of categories, y ic is a sign function, when c=i, y ic is 1, otherwise it is 0. ic represents the predicted probability that the observed sample i belongs to category c. Focal_loss is the first loss function of the SVM double loss function, where p t is the probability corresponding to the category, γ,a t is the weight adjustment factor, and its specific value can be reversely derived from the sample ratio. CrossEntropy_loss represents the second loss function of the SVM double loss function, N represents the number of categories, and weight(class[i]) is the weight of the category of observed sample i, which can be obtained from the sample ratio. The merging of the double loss functions is implemented in Python.

[0031] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0032] The wing skin damage identification method based on support vector machine and optical fiber sensing technology of the present invention is high in efficiency, low in cost, resistant to electromagnetic interference, quick in response, and has a high accuracy rate, and has obvious advantages over traditional manual detection methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a flow chart of the wing skin damage identification method based on support vector machine and fiber optic sensing technology;

[0034] Figure 2 It is the flow chart of sensor placement;

[0035] Figure 3 is a signal processing flow chart;

[0036] Figure 4 It is the effective independent coefficient ranking curve;

[0037] Figure 5 is the modal assurance factor curve;

[0038] Figure 6 It is the response diagram and signal fusion diagram of three fiber Bragg gratings when they are impacted at a certain position of the wing skin;

[0039] Figure 7 It is a flowchart of the dual loss function machine learning model construction and training of GA-SVM. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0041] The present invention discloses a wing skin damage identification method based on support vector machine and fiber optic sensing technology. First, a three-dimensional model of the corresponding wing skin is established by modeling software. After vibration analysis and modal analysis are performed on the three-dimensional model, the sensor point location is selected by an effective independent method and a modal factor hybrid algorithm. A plurality of ribbon fiber grating sensors are attached to the corresponding positions on the wing skin surface to monitor the vibration of the wing skin structure. A sensor data set is obtained by simulating different damages in the outside world. Then, information fusion of multiple sensors is performed by a support matrix method. After that, the fused signal is denoised, features are extracted, and feature dimension reduction is performed to divide the data into a training set and a verification set. Finally, damage identification is performed by an improved support vector machine model. The present invention greatly reduces the size and dimension of the data set through multi-sensor information fusion and feature extraction methods, improves the model accuracy and efficiency, and has better anti-electromagnetic interference and volume advantages compared to traditional electrical sensors.

[0042] When the wing skin is impacted by the outside world, the overall structure will be deformed. By engraving gratings with equal grating periods on the optical fiber, light of a specific wavelength band can be reflected. When the external environment changes, the reflection center wavelength of the grating will shift. This shift can reflect the changes in external physical quantities. Using this method to receive the center wavelength of the fiber Bragg grating, the strain of the entire structure can be received in real time. The multi-fiber Bragg grating sensor of the overall structure is subjected to signal preprocessing, data-level fusion, feature extraction and input into the support vector machine for classification processing, which can quickly and accurately reflect the current state of the wing skin and the location of damage when impacted.

[0043] The present invention mainly comprises 3 parts, such as Figure 1 As shown in the figure, they are: sensor placement, signal processing and establishment of machine learning model.

[0044] like Figure 2 As shown, the fiber grating arrangement process method for wing skin damage identification based on fiber optic sensing technology and support vector machine in the embodiment of the present invention specifically includes the following steps:

[0045] 1. Obtain the specific parameters of the structure or object to be monitored, establish a three-dimensional model, perform finite element simulation through ANSYS Workbench to obtain the modal vibration matrix Φ, obtain the effective independence coefficient E0 of the grid parameters through the effective independence method, and obtain the change weight of each grid point.

[0046] By sorting the effective independent coefficients, Figure 4 The effective independent coefficients are sorted into a curve, and a decreasing function is applied to the sorted curve. After the modal confidence factor of each grid point parameter is obtained, the arrangement points are screened and the fiber Bragg grating is arranged according to the actual situation. Figure 5 is the modal assurance factor curve.

[0047] The calculation method for obtaining E0 through Φ is as follows, where diag represents singular value decomposition, Φ represents the obtained modal vibration matrix, Φ' represents transpose, inv represents the inverse of the matrix, sort represents arranging E0 from small to large, AS represents the size of the effective independent coefficient at the corresponding position, and POS represents the corresponding position of AS.

[0048] E0=diag(Φ*inv(Φ'*Φ)*Φ')

[0049] [AS,POS]=sort(E0)

[0050] The modal confidence factor optimization method used to calculate the screening grid points is implemented by the following formula:

[0051]

[0052] Where Φ1 represents the modal vibration matrix after the decreasing function selection is applied to E0, Φ 1i Φ 1j represents the i-th or j-th order modal vector, Φ 1kj represents the j-th order modal vector of the k-th row of the Φ1 matrix, and k represents the k-th row of the Φ1 matrix.

[0053] 2. If Figure 3 As shown, a method for acquiring, preprocessing, fusion and feature extraction of fiber grating signals for wing skin damage identification based on fiber optic sensing technology and support vector machine includes the following steps:

[0054] Arrange the fiber Bragg grating to ensure the wavelength discrimination of the fiber Bragg grating, connect one end of the fiber Bragg grating to the jumper, connect the jumper to the channel of the fiber Bragg grating demodulator, and obtain the sensing data of multiple gratings through the PC. Divide the wing skin into different areas, use tools to generate impact damage signals and label them accordingly (position, impact intensity, etc.), and obtain different sampling signals through different knocking positions.

[0055] The obtained signals are preprocessed, including normalizing the number of sampling points, denoising, and fusion through the support matrix method. Figure 6 The response signal diagram of three fiber Bragg gratings and their signal fusion diagram when impact damage occurs in a certain area. The preprocessed values ​​are subjected to feature extraction and a feature group matrix is ​​generated. The last column of the matrix represents the label type.

[0056] The method of normalizing the number of signal sampling points is implemented by the following formula, where N represents the number of sampling points before preprocessing, S represents the range of sampling points after preprocessing, and x d Represents the absolute value of the difference between the previous and next signals, index(x d>threshold) represents the first x d The index where the value is greater than the threshold.

[0057] x d =|x(i+1)-x(i)|, (i∈[1,N-1])

[0058] S∈[index(x d >threshold)-100,index(x d >threshold)+1899]

[0059] The collected signal is taken with the first jump as the standard, 100 sampling points are taken forward, and 1899 sampling points are taken backward. The central wavelength offset change threshold is selected as threshold, and the unit is pm.

[0060] like Figure 7 As shown, a GA-SVM dual loss function model for wing skin damage identification based on optical fiber sensing technology and support vector machine is implemented by the following steps:

[0061] The obtained feature group is divided into a training set and a validation set at a ratio of 3:1. The training set is input into the input layer of the SVM. The last column of the training set is one-hot encoded as the output. A basic SVM model is established, and the kernel function is selected as rbf. After optimizing the model's hyperparameter penalty factor C and gamma through the genetic algorithm, the hyperparameter value is brought into the SVM model of the double loss function for training, and the trained model is input into the SVM model as a validation set. The specific formula is implemented by the following method:

[0062]

[0063] Focal_loss=-a t (1-p t ) γ log(p t )

[0064]

[0065]

[0066] Among them, GA_func is the genetic algorithm adaptive function, M represents the number of categories, y ic is a sign function, when c=i, y ic is 1, otherwise it is 0. ic is the predicted probability that the observed sample i belongs to category c. Focal_loss is the first loss function of the SVM double loss function, where p tis the probability corresponding to the category, γ,a t is the weight adjustment factor, and its specific value can be reversely derived from the sample ratio. CrossEntropy_loss represents the second loss function of the SVM double loss function, N represents the number of categories, weight(class[i]) is the weight of the category of the observed sample i, which can be obtained through the sample ratio, x represents the set of samples of this category, and class is the same as c. The SVM model is based on the Python Tensorflow framework. The joint function of the soft cross entropy function and Diceloss is used as the loss function of the experiment. The adaptive function of the genetic algorithm is selected as MSE, and the merging of the double loss functions is implemented through Python's JointLoss function.

[0067] Laboratory conditions required for training: GTX1650 graphics card, Windows system, Python programming language, Tensorflow machine learning framework.

[0068] Data set acquisition method: several fiber Bragg gratings with discrimination, 2000Hz fiber Bragg grating demodulator, demodulation host computer software, monitoring structure or model.

[0069] It will be easily understood by those skilled in the art that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A wing skin damage identification method based on support vector machine and optical fiber sensing technology, characterized in that: The method comprises the following steps: Establish a three-dimensional model of the wing skin, and perform modal analysis on the three-dimensional model of the wing skin to obtain the modal vibration matrix Φ; The effective independent coefficient E0 is calculated for the modal vibration matrix Φ, a decreasing function is applied to the effective independent coefficient E0, and the location of the fiber Bragg grating sensor is determined by the modal confidence factor; Arrange fiber grating sensors on the wing skin; Obtain the fiber Bragg grating sensing signals of the wing skin in normal condition and when different positions are subjected to different degrees of impact, and perform signal fusion of multiple fiber Bragg grating sensing signals using the support matrix method; Extract fusion signal features and construct training sets; Establish a support vector machine model, and perform wing skin damage identification after model training is completed; Among them, the calculation method of effective independence coefficient E0 is as follows: E0=diag(Φ*inv(Φ'*Φ)*Φ') [AS,POS]=sort(E0) In the formula, diag represents singular value decomposition, inv represents the inverse of the matrix, Φ' represents the transpose of the modal vibration matrix, sort represents arranging E0 from small to large, POS represents the corresponding position of AS, and AS represents the size of the effective independent coefficient at the corresponding position; The calculation method of modal assurance factor MAC is as follows: Where Φ1 represents the modal vibration matrix after the effective independent coefficient E0 is selected by the decreasing function, Φ 1i Φ 1j represents the i-th and j-th order modal vectors of the Φ1 matrix, Φ 1kj represents the j-th order modal vector of the k-th row of the Φ1 matrix, T represents the transpose, and k represents the k-th row of the Φ1 matrix.

2. The wing skin damage identification method based on support vector machine and optical fiber sensing technology according to claim 1 is characterized in that: ANSYS Workbench software is used to perform modal analysis on the 3D model of the wing skin.

3. The wing skin damage identification method based on support vector machine and optical fiber sensing technology according to claim 1 is characterized in that: The arrangement of fiber grating sensors on the wing skin includes: According to the actual situation of the wing skin, the fiber Bragg grating is arranged on the wing skin to produce a certain amount of pre-stretching effect, and then the optical fiber is fused with the jumper, and the jumper is connected to the fiber Bragg grating demodulator. The demodulator sends the fiber Bragg grating sensor signal to the PC via Ethernet.

4. The wing skin damage identification method based on support vector machine and optical fiber sensing technology according to claim 1 is characterized in that: After the fiber Bragg grating sensing signal is acquired, the signal is preprocessed; the preprocessing includes normalizing the number of sampling points, differencing and denoising.

5. The wing skin damage identification method based on support vector machine and optical fiber sensing technology according to claim 4 is characterized in that: The preprocessing is as follows: x d =|x(i+1)-x(i)|(i∈[1,N-1]) S∈[index(x d >threshold)-100,index(x d >threshold)+1899] In the formula, N represents the number of sampling points before preprocessing, S represents the range of sampling points after preprocessing, x(i) represents the signal of the i-th sampling point, and x d Represents the absolute value of the difference between the previous and next signals, index(x d >threshold) indicates the first x in the signal d The index where the value is greater than the set impact threshold threshold.

6. The wing skin damage identification method based on support vector machine and optical fiber sensing technology according to claim 1 is characterized in that: Before constructing the training set, PCA feature dimensionality reduction is performed on the extracted features.

7. The wing skin damage identification method based on support vector machine and optical fiber sensing technology according to claim 1 or 6 is characterized in that: The extracted features include root mean square value, variance value, peak-to-peak value, pulse index, kurtosis index, singular spectrum entropy, power spectrum entropy, and LCD time-frequency space entropy.

8. The wing skin damage identification method based on support vector machine and optical fiber sensing technology according to claim 1 is characterized in that: A GA-SVM machine learning model was established, and the optimal C and gamma parameters were obtained through genetic algorithm. The loss function was set as the joint function of soft cross entropy function and Diceloss.

Images