Array excitation ultrasonic guided wave damage positioning method for curved surface component

Abstract

The invention provides an array excitation ultrasonic guided wave damage positioning method for a curved surface component, and belongs to the technical field of non-destructive testing of special-shaped components, and the method comprises the following steps: S1, preparing a damage sample and a blank sample, and respectively arranging a guided wave excitation array and a sensor; s2, respectively exciting and receiving ultrasonic signals of the damaged sample and the blank sample; s3, extracting a first arrival wave time window signal of the damaged sample, and respectively extracting a time-frequency domain parameter of the signal and a correlation coefficient rho l of the damaged sample signal and the blank sample signal; s4, establishing a mapping relation between any position P (x, y) of the curved surface component and the total imaging damage probability Pr (x, y); s5, constructing a damage distance identification and classification model; s6, the scattering delay distance Lscab of each sensor is calculated; s7, determining a possible damage area; and S8, establishing a damage positioning optimization model, and predicting a damage center. Based on the steps, the damage of the complex curved surface component can be accurately positioned.

Description

Technical Field

[0001] The invention relates to the technical field of nondestructive testing of special-shaped components, and in particular to an array-excited ultrasonic guided wave damage locating method for curved surface components. Background Art

[0002] In the field of aerospace, curved surface components are often used in the fuselage, wings, tail and other external structures of aircraft and spacecraft, as well as parts with complex internal and external curves such as fuel tanks, air inlets, and turbine engines. In the long-term use of aircraft and spacecraft, affected by cyclic loads and environmental factors, curved surface components often suffer from fatigue cracks, corrosion damage, impact damage and other types of damage, which seriously threaten the safety of equipment operation. According to the specific application requirements, detection accuracy and environmental conditions of the components, X-ray scanning, acoustic emission detection, laser ultrasound and other means are often used to monitor the damage, but these methods have high requirements for equipment and high detection costs. At the same time, some methods are sensitive to external noise and often lead to reduced detection accuracy. The ultrasonic guided wave monitoring technology based on flexible piezoelectric transducer array can identify the health status of the structure in real time, which is especially suitable for dynamic monitoring and online detection. Therefore, for the damage mode of curved surface components, the use of array-excited ultrasonic guided waves provides new ideas and solutions for the detection of complex structures, which is particularly important for ensuring the safety and reliability of the aerospace field.

[0003] Ultrasonic guided waves have the characteristics of long propagation distance, low energy attenuation, and multi-modal dispersion. They are very sensitive to defects such as cracks and corrosion. However, the propagation of ultrasonic guided waves is not only affected by material parameters, but also by factors such as geometry and boundary conditions. In practical applications, especially the detection of curved components, there are still many challenges. Curved components have complex geometric shapes, which makes guided waves prone to scattering, diffraction and other phenomena, greatly increasing the difficulty of analyzing guided wave modal parameters and signal processing in damage detection. At the same time, the use of array-excited ultrasonic guided waves is usually carried out along a specific path, and single-shot and multi-received ultrasonic signals can easily lead to incomplete detection coverage, which cannot be identified by commonly used damage location methods.

[0004] By analyzing and training the received signals of ultrasonic guided waves, defects and damage prediction of curved components can be performed, and a classification problem of whether the structure is in a damaged state or a healthy state can be formed. However, in damage detection in the aerospace field, the vast majority of aerospace structures are in normal operation without obvious damage. This results in the number of healthy state samples being much larger than the number of damaged state samples in the non-destructive testing data set. This will cause the classifier to learn the features of a large number of healthy samples during the training process, thereby ignoring the subtle differences in damaged samples, causing damaged samples to be misjudged as normal, reducing the performance evaluation of the classifier and affecting subsequent damage location. Therefore, in order to solve the above problems, it is very urgent and important to develop an array-excited ultrasonic guided wave damage location method for curved components. Summary of the invention

[0005] In order to solve the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide an array-excited ultrasonic guided wave damage location method for curved surface components, which can use a flexible piezoelectric transducer to excite ultrasonic guided waves to identify and classify damage at different distances, thereby reducing the influence of the geometric complexity of curved surface components and experimental errors on damage detection; by artificially synthesizing damage category samples, the distribution of the number of samples of healthy categories and damaged categories in the data set is balanced; using a classifier to train and learn the training set after expanding the synthetic samples, the accuracy of damage identification can be improved; by taking the cross-correlation function of the damage scattering signal, a more robust damage location prediction can be achieved, and the accurate location of the damage of the curved surface component can be achieved through the combination of the above methods.

[0006] Specifically, the present invention provides an array-excited ultrasonic guided wave damage locating method for curved surface components, which comprises the following steps:

[0007] S1. Prepare damaged curved surface component specimens with pre-damaged damage at different positions and blank specimens and arrange guided wave excitation arrays and sensors respectively;

[0008] S2. Exciting and receiving ultrasonic signals for damaged samples and blank samples;

[0009] S3, extract the time window signal of the first arrival wave of the damaged sample, and extract the time-frequency domain parameters of the signal and the correlation coefficient ρ between the damaged sample signal and the blank sample signal respectively l ;

[0010] S4, establishing a mapping relationship between any position P(x, y) of the curved surface component and the total imaging damage probability Pr(x, y);

[0011] S5. Construct a damage distance identification and classification model;

[0012] S6. Extract the correlation coefficient of the sensor scattering signals and calculate the scattering delay distance L of each sensor sca_b ;

[0013] S7. Determine the possible damage area: Draw a circle with the sensor as the center and the scattering delay distance as the radius, and extract the arc according to the distribution of the possible damage area:

[0014] S71, respectively, each sensor S b (x b ,y b ) is the center of the circle, and the scattering delay distance is L sca_b Draw a circle with radius O b ;

[0015] S72, in each circle O b Extract arc O b (θ bγα ,θ bγβ ), where θ bγα ,θ bγβ are the starting angle and ending angle of the arc respectively. The area enclosed by the center of the arc and the two radial rays passes through Ω γ The vertex, Ω γ The total imaging damage probability of the curved component is greater than the probability threshold P γ ×[Pr(x,y)] max The area surrounded by the points, γ is Ω γ area;

[0016] S73, for the changing area Ω γ , respectively calculate the arc O b (θ bγα ,θ bγβ )’s weighted starting angle and weighted ending angle θ bα ,θ bβ :

[0017]

[0018] Where ω(γ) is the region Ω γ The total imaging damage probability corresponding to the contour points, γ b is the median of the total imaging damage probability greater than the maximum value in the curved component The area of ​​the region enclosed by the points;

[0019] S8. Using the distance range from the damage position to each arc as the optimization parameter, a damage location optimization model is established to predict the damage center:

[0020] S81. Take any position P(x,y) of the curved surface component and calculate the angle θ between the position and the positive x-axis. p ,like Then discard the point and pick a new point;

[0021] S82, when θ p ∈(θ bα ,θ bβ ), calculate P(x,y) to arc O b (θ bα ,θ bβ ) on the distance d b :

[0022]

[0023] Among them, d b (x,y) is d b Functions about the x-axis and y-axis;

[0024] S83. Establish a damage location optimization model and predict the damage center (x pre ,y pre ):

[0025]

[0026] in, For sensor Medium b The maximum value of the function about the x-axis and y-axis, For sensor Medium b The minimum value of a function about the x-axis and the y-axis.

[0027] Preferably, step S6 specifically includes the following sub-steps:

[0028] S61, the correlation coefficient ρ in the wave propagation path l The sensor on the lowest path is defined as the high probability damage sensor S m (x m ,y m ), calculate the high probability damage scattering signal s' m and the normal scattered signal s' b ;

[0029] S62, calculate different offset times T sca Under these conditions, the high probability scattered signal s' m With the normal scattered signal s' b The cross-correlation function

[0030] S63, defining the offset time that makes the cross-correlation function reach the maximum value as the scattering delay time T sca_b , and calculate the scattering delay distance L sca_b :

[0031] L sca_b =T sca_b ×v;

[0032] Where v is the average velocity of guided wave propagation;

[0033] S64, repeat steps S61 to S63 to obtain the scattering delay distance L of each sensor sca_b .

[0034] Preferably, step S1 specifically includes the following steps:

[0035] S11, the array elements of the waveguide excitation array arranged on the curved surface component are flexible piezoelectric transducers that fit the curved surface, and the number of array elements is greater than 3;

[0036] S12, arranging a total of N sensors on the curved component sample, wherein sensors are arranged on n1 groups of propagation distances, n2 sensors are equidistantly arranged on the same propagation distance, and data samples collected by sensors of the same propagation distance have the same distance number J;

[0037] S13, the excitation ultrasonic signal is a single center frequency sinusoidal pulse signal modulated by a Hanning window. The ultrasonic signal frequency f is determined according to the material and thickness of the curved component, and the number of cycles is N f ;

[0038] S14. Reorganize the three-dimensional coordinates on the curved component into two-dimensional coordinates, wherein the horizontal axis coordinate represents the distance from the surface of the curved component to the origin in the width direction, and the vertical axis coordinate represents the distance from the surface of the curved component to the origin in the length direction.

[0039] Preferably, step S2 specifically includes the following sub-steps:

[0040] S21, testing the sample by means of simultaneous array excitation, the excitation array generates an excitation ultrasonic signal in the sample, and the sensor receives the ultrasonic signal transmitted in the sample;

[0041] S22, for different damaged samples and blank samples, the data samples collected by the sensor have different prefabricated damage numbers I;

[0042] S23. Repeated experiments are performed on single damaged samples or blank samples. The data samples collected by the sensor in different repeated experiments have different repeated experiment numbers K.

[0043] Preferably, step S3 specifically includes the following sub-steps:

[0044] S31, for any sensor position S b (x b ,y b ), extract the first arrival wave signal time window [T start ,T end ]:

[0045]

[0046] Among them, (x cen ,y cen ) is the center coordinate of the excitation array, v is the average velocity of guided wave propagation;

[0047] S32, extracting the ultrasonic signal s of the damaged sample received by the sensor b In the time window [T start ,T end ], and calculate the signal amplitude in the time and frequency domain parameters of the signal including maximum value, peak difference, standard deviation, kurtosis, skewness, energy, center of gravity frequency and frequency standard deviation;

[0048] S33, calculate the ultrasonic signal s of the damaged sample received by the sensor b The segments in the time window are compared with the ultrasonic signals s collected in the corresponding blank sample. bt The correlation coefficient ρ l :

[0049]

[0050] Among them, s b (τ),s bt (τ) are the sensor S in the damaged sample and the blank sample respectively. b The ultrasonic signal amplitude at time τ is, are the sensor S in the damaged sample and the blank sample respectively. b The mean amplitude of the ultrasound signal in the time window.

[0051] Preferably, step S4 specifically includes the following sub-steps:

[0052] S41, for each element position A of the excitation array a (x a ,y a ), a=1,2,3,…a max , array element and sensor S b (x b ,y b ) is:

[0053]

[0054] Select the sensor S in the array b The three array elements with the smallest distance The positions of the three array elements are averaged to form the excitation source position of the propagation path.

[0055]

[0056] S42, the excitation source With sensor S b The wave propagation path between is defined as path l, and the deviation parameter D of any position P of the curved component on path l is l (x,y):

[0057]

[0058] S43, define the damage cutoff parameter R at any position P of the curved component on the path l i (x,y):

[0059]

[0060] Among them, β is the cutoff threshold coefficient;

[0061] S44. For path l, the imaging damage probability of any position P of the curved surface component is expressed as:

[0062]

[0063] Among them, ρ l is the correlation coefficient;

[0064] S45. In the total N wave propagation paths, the damage probability of a single path is superimposed, and the total imaging damage probability of any position P of the curved surface component is expressed as:

[0065]

[0066] S46. Repeat steps S41 to S45 to obtain the total imaging damage probability of all positions of the curved surface component.

[0067] Preferably, step S5 specifically includes the following sub-steps:

[0068] S51, establishing a sensor array data sample with the same distance and retaining damage information:

[0069] S511. Establishing time-frequency domain parameter data set Where n is the number of data sets, the data is standardized so that the mean of each feature is 0 and the variance is 1:

[0070]

[0071] Among them, X ij is one of the elements of the data set X, i represents the data set number, μ j , σ j are the mean and standard deviation of the jth feature respectively;

[0072] S512, Element X' ijTogether they form the standardized time-frequency domain parameter data set X', and calculate the correlation coefficient matrix R of X' X' :

[0073]

[0074] S513, solving the correlation coefficient matrix R X' The eigenvalue λ i and the eigenvector v i , so that R X' v i =λ i v i ;

[0075] S514, select the first k principal components λ1,λ2,…,λ according to the size of the eigenvalue k , select the eigenvectors corresponding to the eigenvalues ​​to form the dimension reduction matrix V k =[v1,v2,…,v k ], and reduce the dimension of the standardized time-frequency domain parameter data set X':

[0076]

[0077] Among them, Y1,Y2,…,Y n is the time-frequency domain parameter data set after dimension reduction of the 1st, 2nd, ..., nth sensors;

[0078] S515, reorganizing the dimension-reduced time-frequency domain parameter data sets of n2 sensors arranged equidistantly at the same propagation distance into sensor array data samples at the same distance Where I, J, and K represent the prefabricated damage number, distance number, and repeated experiment number, respectively;

[0079] S52. Generate synthetic samples to balance the distribution of each category in the data set:

[0080] S53, based on training set D 1new Use the support vector machine classification algorithm to predict the test set category labels;

[0081] S54. Calculate the true positive rate TPR and false positive rate FPR based on the prediction results of the classification algorithm:

[0082]

[0083]

[0084] Among them, TP, FN, FP, and TN represent true positive examples, false negative examples, false positive examples, and true negative examples, respectively;

[0085] S55. Calculate the classification effect parameter AUC formed by the characteristic curves of the true positive rate TPR and the false positive rate FPR and the coordinate axis according to the prediction results:

[0086]

[0087] Among them, TPR ii , FPR ii is the point on the characteristic curve representing the ii-th TPR and FPR, nn is the sample size of the test set;

[0088] S56, repeat steps S52 to S55 20-50 times to obtain the average value of the classification effect parameter AUC

[0089] S57: Repeat step S56 for different distance numbers J to obtain the average classification effect parameters of different distances.

[0090] S58. According to the prediction result of the classifier, the actual imaging damage probability of the position beyond the damage classification prediction distance range is 0, and the actual imaging damage probability Pr of any position P of the curved surface component is real (x,y) is represented as:

[0091]

[0092] in, is the distance from position P to the excitation source, L α ,L β are the lower and upper limits of the damage classification prediction distance range, respectively.

[0093] Preferably, in step S61, the high probability sensor S is damaged. m (x m ,y m ) Ultrasonic signal s collected in the damaged sample m The ultrasonic signal s collected from the corresponding blank sample mt Subtract and get the high probability damage scattering signal s' m =s m -s mt ;

[0094] For any other sensor S b , take the same steps to get the normal scattered signal s' b , b≠m;

[0095] In step S62, the cross-correlation function It is expressed as:

[0096]

[0097] Among them, Ttot is the total duration of the scattered signal, T1 is the unit time step, T start is the starting point of the first arrival wave time window, and τ is the time parameter.

[0098] Preferably, step S52 specifically includes the following sub-steps:

[0099] S521, define the original data set as D = {(Y' 111 ,Z 111 ),(Y' 112 ,Z 112 ),…,(Y' IJK ,Z IJK )}, where Z IJK ∈{t,1} is the category label of a single sensor array data sample at the same distance, the damaged category is 1, and the healthy category is 0. The judgment rules are as follows: the data set sample collected in the blank sample has a category label of 0; the data set sample collected by the first row of sensors in the damaged sample that is greater than the prefabricated damage distance has a category label of 1; the category labels of other samples are 0;

[0100] S522. For the same distance number J, the original data set is split into a training set D1 and a test set D2, the number of samples contained is 9:1, and the test set contains samples in both the injury category and the health category;

[0101] S523. Select a sample set D belonging to the damage category from the training set min , from the set D min Select any damage category sample Y' IJK ∈D min ;

[0102] S524. Calculate the Euclidean distance between the sample and other damage category samples:

[0103]

[0104] in, Represents other damage category samples, sorted from small to large according to the Euclidean distance, and selects k for each damage category sample n The sample with the smallest distance It is called the nearest neighbor sample;

[0105] S525, select two samples Y' at random from the nearest neighbor samples IJKA ,Y' IJKB , combined with the original sample Y' IJK Generate new synthetic samples, A,B∈[1,2,…,k n], the specific steps are: introduce random real number factors c1, c2, c3∈[t, 1], and construct a new sample Y' according to the following formula new :

[0106]

[0107] S526, generate the synthetic sample Y' new and its label Z new =1 is added to the training set;

[0108] S527, repeat steps S523 to S526 until the number of samples in the injury category is the same as that in the healthy category, and record the training set after the expanded synthetic samples as D 1new .

[0109] Preferably, in step S72, P γ is a threshold in the interval [0.5,1] and satisfies P t =1, In step S82, the optimal solution of the damage location optimization model is calculated by the simplex method.

[0110] Compared with the prior art, the present invention has the following beneficial effects:

[0111] (1) The array-excited ultrasonic guided wave damage localization method for curved surface components proposed in the present invention establishes a mapping relationship between the position of the curved surface component and the total imaging damage probability by constructing a virtual excitation source. Compared with the commonly used damage identification algorithm, this method effectively solves the problem of unclear array-excited ultrasonic guided wave transmission and reception paths.

[0112] (2) The array-excited ultrasonic guided wave damage location method for curved components proposed in the present invention achieves the category balance of the training set by expanding the damage data samples through sampling. By transforming the samples, the classifier can extract classification features from sufficient damage state samples during the training process, avoiding overfitting to the healthy state and effectively improving the classification effect parameter AUC.

[0113] (3) The array-excited ultrasonic guided wave damage location method for curved components proposed in the present invention obtains the delay time of the scattered signals received by each sensor by extracting the cross-correlation function of the scattered signals, and further realizes accurate damage location prediction, avoiding the multi-path large sample data processing in traditional damage location methods, and effectively reducing the time and resources required for sample calculation.

[0114] (4) The present invention saves a lot of calculations by more accurately defining the possible damage area and determining the damage center position based on the possible damage area. It can accurately locate the damage of complex curved surface components, thereby solving the problem in the prior art that damage to complex curved surface components cannot be quickly and accurately located. BRIEF DESCRIPTION OF THE DRAWINGS

[0115] Figure 1 It is a flow chart of the array-excited ultrasonic guided wave damage location method for curved surface components of the present invention;

[0116] Figure 2 A schematic diagram of a curved surface component excitation array and a signal receiving sensor in an embodiment of the present invention;

[0117] Figure 3 Schematic diagram of actual imaging damage probability in an embodiment of the present invention;

[0118] Figure 4 It is a schematic diagram of the actual imaging damage probability after classification and identification in an embodiment of the present invention;

[0119] Figure 5 The comparison between the classification effect after oversampling to expand the damaged data sample in the embodiment of the present invention and the classification effect without using this method;

[0120] Figure 6 It is a schematic diagram of damage position prediction in an embodiment of the present invention. DETAILED DESCRIPTION

[0121] Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[0122] The embodiment of the present invention takes the rotating blade of an aircraft engine as an example, and describes in detail the array-excited ultrasonic guided wave damage classification method for curved surface components provided by the present invention. Figure 1 As shown, a damaged sample and a blank sample with prefabricated damage are prepared, and a guided wave excitation array and a sensor are arranged on the damaged sample and the blank sample respectively, and the mode and parameters of the excitation signal are determined to build an ultrasonic guided wave damage detection system for curved surface components; ultrasonic signals are excited and received for curved surface component samples and blank samples with prefabricated damage at different positions, and the blank sample is used for comparison; the first arrival wave time window is extracted, and the time-frequency domain parameters of the time window signal and its correlation coefficient with the corresponding blank sample signal are extracted respectively; a mapping relationship between the position of the curved surface component and the probability of imaging damage based on the correlation coefficient is established; based on the time-frequency domain parameters of repeated experiments, the principal components of the parameters are extracted and a damage distance identification and classification model is constructed; the mutual correlation coefficient of the sensor scattering signal is extracted to calculate the scattering delay distance of the damage; a circle is drawn with the sensor as the center and the scattering delay distance as the radius, and the arc is extracted according to the possible damage area distribution; the distance range from the damage position to each arc is used as the optimization parameter to establish a damage location optimization model. The specific steps are as follows:

[0123] S1. Arrange guided wave excitation arrays and sensors on damaged samples and blank samples respectively, determine the mode and parameters of the excitation signal, and build an ultrasonic guided wave damage detection system for curved components. The specific steps are as follows:

[0124] S11, prefabricate damage at different positions on the curved surface component to obtain damaged samples, and set blank samples for comparison. Arrange the guided wave excitation array on the damaged sample and the blank sample respectively, the array elements of the guided wave excitation array are flexible piezoelectric transducers that completely fit the curved surface, the number of array elements is greater than 3, and is set to 5 in this embodiment.

[0125] S12, such as Figure 2 As shown, 21 sensors are arranged on the damaged sample and the blank sample respectively, wherein the sensors are arranged at 3 groups of propagation distances, which are 10 cm, 15 cm, and 20 cm respectively. Seven sensors are equidistantly arranged at the same propagation distance, and the data samples collected by the sensors at the same propagation distance have the same distance number J, J=1, 2, 3.

[0126] S13. The excitation ultrasonic signal is a single center frequency sinusoidal pulse signal modulated by a Hanning window. The ultrasonic signal frequency is determined to be 320 kHz and the number of cycles is 5 cycles based on the material and thickness information of the curved component.

[0127] S14. Reorganize the three-dimensional coordinates on the curved component into two-dimensional coordinates, wherein the horizontal axis coordinate represents the distance from the surface of the curved component to the origin in the width direction, and the vertical axis coordinate represents the distance from the surface of the curved component to the origin in the length direction.

[0128] S2. Excite and receive ultrasonic signals for the damaged sample and the blank sample respectively, specifically:

[0129] S21. The sample is tested by simultaneously exciting the array, the exciting array generates an excitation ultrasonic signal in the sample, and the sensor receives the ultrasonic signal transmitted in the sample.

[0130] S22, such as Figure 2 As shown in the figure, in case 1, a blank sample without prefabricated damage, the array excites ultrasonic guided waves and uses sensors to receive guided wave signals; in three different damaged samples, Teflon tape with a radius of 1.6 cm is pasted at 7.2 cm (case 2), 12.1 cm (case 3), and 17 cm (case 4) from the excitation center to simulate damage. For different damaged samples and blank samples, the data samples collected by the sensor have different prefabricated damage numbers I, I = 1, 2, 3, 4.

[0131] S23. To eliminate experimental errors, a single damaged sample or blank sample is subjected to 10 repeated experiments. The data samples collected by the sensor in different repeated experiments have different repeated experiment numbers K, where K = 1, 2, ..., 1t.

[0132] S3, extract the time window of the first arrival wave of the damaged sample, and respectively extract the time-frequency domain parameters of the time window signal and the correlation coefficient between the damaged sample signal and the corresponding blank sample signal, specifically:

[0133] S31, for any sensor position S b (x b ,y b ), extract the first arrival wave time window [T start ,T end ]:

[0134]

[0135] Among them, (x cen ,y cen ) = [9.2, 25.4] are the coordinates of the center of the excitation array, and v is the average velocity of the guided wave propagation, which is determined by the geometric characteristics of the path from the sensor location to the center of the array.

[0136] S32, extracting the ultrasonic signal s received by the sensor b In the time window [T start ,T end ] and calculate the signal amplitude in the time-frequency domain parameters of the signal. The time-frequency domain parameters include: maximum value, peak difference, standard deviation, kurtosis, skewness, energy, center of gravity frequency and frequency standard deviation.

[0137] S33, calculating the ultrasonic signal s received by the sensor b The segments in the time window are compared with the ultrasonic signals s collected in the corresponding blank sample. bt The correlation coefficient ρ l :

[0138]

[0139] Among them, s b (τ),s bt (τ) are the sensor S in the damaged sample and the blank sample respectively. b The amplitude of the ultrasonic signal at time τ is: are the sensor S in the damaged sample and the blank sample respectively. b The mean amplitude of the ultrasound signal in the time window.

[0140] S4. Establish a mapping relationship between any position P(x, y) of the curved surface component and the total imaging damage probability Pr(x, y) based on the correlation coefficient, specifically:

[0141] S41, for each element position A of the excitation array a (x a ,y a )(a=1,2,3,…5), which is connected to the sensor Sb (x b ,y b ) is:

[0142]

[0143] Select the array with S b The three array elements with the smallest distance The average of their positions constitutes the excitation source position of the propagation path

[0144]

[0145] S42, the excitation source With sensor S b The wave propagation path between them is defined as path l, and the deviation parameter D of any position P of the curved surface component on path l is l (x,y):

[0146]

[0147] S43, define the damage cutoff parameter R of any position P of the curved component on the path l (l = 1, 2, ..., 21) i (x,y):

[0148]

[0149] Where β is the cutoff threshold coefficient, in this example β = 1.t5.

[0150] S44. For path l, the imaging damage probability of any position P of the curved surface component can be expressed as:

[0151]

[0152] Among them, ρ l is the correlation coefficient.

[0153] S45. In the total of 21 wave propagation paths, the damage probability of a single path is superimposed, and the total imaging damage probability of any position P of the curved surface component can be expressed as:

[0154]

[0155] S46. Repeat steps S41 to S45, and obtain the total imaging damage probability of all positions of the curved surface component. Figure 3 The figure shows the total imaging damage probability of the curved surface component in the damaged samples Case 2, Case 3, and Case 4. The bright spots in the figure are where the probability is higher, indicating that the damage probability is higher here.

[0156] S5. Based on the time-frequency domain parameters of repeated experiments, extract the principal components of the parameters and construct a damage distance identification and classification model. The specific steps are as follows:

[0157] S51, establishing a sensor array data sample with the same distance and retaining damage information:

[0158] S511. Establishing time-frequency domain parameter data set Where n is the number of data sets. The data is standardized according to the following formula so that the mean of each feature is 0 and the variance is 1:

[0159]

[0160] Among them, X ij is one of the elements of the data set X, i represents the data set number, μ j , σ j are the mean and standard deviation of the j-th feature, respectively.

[0161] S512, Element X' ij Together they form the standardized time-frequency domain parameter data set X', and calculate the correlation coefficient matrix R of X' X' :

[0162]

[0163] S513, solving the correlation coefficient matrix R X' The eigenvalue λ i and the eigenvector v i , so that:

[0164] R X' v i =λ i v i .

[0165] S514, select the first three principal components λ1, λ2, λ3 according to the size of the eigenvalue, select the eigenvectors corresponding to the eigenvalues ​​to form a dimensionality reduction matrix V3 = [v1, v2, v3], and reduce the dimension of the standardized time-frequency domain parameter data set X':

[0166]

[0167] Among them, Y1,Y2,…,Y n is the time-frequency domain parameter data set after dimension reduction of the 1st, 2nd,…,nth sensors.

[0168] S515, reorganize the reduced dimension time-frequency domain parameter data sets of the seven sensors arranged equidistantly at the same propagation distance into the same-distance sensor array data samples Y' IJK=[Y1,Y2,…Y7], where I,J,K represent the prefabricated damage number, distance number, and repeated experiment number respectively.

[0169] S52. Generate synthetic samples to balance the distribution of each category in the data set:

[0170] S521. Define the original data set as:

[0171] D={(Y' 111 ,Z 111 ),(Y' 112 ,Z 112 ),…,(Y' IJK ,Z IJK )},

[0172] Among them, Z IJK ∈{t,1} is the category label of a single sensor array data sample at the same distance, the damaged category is 1, and the healthy category is 0. The judgment rules are as follows: the data set samples collected in the blank sample have a category label of 0; the data set samples collected by the first row of sensors in the damaged sample that are greater than the prefabricated damage distance have a category label of 1; the category labels of other samples are 0. At this time, the 10cm, 15cm, and 20cm data sets each contain 30 healthy category samples and 10 damaged category samples.

[0173] S522. For the same distance number J, the original data set is split into a training set D1 containing 36 samples and a test set D2 containing 4 samples, and the test set has samples from 1 injury category and 3 healthy categories.

[0174] S523. Select a sample set D belonging to the damage category from the training set min From the set D min , select any damage category sample Y' IJK ∈D min .

[0175] S524. Calculate the Euclidean distance between the sample and other damage category samples:

[0176]

[0177] Other injury category samples are assumed to be Sort by Euclidean distance from small to large, and select the 5 samples with the smallest distance for each damage category sample It is called the nearest neighbor sample.

[0178] S525, select two samples Y' at random from the nearest neighbor samples IJKA ,Y' IJKB, A, B∈[1,2,…,5], combined with the original sample Y' IJK Generate new synthetic samples. The specific steps are: introduce random real number factors c1, c2, c3∈[t, 1], and construct a new sample Y' according to the following formula new :

[0179]

[0180] S526, generate the synthetic sample Y' new and its label Z new = 1 is added to the training set.

[0181] S527, repeat steps S523 to S526 until the number of injury category samples and healthy category samples reaches 27, and the training set after the expanded synthetic samples is recorded as D 1new .

[0182] S53, based on training set D 1new Use the support vector machine classification algorithm to predict the class labels of the test set.

[0183] S54. Calculate the true positive rate TPR and false positive rate FPR based on the prediction results of the classification algorithm:

[0184]

[0185] Among them, TP, FN, FP, and TN respectively represent true positive examples, i.e., the number of times the algorithm predicts damaged samples as damaged samples, false negative examples, i.e., the number of times the algorithm predicts healthy samples as damaged samples, false positive examples, i.e., the number of times the algorithm predicts damaged samples as healthy samples, and true negative examples, i.e., the number of times the algorithm predicts healthy samples as healthy samples.

[0186] S55. Calculate the classification effect parameter AUC formed by the characteristic curves of the true positive rate TPR and the false positive rate FPR and the coordinate axis according to the prediction results.

[0187]

[0188] Among them, TPR ii , FPR ii is the point on the characteristic curve representing the ii-th TPR and FPR, and nn is the sample size of the test set.

[0189] S56, repeat steps S52 to S55 20 times to obtain the average value of the classification effect parameter AUC

[0190] S57: Repeat step S56 for different distance numbers J to obtain the average classification effect parameters of different distances.

[0191] like Figure 5 Shown are the average classification effect parameters of the classifiers in Case 2, Case 3, and Case 4 after training with and without the oversampling algorithm. By comparing the values, we can see that the use of oversampling algorithm can significantly improve the accuracy of the classifier in identifying damage.

[0192] S58. According to the prediction result of the classifier, the actual imaging damage probability of the position beyond the damage classification prediction distance range is 0. The actual imaging damage probability Pr of any position P of the curved surface component is real (x,y) can be expressed as:

[0193]

[0194] in, is the distance from position P to the excitation source, L α ,L β are the lower and upper bounds of the damage classification prediction distance range, respectively. Figure 4 As shown in the figure, after classification and identification, the high probability area of ​​damage is further reduced, which helps to identify damage more accurately.

[0195] S6. Extract the correlation coefficient of the sensor scattering signals and calculate the scattering delay distance of each sensor, specifically:

[0196] S61, the correlation coefficient ρ in the wave propagation path l l The sensor on the lowest path is defined as a high probability sensor of damage. In this example, the high probability sensors of damage in Case 2, Case 3, and Case 4 are S2(5.64,16.19), S 1t (7.37,1t.54), S 19 (1t.94,5.42). The ultrasonic signal s collected in the damaged sample m The ultrasonic signal s collected from the corresponding blank sample mt Subtract and get the high probability damage scattering signal s' m =s m -s mt ; For any other sensor S b (b≠m) Take the same steps to get the scattered signal s' b .

[0197] S62, calculate different offset times T sca Under these conditions, the high probability scattered signal s' m and the scattered signal s' b The cross-correlation function

[0198]

[0199] Among them, T tot is the total duration of the scattered signal, T1 is the unit time step, T start is the starting point of the first arrival wave time window, and τ is the time parameter.

[0200] S63, defining the offset time that makes the cross-correlation function reach the maximum value as the scattering delay time T sca_b , and calculate the scattering delay distance L sca_b :

[0201] L sca_b =T sca_b ×v.

[0202] S64, repeat steps S61 to S63 to obtain the scattering delay distance L of each sensor sca_b ; In this example, the scattering delay distance of each sensor in Case 2 is: L sca_1 =3.9cm, L sca_3 =1.8cm, L sca_4 =2.5cm, L sca_5 =4.3cm, L sca_6 =6.8cm, L sca_7 =7.9cm; the scattering delay distance of each sensor in Case 3 is: L sca_8 =1.5cm, L sca_9 =t.7cm, L sca_11 =t.8cm, L sca_12 =1.5cm, L sca_13 =1.9cm, L sca_14 =4.6cm; the scattering delay distance of each sensor in Case 4 is: L sca_15 =3.6cm, L sca_16 =1.6cm, L sca_17 =2.6cm, L sca_18 =1.3cm, L sca_2t =t.t1cm,L sca_21 =t.1cm.

[0203] S7, draw a circle with the sensor as the center and the scattering delay distance as the radius, and extract the arc according to the possible damage area distribution. The specific sub-steps are:

[0204] S71, respectively, each sensor S b (x b ,y b ) is the center of the circle, and the scattering delay distance is L sca_b Draw a circle with radius O b .

[0205] S72, in each circle O bExtract arc O b (θ bγα ,θ bγβ ), where θ bγα ,θ bγβ are the starting angle and ending angle of the arc, respectively, and 0 degrees is the positive coordinate of the x-axis. The area enclosed by the center of the arc and the two radial rays passes through Ω γ The vertex, Ω γ The total imaging damage probability of the curved component is greater than the probability threshold P γ ×[Pr(x,y] max The area surrounded by the points, γ is Ω γ The area of ​​​​P γ is a threshold in the interval [0.5,1] and satisfies P t =1, [Pr(x,y)] max is the maximum value of the total imaging damage probability.

[0206] S73, P γ Constant changes lead to Ω γ Constantly changing, for a constantly changing areaΩ γ , respectively calculate the arc O b (θ bγα ,θ bγβ )’s weighted starting angle and weighted ending angle θ bα ,θ bβ :

[0207]

[0208]

[0209] Where ω(γ) is the region Ω γ The total imaging damage probability corresponding to the contour points, γ b is the median of the total imaging damage probability greater than the maximum value in the curved component The area of ​​the region enclosed by the points.

[0210] S8. Using the distance range from the damage position to each arc as the optimization parameter, a damage location optimization model is established to predict the damage center:

[0211] S81. Take any position P(x,y) of the curved surface component and calculate the angle θ between the position and the positive x-axis. p ,like Then discard the point and pick a new one.

[0212] S82, when θ p ∈(θ bα ,θ bβ ), calculate P(x,y) to arc Ob (θ bα ,θ bβ ) on the distance d b :

[0213]

[0214] Among them, d b (x,y) is d b Function about the x-axis and y-axis.

[0215] S83. Establish a damage location optimization model and predict the damage center (x pre ,y pre ):

[0216]

[0217] in, For sensor Medium b The maximum value of the function about the x-axis and y-axis, For sensor Medium b The minimum value of a function about the x-axis and the y-axis.

[0218] like Figure 6 As shown in the figure, the red dot is the weighted damage center predicted by the optimization model, and the area surrounded by the yellow circle is the damage predicted in the experiment. The errors between the predicted weighted damage center and the actual damage center in Case 2, Case 3, and Case 4 are 1.28 cm, 0.88 cm, and 0.66 cm, respectively. It can be seen that the predicted damage centers all fall within the actual prefabrication damage range, indicating that this method achieves accurate damage positioning and can be used for accurate damage positioning of curved components.

[0219] The embodiments described above are only descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for locating damage of curved surface components by using array-excited ultrasonic guided waves, characterized in that: It includes the following steps: S1. Prepare damaged curved surface component specimens with pre-damaged damage at different positions and blank specimens and arrange guided wave excitation arrays and sensors respectively; S2. Exciting and receiving ultrasonic signals for damaged samples and blank samples; S3, extract the time window signal of the first arrival wave of the damaged sample, and extract the time-frequency domain parameters of the signal and the correlation coefficient ρ between the damaged sample signal and the blank sample signal respectively l ; S4, establishing a mapping relationship between any position P(x, y) of the curved surface component and the total imaging damage probability Pr(x, y); S5. Construct a damage distance identification and classification model; S6. Extract the correlation coefficient of the sensor scattering signals and calculate the scattering delay distance L of each sensor sca_b ; S7. Determine the possible damage area: Draw a circle with the sensor as the center and the scattering delay distance as the radius, and extract the arc according to the distribution of the possible damage area: S71, respectively, each sensor S b (x b ,y b ) is the center of the circle, and the scattering delay distance is L sca_b Draw a circle with radius O b ; S72, in each circle O b Extract arc O b (θ bγα ,θ bγβ ), where θ bγα ,θ bγβ are the starting angle and ending angle of the arc respectively. The area enclosed by the center of the arc and the two radial rays passes through Ω γ The vertex, Ω γ The total imaging damage probability of the curved component is greater than the probability threshold P γ ×[Pr(x,y)] max The area surrounded by the points, γ is Ω γ area; S73, for the changing area Ω γ , respectively calculate the arc O b (θ bγα ,θ bγβ )’s weighted starting angle and weighted ending angle θ bα ,θ bβ : Where ω(γ) is the region Ω γ The total imaging damage probability corresponding to the contour point, γ b is the median of the total imaging damage probability greater than the maximum value in the curved component The area of ​​the region enclosed by the points; S8. Using the distance range from the damage position to each arc as the optimization parameter, a damage location optimization model is established to predict the damage center: S81. Take any position P(x,y) of the curved surface component and calculate the angle θ between the position and the positive x-axis. p ,like Then discard the point and pick a new point; S82, when θ p ∈(θ bα ,θ bβ ), calculate P(x,y) to arc O b (θ bα ,θ bβ ) on the distance d b : Among them, d b (x,y) is d b Functions about the x-axis and y-axis; S83. Establish a damage location optimization model and predict the damage center (x pre ,y pre ): in, For sensor Medium b The maximum value of the function about the x-axis and y-axis, For sensor Medium b The minimum value of a function about the x-axis and y-axis.

2. The array-excited ultrasonic guided wave damage location method for curved surface components according to claim 1, characterized in that: Step S6 specifically includes the following sub-steps: S61, the correlation coefficient ρ in the wave propagation path l The sensor on the lowest path is defined as the high probability damage sensor S m (x m ,y m ), calculate the high probability damage scattering signal s' m and the normal scattered signal s' b ; S62, calculate different offset times T sca Under these conditions, the high probability scattered signal s' m With the normal scattered signal s' b The cross-correlation function S63, defining the offset time that makes the cross-correlation function reach the maximum value as the scattering delay time T sca_b , and calculate the scattering delay distance L sca_b : L sca_b =T sca_b ×v; Where v is the average velocity of guided wave propagation; S64, repeat steps S61 to S63 to obtain the scattering delay distance L of each sensor sca_b .

3. The array-excited ultrasonic guided wave damage location method for curved surface components according to claim 1, characterized in that: Step S1 specifically includes the following steps: S11, the array elements of the waveguide excitation array arranged on the curved surface component are flexible piezoelectric transducers that fit the curved surface, and the number of array elements is greater than 3; S12, arranging a total of N sensors on the curved component sample, wherein sensors are arranged on n1 groups of propagation distances, n2 sensors are equidistantly arranged on the same propagation distance, and data samples collected by sensors of the same propagation distance have the same distance number J; S13, the excitation ultrasonic signal is a single center frequency sinusoidal pulse signal modulated by a Hanning window. The ultrasonic signal frequency f is determined according to the material and thickness of the curved component, and the number of cycles is N f ; S14. Reorganize the three-dimensional coordinates on the curved component into two-dimensional coordinates, wherein the horizontal axis coordinate represents the distance from the surface of the curved component to the origin in the width direction, and the vertical axis coordinate represents the distance from the surface of the curved component to the origin in the length direction.

4. The array-excited ultrasonic guided wave damage location method for curved surface components according to claim 3, characterized in that: Step S2 specifically includes the following sub-steps: S21, testing the sample by means of simultaneous array excitation, the excitation array generates an excitation ultrasonic signal in the sample, and the sensor receives the ultrasonic signal transmitted in the sample; S22, for different damaged samples and blank samples, the data samples collected by the sensor have different prefabricated damage numbers I; S23. Repeated experiments are performed on single damaged samples or blank samples. The data samples collected by the sensor in different repeated experiments have different repeated experiment numbers K.

5. The array-excited ultrasonic guided wave damage location method for curved surface components according to claim 4, characterized in that: Step S3 specifically includes the following sub-steps: S31, for any sensor position S b (x b ,y b ), extract the first arrival wave signal time window [T start ,T end ]: Among them, (x cen ,y cen ) is the center coordinate of the excitation array, v is the average velocity of guided wave propagation; S32, extracting the ultrasonic signal s of the damaged sample received by the sensor b In the time window [T start ,T end ], and calculate the signal amplitude in the time and frequency domain parameters of the signal including maximum value, peak difference, standard deviation, kurtosis, skewness, energy, center of gravity frequency and frequency standard deviation; S33, calculate the ultrasonic signal s of the damaged sample received by the sensor b The segments in the time window are compared with the ultrasonic signals s collected in the corresponding blank sample. bt The correlation coefficient ρ l : Among them, s b (τ),s bt (τ) are the sensor S in the damaged sample and the blank sample respectively. b The ultrasonic signal amplitude at time τ is, are the sensor S in the damaged sample and the blank sample respectively. b The mean amplitude of the ultrasound signal in the time window.

6. The array-excited ultrasonic guided wave damage location method for curved surface components according to claim 5, characterized in that: Step S4 specifically includes the following sub-steps: S41, for each array element position A of the excitation array a (x a ,y a ), a=1,2,3,…a max , array element and sensor S b (x b ,y b ) is: Select the sensor S in the array b The three array elements with the smallest distance The positions of the three array elements are averaged to form the excitation source position of the propagation path. S42, the excitation source With sensor S b The wave propagation path between is defined as path l, and the deviation parameter D of any position P of the curved component on path l is l (x,y): S43, define the damage cutoff parameter R at any position P of the curved component on the path l i (x,y): Among them, β is the cutoff threshold coefficient; S44. For path l, the imaging damage probability of any position P of the curved surface component is expressed as: Among them, ρ l is the correlation coefficient; S45. In the total N wave propagation paths, the damage probability of a single path is superimposed, and the total imaging damage probability of any position P of the curved surface component is expressed as: S46. Repeat steps S41 to S45 to obtain the total imaging damage probability of all positions of the curved surface component.

7. The array-excited ultrasonic guided wave damage location method for curved surface components according to claim 6, characterized in that: Step S5 specifically includes the following sub-steps: S51, establishing a sensor array data sample with the same distance and retaining damage information: S511. Establishing time-frequency domain parameter data set Where n is the number of data sets, the data is standardized so that the mean of each feature is 0 and the variance is 1: Among them, X ij is one of the elements of the data set X, i represents the data set number, μ j , σ j are the mean and standard deviation of the jth feature respectively; S512, Element X' ij Together they form the standardized time-frequency domain parameter data set X', and calculate the correlation coefficient matrix R of X' X' : S513, solving the correlation coefficient matrix R X' The eigenvalue λ i and the eigenvector v i , so that R X' v i =λ i v i ; S514, select the first k principal components λ1,λ2,…,λ according to the size of the eigenvalue k , select the eigenvectors corresponding to the eigenvalues ​​to form the dimension reduction matrix V k =[v1,v2,…,v k ], and reduce the dimension of the standardized time-frequency domain parameter data set X': Y=X'V k =[Y1,Y2,…,Y n ] T Among them, Y1,Y2,…,Y n is the time-frequency domain parameter data set after dimension reduction of the 1st, 2nd, ..., nth sensors; S515, reorganize the dimension-reduced time-frequency domain parameter data set of n2 sensors arranged equidistantly at the same propagation distance into the same-distance sensor array data sample Y' IJK =[Y1,Y2,…Y n2 ], where I, J, and K represent the prefabricated damage number, distance number, and repeated experiment number, respectively; S52. Generate synthetic samples to balance the distribution of each category in the data set: S53, based on training set D 1new Use the support vector machine classification algorithm to predict the test set category labels; S54. Calculate the true positive rate TPR and false positive rate FPR based on the prediction results of the classification algorithm: Among them, TP, FN, FP, and TN represent true positive examples, false negative examples, false positive examples, and true negative examples, respectively; S55. Calculate the classification effect parameter AUC formed by the characteristic curves of the true positive rate TPR and the false positive rate FPR and the coordinate axis according to the prediction results: Among them, TPR ii , FPR ii is the point on the characteristic curve representing the ii-th TPR and FPR, nn is the sample size of the test set; S56, repeat steps S52 to S55 20-50 times to obtain the average value of the classification effect parameter AUC S57: Repeat step S56 for different distance numbers J to obtain the average classification effect parameters of different distances. S58. According to the prediction result of the classifier, the actual imaging damage probability of the position beyond the damage classification prediction distance range is 0, and the actual imaging damage probability Pr of any position P of the curved surface component is real (x,y) is represented as: in, is the distance from position P to the excitation source, L α ,L β are the lower and upper limits of the damage classification prediction distance range, respectively.

8. The array-excited ultrasonic guided wave damage location method for curved surface components according to claim 2, characterized in that: In step S61, the high probability sensor S is damaged. m (x m ,y m ) Ultrasonic signal s collected in the damaged sample m The ultrasonic signal s collected from the corresponding blank sample mt Subtract and get the high probability damage scattering signal s' m =s m -s mt ; For any other sensor S b , take the same steps to get the normal scattered signal s' b , b≠m; In step S62, the cross-correlation function It is expressed as: Among them, T tot is the total duration of the scattered signal, T1 is the unit time step, T start is the starting point of the first arrival wave time window, and τ is the time parameter.

9. The array-excited ultrasonic guided wave damage location method for curved surface components according to claim 7, characterized in that: Step S52 specifically includes the following sub-steps: S521, define the original data set as D = {(Y' 111 ,Z 111 ),(Y' 112 ,Z 112 ),…,(Y' IJK ,Z IJK )}, where Z IJK ∈{t,1} is the category label of a single sensor array data sample at the same distance, the damaged category is 1, and the healthy category is 0. The judgment rules are as follows: the data set sample collected in the blank sample has a category label of 0; the data set sample collected by the first row of sensors in the damaged sample that is greater than the prefabricated damage distance has a category label of 1; the category labels of other samples are 0; S522. For the same distance number J, the original data set is split into a training set D1 and a test set D2, the number of samples contained is 9:1, and the test set contains samples in both the injury category and the health category; S523. Select a sample set D belonging to the damage category from the training set min , from the set D min Select any damage category sample Y' IJK ∈D min ; S524. Calculate the Euclidean distance between the sample and other damage category samples: in, Represents other damage category samples, sorted from small to large according to the Euclidean distance, and selects k for each damage category sample n The sample with the smallest distance It is called the nearest neighbor sample; S525, select two samples Y' at random from the nearest neighbor samples IJKA ,Y' IJKB , combined with the original sample Y' IJK Generate new synthetic samples, A,B∈[1,2,…,k n ], the specific steps are: introduce random real number factors c1, c2, c3∈[t, 1], and construct a new sample Y' according to the following formula new : S526, generate the synthetic sample Y' new and its label Z new =1 is added to the training set; S527, repeat steps S523 to S526 until the number of samples in the injury category is the same as that in the healthy category, and record the training set after the expanded synthetic samples as D 1new .

10. The array-excited ultrasonic guided wave damage location method for curved surface components according to claim 1, characterized in that: In step S72, γ is a threshold in the interval [0.5,1] and satisfies P t =1, In step S82, the optimal solution of the damage location optimization model is calculated by the simplex method.

Images