Ultrasound guided wave damage localization imaging method based on convolutional auto-encoding

Abstract

The application provides an ultrasonic guided wave damage positioning imaging method based on convolutional auto-encoding, which comprises the following steps: constructing a simulation multi-wave packet ultrasonic guided wave signal and an input data set for training of the ultrasonic guided wave according to A0 mode prior dispersion curve information of the ultrasonic guided wave and in combination with an excitation signal; constructing a label data set for training of the ultrasonic guided wave; constructing a network model of the ultrasonic guided wave damage positioning imaging method based on convolutional auto-encoding and training by means of the convolutional auto-encoding; constructing a test data set of the ultrasonic guided wave; predicting a time sequence corresponding to the test data set; and determining a distribution position of damage in a plate corresponding to the test data set. The network model of the ultrasonic guided wave damage positioning imaging method based on convolutional auto-encoding can overcome the problem of inaccurate positioning caused by guided wave dispersion, has more excellent performance in the case of inaccurate prior dispersion data, and has good generalization ability.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of nondestructive testing, and particularly relates to an ultrasonic guided wave damage positioning and imaging method based on convolution auto-encoding. BACKGROUND

[0002] Composite materials have high specific strength, corrosion resistance and design flexibility, and have wide application prospects in the aerospace industry. However, various damages such as matrix cracking, delamination and fiber fracture may occur in the composite material, which leads to the decline of mechanical properties and thus damages the reliability of the structure. Therefore, in order to avoid serious accidents, damage detection and positioning of the composite material structure are needed.

[0003] The ultrasonic guided wave detection method has the advantages of long propagation distance, large detection area and high sensitivity, and can realize damage positioning and evaluation of plate and shell structures. Therefore, ultrasonic guided waves have been widely concerned in the field of nondestructive testing and structural health monitoring. Two key characteristics of ultrasonic guided waves are dispersion and multimode, and the phase velocity and group velocity depend on the product of the excitation frequency and the structure thickness. The nonlinear relationship between wave number and frequency causes wave packet expansion and waveform distortion. The dispersed wave packet is elongated in the time domain, which is more likely to be superimposed, thus reducing the time resolution of the guided wave and greatly affecting the damage positioning and imaging results. Deep learning methods have received increasing attention in recent years, and they can extract relevant features at different abstraction levels from data. They have been preliminarily applied in nondestructive testing and structural health monitoring. However, current deep learning models based on ultrasonic guided waves mostly directly perform damage classification or coordinate estimation tasks on signals, and do not extract effective waveform features from the signals. The advantage of deep learning methods is that the model has stronger generalization ability for data, which can weaken the influence of prior dispersion information and improve the damage positioning resolution. Therefore, in order to overcome the shortcomings of the above methods, it is very urgent and necessary to seek an ultrasonic guided wave damage positioning and imaging method based on convolution auto-encoding to realize high-resolution damage positioning and imaging and solve the problem of low imaging precision on composite material plates. SUMMARY

[0004] The present application is directed to the defects in the prior art, and proposes an ultrasonic guided wave damage positioning imaging method based on convolutional auto-encoding. The method comprises: according to the A0 mode prior dispersion curve information of the ultrasonic guided wave, combining the excitation signal to simulate a multi-wave packet ultrasonic guided wave signal and constructing an input data set for training of the ultrasonic guided wave, constructing a label data set for training of the ultrasonic guided wave, constructing a network model of the ultrasonic guided wave damage positioning imaging method based on convolutional auto-encoding with the aid of convolutional auto-encoding and training, constructing a test data set of the ultrasonic guided wave, predicting the time sequence corresponding to the test data set, and determining the distribution position of the damage in the plate corresponding to the test data set. The network model of the ultrasonic guided wave damage positioning imaging method based on convolutional auto-encoding can overcome the problem of inaccurate positioning caused by guided wave dispersion, has more excellent performance in the case of inaccurate prior dispersion data, and has good generalization ability.

[0005] The present application provides an ultrasonic guided wave damage positioning imaging method based on convolutional auto-encoding, which comprises the following steps:

[0006] S1, simulate a multi-wave packet ultrasonic guided wave signal and construct an input data set for training of the ultrasonic guided wave: according to the A0 mode prior dispersion curve information of the ultrasonic guided wave, combine the excitation signal f(t) to simulate ultrasonic guided wave signals y1(t), y2(t) and y3(t) with a single wave packet, two wave packets and three wave packets respectively existing in the A0 mode under different propagation distances, and construct an input data set for training of the ultrasonic guided wave;

[0007] S2, construct a label data set for training of the ultrasonic guided wave: according to the known propagation distance x and the group velocity of the A0 mode, calculate the time of flight of each wave packet in the ultrasonic guided wave signal with a single wave packet, two wave packets and three wave packets respectively existing in the A0 mode, generate a first time sequence label corresponding to the ultrasonic guided wave signal y1(t) with a single wave packet existing in the A0 mode generate a second time sequence label corresponding to the ultrasonic guided wave signal y2(t) with two wave packets existing in the A0 mode generate a third time sequence label corresponding to the ultrasonic guided wave signal y3(t) with three wave packets existing in the A0 mode construct a label data set for training of the ultrasonic guided wave;

[0008] S3, construct a network model of the ultrasonic guided wave damage positioning imaging based on convolutional auto-encoding and train it;

[0009] S31, construct a network model of ultrasonic guided wave damage positioning imaging based on convolutional auto-encoding, the network model of ultrasonic guided wave damage positioning imaging based on convolutional auto-encoding comprises a convolution module and a deconvolution module; the convolution module is provided with a plurality of convolution modules for encoding; the deconvolution module is provided with a plurality of deconvolution modules for up-sampling decoding; the convolution module comprises a convolution layer, a batch normalization layer and an activation function ReLU function, and the deconvolution module comprises a deconvolution layer, a BN layer and a ReLU function; a dropout module is arranged in front of the deconvolution module to avoid overfitting, and a time sequence y is regenerated after feature extraction through the deconvolution module ToF ;

[0010] S32, input the training input data set in step S1 and the training label data set in step S2 as a training set into the network model of the ultrasonic guided wave damage positioning imaging method based on convolutional auto-encoding for training;

[0011] S4, construct a test data set of ultrasonic guided waves;

[0012] S41, place a piezoelectric ceramic sensor on the surface of a carbon fiber reinforced composite laminated plate, set a sensor array with a sensors, sequentially excite and collect ultrasonic guided wave signals, collect b groups of ultrasonic guided wave signals and pre-process them as reference signals;

[0013] S42, set a delamination damage on the composite plate, collect multiple groups of ultrasonic guided wave signals again, pre-process them and subtract them from the reference signals to generate b groups of residual signals containing damage reflected waves, thereby forming a test data set of ultrasonic guided waves;

[0014] S5, predict the time sequence corresponding to the test data set: input the test data set constructed in step S4 into the network model of the ultrasonic guided wave damage positioning imaging method based on convolutional auto-encoding trained in step S3, estimate the time of flight of each wave packet in the ultrasonic guided wave signal, and output a time sequence y ToF ;

[0015] S6, determine the distribution position of the damage in the test data set: input the time sequence y ToF in step S5 into a delay and stack algorithm to perform high-resolution damage positioning imaging of guided waves and obtain the distribution position of the damage in the plate.

[0016] Further, the step S1 specifically comprises the following steps:

[0017] S11, the excitation signal f(t) takes a 3-period narrow-band comb signal with a center frequency of 40 kHz, when the ultrasonic guided wave signal is in a frequency dispersion state and the propagation distance of the ultrasonic guided wave is x0, the response signal y0(t) collected by the sensor is represented as:

[0018]

[0019] wherein F(ω) represents Fourier transform of the excitation signal f(t); ω represents frequency; k represents wave number of the guided wave mode and is related to the frequency ω, i.e. k = k(ω); j represents imaginary unit; e is natural constant; ∞ represents infinity; t represents time; d represents differential;

[0020] S12, simulating the ultrasonic guided wave signal y1(t) with a single wave packet of A0 mode:

[0021]

[0022] wherein k A represents prior wave number information of A0 mode; x A1 represents propagation distance of the first wave packet of A0 mode;

[0023] S13, simulating the ultrasonic guided wave signal y2(t) with two wave packets of A0 mode:

[0024]

[0025] wherein x A2 represents propagation distance of the second wave packet of A0 mode;

[0026] S14, simulating the ultrasonic guided wave signal y3(t) with three wave packets of A0 mode:

[0027]

[0028] wherein x A3 represents propagation distance of the third wave packet of A0 mode.

[0029] Preferably, the step S2 specifically comprises the following steps:

[0030] S21, when the ultrasonic guided wave signal is in a non-dispersive state, the wave number k of the guided wave mode is a linear function of frequency k', at this time, the phase velocity c p (ω) and the group velocity c g (ω) of the A0 mode satisfy the following conditions:

[0031]

[0032] S22, when the propagation distance is x i , the non-dispersive guided wave signal y i_c (t) is represented as:

[0033]

[0034] wherein δ(·) represents impulse function; t i represents xi corresponding time instant;

[0035] S23, obtaining a first time series label corresponding to the first wave packet of the A0 mode according to the propagation distance x A1 of the first wave packet of the A0 mode

[0036]

[0037] wherein, denotes the group velocity of the A0 mode; f c denotes the center frequency of the excitation signal; denotes the propagation distance x A1 corresponding time instant;

[0038] S24, obtaining a second time series label corresponding to the second wave packet of the A0 mode according to the propagation distance x A1 of the third wave packet of the A0 mode A2 of the second wave packet of the A0 mode

[0039]

[0040] wherein, denotes the propagation distance x A2 corresponding time instant;

[0041] S25, obtaining a third time series label corresponding to the third wave packet of the A0 mode according to the propagation distance x A1 of the first wave packet of the A0 mode A2 of the third wave packet of the A0 mode A3 of the third wave packet of the A0 mode

[0042]

[0043] wherein, denotes the propagation distance x A3 corresponding time instant;

[0044] S26, taking the first time series label the second time series label and the third time series label as the label of the input data set for training of the guided wave signal.

[0045] Preferably, the pixel value of the distribution position I(x, y) of the damage in the plate in the step S6 is represented as:

[0046]

[0047] wherein, P denotes the number of excitation-receiving sensor pairs; respectively represent the distance from the excitation and receiving sensor to I(x,y).

[0048] Preferably, two convolution modules are provided in the step S3, the kernel sizes of the corresponding convolution layers are 1x10 and 1x3 respectively, and the numbers of the corresponding convolution kernels are 4 and 8 respectively; two deconvolution modules are provided, the kernel sizes of the corresponding deconvolution layers are 1x3 and 1x10 respectively, and the numbers of the corresponding convolution kernels are 8 and 4 respectively; and the ratio of the dropout module is 0.6.

[0049] Preferably, the network hyperparameters of the network model of the ultrasonic guided wave damage positioning and imaging method based on convolution self-encoding in the step S3 are set as follows: mean square error as a loss function, stochastic gradient descent as an optimizer, learning rate set to 5x10 -6 , and batch size set to 64.

[0050] Compared with the prior art, the technical effects of the present application are:

[0051] 1. The ultrasonic guided wave damage positioning and imaging method based on convolution self-encoding proposed in the present application considers that the amplitude of guided waves will rapidly attenuate when propagating on a composite material plate structure, and the signal-to-noise ratio of damage reflection signals is low. The proposed method can weaken the influence of prior dispersion information by means of deep learning method, improve damage positioning resolution by using convolution self-encoding, and realize high-resolution positioning of damage of composite materials in the case.

[0052] 2. The ultrasonic guided wave damage positioning and imaging method based on convolution self-encoding proposed in the present application can overcome the problem of inaccurate positioning caused by guided wave dispersion, has more excellent performance in the case of inaccurate prior dispersion data, and provides a new way for solving ultrasonic guided wave signal processing tasks by using deep learning technology. BRIEF DESCRIPTION OF DRAWINGS

[0053] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments made with reference to the accompanying drawings.

[0054] Figure 1 is a flow chart of the ultrasonic guided wave damage positioning and imaging method based on convolution self-encoding of the present application;

[0055] Figure 2 is a network model schematic diagram of the ultrasonic guided wave damage positioning and imaging method based on convolution self-encoding of the present application;

[0056] Figure 3 is a piezoelectric ceramic sensor arrangement schematic diagram of the present application;

[0057] Figure 4a is the residual signal and corresponding time series label map of the sensor PZT#1-PZT#3 path of the present application;

[0058] Figure 4b is the corresponding time series result of the sensor PZT#1-PZT#3 output by the network of the present application;

[0059] Figure 5a is the residual signal and corresponding time series label map of the sensor PZT#2-PZT#5 path of the present application;

[0060] Figure 5b is the corresponding time series result of the sensor PZT#2-PZT#5 output by the network of the present application;

[0061] Figure 6 is the schematic diagram of the guided wave damage high-resolution positioning imaging result of the present application. DETAILED DESCRIPTION

[0062] The present application will be further described below in conjunction with the accompanying drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the related application, and not to limit the application. In addition, it should be noted that only the parts related to the application are shown in the drawings for ease of description. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0063] Figure 1 The ultrasonic guided wave damage positioning imaging method based on convolutional auto-encoding of the present application is shown, which comprises the following steps:

[0064] S1, simulate multi-wave packet ultrasonic guided wave signals and construct training input data set of guided wave signals: according to the A0 mode prior dispersion curve information of ultrasonic guided wave, combined with the excitation signal f(t), simulate ultrasonic guided wave signals y1(t), y2(t) and y3(t) with single wave packet, two wave packets and three wave packets of A0 mode respectively at different propagation distances, and construct training input data set of guided wave signals.

[0065] S11, considering that A0 mode guided wave is more suitable for detecting delamination damage in composite plates than other modes, the excitation signal f(t) takes a 3-period narrow-band comb signal with a center frequency of 40 kHz, and the signal is dominated by A0 mode at this frequency. Neglecting the amplitude change, the nonlinear characteristics of wave number cause dispersion, wave packet broadening of response waveform, and amplitude reduction. When the ultrasonic guided wave signal is in the dispersion state and the propagation distance of the ultrasonic guided wave is x0, the response signal y0(t) collected by the sensor is represented as:

[0066]

[0067] wherein F(ω) represents the Fourier transform of the excitation signal f(t); ω represents frequency; k represents the wave number of the guided wave mode and is related to the frequency ω, i.e. k = k(ω); j represents the imaginary unit; e is the natural constant; ∞ represents infinity; t represents time; and d represents differentiation.

[0068] S12, simulate the ultrasonic guided wave signal y1(t) in which the A0 mode exists a single wave packet:

[0069]

[0070] wherein k A represents the prior wave number information of the A0 mode; x A1 represents the propagation distance of the first wave packet of the A0 mode.

[0071] S13, simulate the ultrasonic guided wave signal y2(t) in which the A0 mode exists two wave packets:

[0072]

[0073] wherein x A2 represents the propagation distance of the second wave packet of the A0 mode.

[0074] S14, simulate the ultrasonic guided wave signal y3(t) in which the A0 mode exists three wave packets:

[0075]

[0076] wherein x A3 represents the propagation distance of the third wave packet of the A0 mode.

[0077] In one specific embodiment, x A1 , x A2 , x A3 ∈ [0, 0.6], and in the single wave packet condition, the value is randomly taken 1000 times, in the two wave packet condition, the value is randomly taken 1000 times, and in the three wave packet condition, the value is randomly taken 1000 times, thereby constituting 3000 ultrasonic dispersion guided wave signal samples as the training input data set of the guided wave signal.

[0078] S2, construct the training label data set of the guided wave signal: according to the known propagation distance x and the group velocity of the A0 mode, respectively calculate the time of flight of each wave packet in the ultrasonic guided wave signal in which the A0 mode exists a single wave packet, two wave packets and three wave packets, and generate the first time sequence label generate the second time sequence label The third time sequence label corresponding to the ultrasonic guided wave signal y3(t) generated by the A0 mode existing three wave packets The training label data set of the guided wave signal is constructed.

[0079] S21, when the ultrasonic guided wave signal is in a non-dispersive state, the wave number k of the guided wave mode is a linear function of the frequency k', at this time, the phase velocity c p (ω) and the group velocity c g (ω) satisfy the following conditions:

[0080]

[0081] S22, when the propagation distance is x i The non-dispersive guided wave signal y i_c (t) is expressed as:

[0082]

[0083] Where δ(·) represents the impulse function; t i represents the x i corresponding time.

[0084] S23, according to the propagation distance x A1 of the first wave packet of the A0 mode, the first time sequence label corresponding to a single wave packet of formula (2) is obtained

[0085]

[0086] Where c represents the group velocity of the A0 mode; f c represents the center frequency of the excitation signal; represents the x A1 corresponding time.

[0087] S24, according to the propagation distance x A1 of the first wave packet of the A0 mode and the propagation distance x A2 of the third wave packet of the A0 mode, the second time sequence label corresponding to the two wave packets of formula (3) is obtained

[0088]

[0089] Where c represents the x A2 corresponding time.

[0090] S25, according to the propagation distance x A1 of the first wave packet of the A0 mode, the propagation distance x A2 of the third wave packet of the A0 mode and the propagation distance xA3 obtaining a third time sequence label corresponding to formula (4) containing three wave packets

[0091]

[0092] wherein, x A3 corresponding to the moment.

[0093] S26, the first time sequence label the second time sequence label and the third time sequence label as the label of the input data set for training of the guided wave signal.

[0094] S3, constructing a network model of ultrasonic guided wave damage positioning imaging based on convolutional auto-encoding and training.

[0095] S31, as shown in Figure 2 , a network model of ultrasonic guided wave damage positioning imaging based on convolutional auto-encoding is constructed, the network model of ultrasonic guided wave damage positioning imaging based on convolutional auto-encoding includes convolution modules and deconvolution modules; 2 convolution modules are arranged for encoding; 2 deconvolution modules are arranged for up-sampling decoding; the convolution module includes a convolution layer, a batch normalization layer and an activation function ReLU function, and the deconvolution module includes a deconvolution layer, a BN layer and a ReLU function; a dropout module is arranged in front of the deconvolution module to avoid overfitting, and the ratio is set to 0.6; after feature extraction by the deconvolution module, a time sequence y ToF is regenerated.

[0096] The kernel size of the convolution layer corresponding to the two convolution modules is 1x10 and 1x3 respectively, and the number of the convolution kernel corresponding to the two convolution modules is 4 and 8 respectively; the kernel size of the deconvolution layer corresponding to the two deconvolution modules is 1x3 and 1x10 respectively, and the number of the convolution kernel corresponding to the two deconvolution modules is 8 and 4 respectively; the ratio of the dropout module is set to 0.6, as shown in Table 1.

[0097] Table 1

[0098]

[0099] The network hyperparameters of the network model of the ultrasonic guided wave damage positioning imaging method based on convolutional auto-encoding are set as follows: mean square error as the loss function, stochastic gradient descent as the optimizer, learning rate set to 5x10 -6 , and batch size set to 64.

[0100] S32, input the training input data set in step S1 and the training label data set in step S2 as a training set into the network model of the ultrasonic guided wave damage positioning and imaging method based on convolutional auto-encoding for training.

[0101] S4, construct a test data set.

[0102] S41, place the piezoelectric ceramic sensor on the surface of the carbon fiber reinforced composite laminate, set a sensor to form a sensor array, sequentially excite and collect ultrasonic guided wave signals, collect b groups of ultrasonic guided wave signals and pre-process them as reference signals. In one specific embodiment, a is 8 and b is 28.

[0103] S42, set a delaminated damage on the composite plate, collect multiple groups of ultrasonic guided wave signals again, pre-process them and subtract them from the reference signals to generate b groups of residual signals containing damage reflected waves, thereby forming a test data set of ultrasonic guided waves.

[0104] In one specific embodiment, a carbon fiber reinforced composite laminate with a size of 700mm x 700mm x 2mm is used as the experimental specimen, and the lay-up direction is [+45 / -45 / 0 / 90] 2s As shown in Figure 3 , it is a schematic diagram of the arrangement of the piezoelectric ceramic sensor.

[0105] S5, predict the time sequence corresponding to the test data set: input the test data set constructed in step S4 into the network model of the ultrasonic guided wave damage positioning and imaging method based on convolutional auto-encoding trained in step S3, estimate the time of flight of each wave packet in the ultrasonic guided wave signal, and output the time sequence y ToF .

[0106] In one specific embodiment, the residual signals obtained by the sensors PZT#1-PZT#3, the corresponding time sequence labels are shown in Figure 4a , and the time sequence results output by the network are shown in Figure 4b ; the residual signals obtained by the sensors PZT#2-PZT#5, the corresponding time sequence labels are shown in Figure 5a , and the time sequence results output by the network are shown in Figure 5b . The first wave packet in the residual signal is the wave packet generated by the electromagnetic pulse, which is ignored here. It can be found that the pulse function output by the network can correspond to the position of the damage reflected wave, thereby improving the accuracy of subsequent damage positioning.

[0107] S6, determine the distribution position of the damage in the plate corresponding to the test data set: input the time sequence y ToF into the time delay superposition algorithm to perform high-resolution damage positioning and imaging of guided waves, and obtain the distribution position of the damage in the plate.

[0108] The pixel value of the distribution position I(x, y) of the damage in the plate is expressed as:

[0109]

[0110] Wherein, P represents the excitation-receiving sensor pair number; Respectively represent the distance from the excitation and receiving sensor to I(x, y).

[0111] The final guided wave damage high-resolution positioning imaging result is as shown in the figure Figure 6 The black circle represents the real damage position, and the black cross is the damage center position. The results show that the method can realize the accurate positioning imaging of the damage on the composite plate.

[0112] The present application designs an ultrasonic guided wave damage positioning imaging method based on convolution auto-encoding. Considering that the amplitude of the guided wave will rapidly attenuate when propagating on the composite material plate structure, and the signal-to-noise ratio of the damage reflection signal is low, the method takes advantage of the deep learning method to weaken the influence of prior dispersion information, uses convolution auto-encoding to improve the damage positioning resolution, and realizes the high-resolution positioning of the damage of the composite material under the condition; by constructing the network model of the ultrasonic guided wave damage positioning imaging method based on convolution auto-encoding, the problem of inaccurate positioning caused by guided wave dispersion can be overcome, and better performance can be achieved under the condition that the prior dispersion data is not accurate; at the same time, it provides a new way for using deep learning technology to solve the ultrasonic guided wave signal processing task.

[0113] Finally, it should be explained that: the above examples are only for illustration, not for limiting the technical solutions of the present application. Although the present application has been described in detail with reference to the above examples, those skilled in the art should understand that the present application can still be modified or replaced by equivalents without departing from the spirit and scope of the present application. Any modification or partial replacement should be covered in the scope of the claims of the present application.

Claims

1. A method for ultrasonic guided wave damage localization imaging based on convolutional autoencoders, characterized in that, It includes the following steps: S1. Simulate multi-wave packet ultrasonic guided wave signals and construct the training input dataset for ultrasonic guided waves: Based on the prior dispersion curve information of the A0 mode of ultrasonic guided waves, and combined with the excitation signal f(t), simulate ultrasonic guided wave signals y1(t), y2(t), and y3(t) with a single wave packet, two wave packets, and three wave packets in the A0 mode at different propagation distances, and construct the training input dataset for ultrasonic guided waves; S2. Construct a training label dataset for ultrasonic guided waves: Based on the known propagation distance x and the group velocity of the A0 mode, calculate the flight time of each wave packet in the ultrasonic guided wave signals with a single wave packet, two wave packets, and three wave packets in the A0 mode, respectively, and generate the first time series label corresponding to the ultrasonic guided wave signal y1(t) with a single wave packet in the A0 mode. The second time series label corresponding to the ultrasonic guided wave signal y2(t) with two wave packets in the A0 mode is generated. The third time series label corresponding to the ultrasonic guided wave signal y3(t) with three wave packets in the A0 mode is generated. Construct a labeled dataset for training ultrasonic guided waves; S3. Construct and train a network model for ultrasonic guided wave damage localization imaging based on convolutional autoencoder; S31. Construct a network model for ultrasonic guided wave damage localization imaging based on convolutional autoencoders. This network model includes convolutional modules and deconvolutional modules. Multiple convolutional modules are used for encoding; multiple deconvolutional modules are used for upsampling decoding. Each convolutional module includes a convolutional layer, a batch normalization layer, and a ReLU activation function. Each deconvolutional module includes a deconvolutional layer, a BN layer, and a ReLU function. A dropout module precedes each deconvolutional module to prevent overfitting. The time series y is regenerated after feature extraction through the deconvolutional module. ToF ; S32. The training input dataset described in step S1 and the training label dataset described in step S2 are used as training sets and input into the network model of the ultrasonic guided wave damage localization imaging method based on convolutional autoencoder for training. S4. Construct a test dataset for ultrasonic guided waves; S41. Place the piezoelectric ceramic sensor on the surface of the carbon fiber reinforced composite laminate, set up a sensors to form a sensor array, sequentially excite and collect ultrasonic guided wave signals, collect b sets of ultrasonic guided wave signals and use them as reference signals after preprocessing; S42. Set up delamination damage on the composite material plate, collect multiple sets of ultrasonic guided wave signals again, preprocess them and subtract them from the reference signal to generate b sets of residual signals containing damage reflection waves, thus forming the test dataset of ultrasonic guided waves. S5. Predict the time series corresponding to the test dataset: Input the test dataset constructed in step S4 into the network model of the ultrasonic guided wave damage localization imaging method based on convolutional autoencoder trained in step S3, estimate the flight time of each wave packet in the ultrasonic guided wave signal, and output the time series y. ToF ; S6. Determine the distribution location of damage in the plate corresponding to the test dataset: The time series y described in step S5... ToF The input delay superposition algorithm is used to perform high-resolution localization imaging of guided wave damage, obtaining the distribution location of the damage in the plate. The pixel value of the distribution location I(x,y) is represented as: Where P represents the number of excitation-receiver sensor pairs; Let c represent the distances from the excitation and receiving sensors to I(x,y), respectively. g This represents the group velocity of mode A0.

2. The ultrasonic guided wave damage localization imaging method based on convolutional autoencoder according to claim 1, characterized in that, Step S1 specifically includes the following steps: S11. The excitation signal f(t) is a 3-cycle narrowband comb signal with a center frequency of 40kHz. When the ultrasonic guided wave signal is in a dispersed state and the propagation distance of the ultrasonic guided wave is x0, the response signal y0(t) collected by the sensor is expressed as: Where F(ω) represents the Fourier transform of the excitation signal f(t); ω represents the frequency; k represents the wave number of the guided wave mode and is related to the frequency ω, i.e., k = k(ω); j represents the imaginary part; e is the natural constant; ∞ represents infinity; t represents time; and d represents the differential. S12. Simulation of ultrasonic guided wave signal y1(t) with a single wave packet in A0 mode: Where, k A This represents the prior wavenumber information of the A0 mode; x A1 This represents the propagation distance of the first wave packet of mode A0; S13. Simulation of ultrasonic guided wave signal y2(t) with two wave packets in A0 mode: Where, x A2 This represents the propagation distance of the second wave packet of mode A0; S14. Simulation of ultrasonic guided wave signal y3(t) with three wave packets in A0 mode: Where, x A3 This represents the propagation distance of the third wave packet of mode A0.

3. The ultrasonic guided wave damage localization imaging method based on convolutional autoencoder according to claim 1, characterized in that, Step S2 specifically includes the following steps: S21. When the ultrasonic guided wave signal is in a non-dispersion state, the wave number k of the guided wave mode is a linear function of the frequency k'. At this time, the phase velocity c of the A0 mode is... p (ω) and group velocity c g (ω) satisfies the following condition: S22, When the propagation distance is x i Time-independent non-dispersion guided wave signal y i_c (t) is represented as: Where δ(·) represents the impulse function; t i x represents i Corresponding time; S23. Based on the propagation distance x of the first wave packet of mode A0 A1 To obtain the first time series label of a single wave packet corresponding to equation (2). in, The group velocity of mode A0 is represented by f. c Indicates the center frequency of the excitation signal; x represents A1 Corresponding time; S24. Based on the propagation distance x of the first wave packet of mode A0 A1 The propagation distance x of the third wave packet of mode A0 A2 Equation (3) is obtained by taking the second time series label containing two wave packets. in, x represents A2 Corresponding time; S25. Based on the propagation distance x of the first wave packet of mode A0 A1 The propagation distance x of the third wave packet of mode A0 A2 The propagation distance x of the third wave packet of mode A0 A3 Equation (4) is obtained by taking the third time series label containing three wave packets. in, x represents A3 Corresponding time; S26, Label the first time series Second time series label and third time series labels Labels used as input datasets for training guided wave signals.

4. The ultrasonic guided wave damage localization imaging method based on convolutional autoencoder according to claim 1, characterized in that, In step S3, there are two convolutional modules, with kernel sizes of 1×10 and 1×3 for the corresponding convolutional layers, and the number of kernels of 4 and 8 for the corresponding convolutional layers, respectively; there are two deconvolutional modules, with kernel sizes of 1×3 and 1×10 for the corresponding deconvolutional layers, and the number of kernels of 8 and 4 for the corresponding deconvolutional layers, respectively; the dropout module ratio is set to 0.

6.

5. The ultrasonic guided wave damage localization imaging method based on convolutional autoencoder according to claim 1, characterized in that, In step S3, the network hyperparameters of the network model for the ultrasonic guided wave damage localization imaging method based on convolutional autoencoder are set as follows: mean squared error as the loss function, stochastic gradient descent as the optimizer, and a learning rate of 5×10⁻⁶. -6 Batch size is set to 64.

Images