Laser-ultrasonic electrical sleeve micro-defect imaging method based on deep learning

Abstract

The application relates to a laser ultrasonic electric sleeve micro-defect imaging method based on deep learning, which comprises the following steps: applying laser ultrasonic body waves to an electric sleeve and collecting signals; performing time domain and frequency domain analysis on the collected signals, extracting characteristic parameters of damage scattering signals, and constructing a micro-damage representation vector fusing multi-dimensional characteristic parameters; using a sparse representation algorithm to realize harmonic signal reconstruction and feature extraction on the micro-damage representation vector through a straight-through wave dictionary, a fundamental frequency scattering wave dictionary and a second harmonic wave dictionary, obtaining reconstructed micro-damage scattering signals, and converting the signals into a two-dimensional spatial domain; inputting the reconstructed micro-damage scattering signals and the two-dimensional spatial domain signals obtained through conversion into a multi-feature fusion neural network model, respectively performing feature extraction, integrating the features obtained from the two parts, and obtaining defect classification and imaging results. Compared with the prior art, the application can realize rapid and accurate imaging of micro-defects in an electric sleeve device.

Description

Technical Field

[0001] This invention relates to the field of electrical bushing defect detection technology, and in particular to a laser ultrasonic imaging method for minute defects in electrical bushings based on deep learning. Background Technology

[0002] Currently, the large-scale integration of new energy sources is making the power grid structure increasingly complex, requiring more generator units to participate in deep peak shaving more frequently, leading to accelerated wear and tear on transmission and transformation equipment. The adverse effects of transmission and transformation equipment failures on the increasingly complex power grid structure are becoming more severe, with the scope of impact continuously expanding, seriously threatening the safe and reliable operation of the power grid. Electrical bushings are fundamental but critical electrical equipment in transmission and transformation systems. However, due to factors such as substandard manufacturing quality, harsh external environmental conditions, and long-term operating conditions, defects can occur in electrical bushings, which are a significant source of transmission and transformation equipment failures, posing a serious threat to production safety and even triggering catastrophic accidents. Therefore, emphasizing and strengthening the research, development, and application of electrical bushing defect detection technology is of great significance for timely detection of hidden equipment defects, reducing equipment failures and potential power grid safety hazards, and ensuring the safe operation of the power grid.

[0003] Non-destructive testing (NDT) technologies, such as ultrasonic testing, X-ray testing, eddy current testing, magnetic flux leakage testing, and penetrant testing, are widely used for the detection and evaluation of defects in power grid transmission and transformation equipment. However, conventional NDT methods are either low in sensitivity, complex and time-consuming, susceptible to environmental influences, or destructive, thus limiting their application to the stringent requirements of safe operation and maintenance of power transmission and transformation equipment. Laser ultrasonic testing technology, with its advantages of high resolution and ease of integration, has become a research hotspot in the field of NDT, providing new methods for equipment damage detection and life assessment.

[0004] Laser ultrasonic testing uses laser light as the excitation source for ultrasonic waves, enabling the detection and evaluation of material structural parameters such as minute defects, hardness, residual stress, elastic modulus, and grain size without damaging the material's structural properties and physical characteristics. Conducting laser ultrasonic non-destructive testing on electrical bushings can pinpoint the location of defects, determine their size, and predict their propagation direction, thereby estimating the remaining lifespan of the equipment. This allows for timely and targeted measures to prevent and control defect expansion, reducing maintenance costs, ensuring the efficient operation of power transmission and transformation equipment, and providing safety assurance for power production. Therefore, research on laser ultrasonic-based detection and imaging technology for minute defects within electrical bushings has significant academic and engineering application value. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology by providing a deep learning-based laser ultrasonic electrical conduit micro-defect imaging method, thereby enabling the imaging of micro-defects in electrical conduit equipment.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A deep learning-based laser ultrasonic electrical bushing micro-defect imaging method includes the following steps:

[0008] Laser ultrasonic waves are applied to the electrical bushing, and the signal is acquired.

[0009] The acquired signals were analyzed in the time and frequency domains to extract the characteristic parameters of the damage scattering signals and construct a micro-damage characterization vector that integrates multi-dimensional characteristic parameters.

[0010] A sparse representation algorithm is used to reconstruct harmonic signals and extract features from the micro-damage characterization vector by using a pre-constructed direct wave dictionary, fundamental frequency scattered wave dictionary and second harmonic dictionary, so as to obtain the reconstructed micro-damage scattered signal.

[0011] The reconstructed micro-damage scattering signal is transformed from the time domain to the two-dimensional spatial domain. The reconstructed micro-damage scattering signal and the transformed two-dimensional spatial domain signal are input into a pre-trained multi-feature fusion neural network model for feature extraction. The features obtained from the two parts are then integrated to obtain the defect classification and imaging results.

[0012] Furthermore, the time-domain and frequency-domain analysis of the acquired signal specifically involves:

[0013] If the acquired signal has stationary signal characteristics, then the time domain characteristics of the signal are extracted through time domain analysis, and then frequency domain analysis is performed through fast Fourier transform to obtain frequency domain characteristics, thereby obtaining the characteristic parameters of the damage scattering signal.

[0014] If the acquired signal has non-stationary characteristics, short-time Fourier transform is used for time-frequency joint analysis to capture the dynamic migration characteristics of signal energy in the time-frequency plane and obtain the characteristic parameters of the damage scattering signal.

[0015] Furthermore, the expression for the through-wave dictionary is:

[0016]

[0017] In the formula, This is the output of the direct-pass dictionary. This is the spectral function of the original signal. For wave number, The spatial distance from the laser excitation point to the signal receiving point. j The imaginary unit, Angular frequency, is the base of the natural logarithm. Pi t For time;

[0018] The expression for the fundamental frequency scattered wave dictionary is:

[0019]

[0020] In the formula, This is the output of the fundamental frequency scattered wave dictionary. The spatial distance from the excitation point to the defect. The spatial distance from the defect to the receiving point;

[0021] The expression for the second harmonic dictionary is:

[0022]

[0023] In the formula, This is the output of the second harmonic dictionary. The center frequency of the excitation signal;

[0024] The expression for the reconstructed micro-damage scattering signal is:

[0025]

[0026] In the formula, This is the reconstructed micro-damage scattering signal.

[0027] Furthermore, the multi-feature fusion neural network model is divided into one branch that extracts features from the reconstructed micro-damage scattering signal through a convolutional layer, reduces the feature dimension through a pooling layer, and finally integrates the features of the first branch through a global average pooling layer.

[0028] The second branch is divided into two branches. The features of the two-dimensional spatial domain signal are extracted through the convolutional layer, the feature dimension is reduced through the pooling layer, and finally the features of the second branch are integrated through the global average pooling layer.

[0029] The features of the first and second branches are integrated, and overfitting is prevented by a Dropout layer. Finally, the classification and imaging tasks are completed by a fully connected layer.

[0030] Furthermore, the training process of the multi-feature fusion neural network model includes:

[0031] A finite element simulation model of electrical bushing equipment with typical defects was established using simulation software. Excitation array elements were set up, and laser sources were placed at the positions of the excitation array elements. The interaction signals between the laser ultrasound and the defects of the finite element simulation model of the electrical bushing equipment were collected to construct a training dataset for training a multi-feature fusion neural network model.

[0032] Furthermore, during the training process, different acquisition signals are obtained by constructing different defect morphologies, different defect depths, and different distances between the laser source and the defect to construct a training dataset.

[0033] Furthermore, the typical defects include cracks, porosity, and inclusions.

[0034] Furthermore, the acquired signal is the micro-damage scattering signal of the transmitted surface wave of the electrical bushing.

[0035] Furthermore, the method is implemented by a system consisting of a laser high-energy excitation hardware platform and a visualization analysis module;

[0036] The laser high-energy excitation hardware platform is used to apply laser ultrasonic body waves to the electrical bushing;

[0037] The visualization analysis module is used to acquire signals from electrical bushings subjected to laser ultrasonic body waves, perform time-domain and frequency-domain analysis on the acquired signals, extract feature parameters of the damage scattering signals, and construct a micro-damage characterization vector that integrates multi-dimensional feature parameters. A sparse representation algorithm is used to reconstruct harmonic signals and extract features from the micro-damage characterization vector using a pre-constructed direct-wave dictionary, fundamental frequency scattering wave dictionary, and second harmonic dictionary, resulting in a reconstructed micro-damage scattering signal. The reconstructed micro-damage scattering signal is then transformed from the time domain to the two-dimensional spatial domain. Both the reconstructed micro-damage scattering signal and the transformed two-dimensional spatial domain signal are input into a pre-trained multi-feature fusion neural network model for feature extraction. The features obtained from the two parts are then integrated to obtain defect classification and imaging results.

[0038] Compared with the prior art, the present invention has the following advantages:

[0039] (1) This invention studies a weak signal extraction method. First, the characteristic parameters of the collected signal are extracted in the time domain, frequency domain and time-frequency domain. Then, the harmonic signal reconstruction and feature extraction of the micro-damage characterization vector are realized by constructing a direct wave dictionary, a fundamental frequency scattering wave dictionary and a second harmonic dictionary. This enables accurate reconstruction of the scattering signal generated by the micro-defect, and then extraction of damage characteristic parameters. Finally, the time domain signal and the two-dimensional spatial domain signal are used together through a neural network model to predict the defect imaging, so as to realize fast and accurate damage inversion imaging.

[0040] (2) This invention proposes a research method on the propagation characteristics of laser ultrasound in electrical bushing equipment and the coupling mechanism with micro-damage. Through the simulation of electrical bushing equipment and laser source, and by setting different defect morphologies, different defect depths and different distances between the laser source and the defect, a rich training dataset is constructed, which provides theoretical support for the effective extraction of feature parameters and damage imaging. Attached Figure Description

[0041] Figure 1 This is a flowchart illustrating a deep learning-based laser ultrasonic electrical conduit micro-defect imaging method provided in an embodiment of the present invention.

[0042] Figure 2 This is a schematic diagram of the processing results of a laser ultrasonic electrical bushing micro-defect imaging method based on deep learning provided in an embodiment of the present invention;

[0043] Figure 3 This is a schematic diagram of the structure of a multi-feature fusion neural network model provided in an embodiment of the present invention;

[0044] Figure 4 This is a schematic diagram of a system for performing damage detection experiments on electrical bushings under test, provided in an embodiment of the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0046] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0047] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0048] Example 1

[0049] like Figure 1 and Figure 2 As shown, this embodiment provides a deep learning-based method for imaging minute defects in laser ultrasonic electrical bushings, including the following steps:

[0050] S1: Apply laser ultrasonic waves to the electrical bushing and acquire the signal;

[0051] S2: Perform time-domain and frequency-domain analysis on the acquired signals, extract the characteristic parameters of the damage scattering signals, and construct a micro-damage characterization vector that integrates multi-dimensional characteristic parameters;

[0052] S3: The sparse representation algorithm is used to reconstruct the harmonic signal and extract the features of the micro-damage characterization vector by using a pre-constructed direct wave dictionary, fundamental frequency scattering wave dictionary and second harmonic dictionary, so as to obtain the reconstructed micro-damage scattering signal.

[0053] S4: The reconstructed micro-damage scattering signal is transformed from the time domain to the two-dimensional spatial domain. The reconstructed micro-damage scattering signal and the transformed two-dimensional spatial domain signal are input into a pre-trained multi-feature fusion neural network model for feature extraction. The features obtained from the two parts are then integrated to obtain the defect classification and imaging results.

[0054] The training process for a multi-feature fusion neural network model includes:

[0055] S5: Establish a finite element simulation model of electrical bushing equipment containing typical defects using simulation software, set up excitation array elements, set up laser sources at the excitation array element positions, and collect the interaction signals between laser ultrasound and the defects of the finite element simulation model of electrical bushing equipment to construct a training dataset.

[0056] The data in the training dataset are processed according to steps S2 and S3 respectively, and then input into the multi-feature fusion neural network model for model training using the method in step S4.

[0057] Specifically, the above training process studied the propagation characteristics of laser ultrasonic bulk waves in multilayer heterogeneous complex electrical bushing equipment based on theoretical and simulation methods, and determined the interaction mechanism between laser ultrasound and typical defects such as cracks, pores, and inclusions, as well as defects with different morphologies, cracking directions, and spatial dimensions.

[0058] Specifically, a finite element simulation model of electrical bushing equipment containing typical defects such as cracks, pores, and inclusions was established to determine the propagation path and detection range of laser ultrasonic bulk waves inside the electrical bushing equipment.

[0059] Simulation models of electrical conduit equipment defects with different shapes, crack directions, and spatial dimensions were established. The sound field propagation law was analyzed, and the interaction mechanism between laser ultrasound and micro-defects in electrical conduit equipment was established.

[0060] Finally, by constructing different defect morphologies, different defect depths, and different distances between the laser source and the defect, we studied the impact of these defects on sound field propagation.

[0061] In this embodiment, the training data collection process includes:

[0062] 1) Model structure parameter settings

[0063] To verify the effectiveness of the proposed method, a simplified finite element model of a multilayer heterogeneous complex structure (total size 300×200×12 mm³, containing four alternating metal / ceramic composite layers with a thickness of 3 mm) was established using simulation software. The material parameters were as follows: alternating metal layers (density 7850 kg / m³, Young's modulus 210 GPa, Poisson's ratio 0.30) and ceramic layers (density 4200 kg / m³, Young's modulus 380 GPa, Poisson's ratio 0.22). During the simulation, the material constitutive model was strictly limited to linear elasticity, and geometric nonlinearity was ignored. Rectangular defects of 10×10 mm² were pre-defined in the critical load-bearing region of the model. The defect boundary conditions were set to hard contact, frictionless, and allowed to open / close freely. The numerical model was discretized using a fine mesh to ensure accurate characterization of interlayer stress transfer and the strain field around the defects.

[0064] 2) Mesh parameter settings

[0065] Taking into account factors such as model calculation accuracy and computational efficiency, the element size was set to 1 mm during mesh generation, the mesh of the damaged element was refined, and the defect was locally meshed, with 60 nodes used to divide the defect area equally.

[0066] 3) Excitation signal parameter settings

[0067] In this simulation, the excitation array element is single and its position is fixed. An out-of-plane displacement signal is symmetrically applied at the position of the excitation array element. The excitation signal is a sinusoidal time-domain pulse signal modulated by a 5-cycle Hanning window. The amplitude of the displacement signal is 1e-6 m, and the excitation frequency is set to 300 kHz.

[0068] During the experiment, the position point A of the excitation array element was kept stationary. A laser line scan was performed by moving the position point B of the laser receiving array element along the x-axis on the sample surface. The starting point was located 5 mm to the right of the excitation array element, and the scanning range was 50 mm. The ultrasonic bulk wave signal excited by the excitation array element propagated inside the sample. When it encountered a defect, it was reflected and received by the receiving array element. During the laser line scan, Defined as the spatial distance from the excitation point A to the defect. The distance from the defect to receiving point B; Let A be the spatial distance from the laser excitation point A to the signal receiving point B.

[0069] In step S2, the time-domain and frequency-domain analysis of the acquired signal is specifically performed as follows:

[0070] If the acquired signal has stationary signal characteristics, then the time domain characteristics of the signal are extracted through time domain analysis, and then frequency domain analysis is performed through fast Fourier transform to obtain frequency domain characteristics, thereby obtaining the characteristic parameters of the damage scattering signal.

[0071] If the acquired signal has non-stationary characteristics, short-time Fourier transform is used for time-frequency joint analysis to capture the dynamic migration characteristics of signal energy in the time-frequency plane and obtain the characteristic parameters of the damage scattering signal.

[0072] Specifically, the micro-damage scattering signal of the transmitted surface wave is extracted, and the characteristic parameters of the damage scattering signal are extracted in the time domain, frequency domain, and time-frequency domain.

[0073] First, time-domain features of the signal are extracted based on time-domain analysis, including mean and root mean square (RMS) values, to characterize the signal amplitude distribution. Then, frequency-domain analysis is performed using Fast Fourier Transform (FFT) to reveal the signal's energy concentration characteristics and frequency component evolution patterns.

[0074] To address the characteristics of non-stationary signals, short-time Fourier transform (STFT) is employed for joint time-frequency analysis to capture the dynamic migration characteristics of signal energy in the time-frequency plane. Feature parameters of the damage scattering signal are extracted from the time domain, frequency domain, and time-frequency domain, and a micro-damage characterization vector integrating multi-dimensional feature parameters is constructed.

[0075] In step S3, the process of harmonic signal reconstruction and feature extraction is as follows:

[0076] Based on mathematical models of direct-pass waves, fundamental frequency scattered waves, and second harmonics, three types of dictionaries were constructed, and a sparse representation algorithm was used to extract features from the harmonic signals. The excitation signal is S( t The received signal is G( t )= G1( t ) +G2( t )+ G3( t ),in:

[0077] The expression for the direct-through wave dictionary is:

[0078]

[0079] In the formula, This is the output of the direct-pass dictionary. This is the spectral function of the original signal. For wave number, The spatial distance from the laser excitation point to the signal receiving point. t For time;

[0080] The expression for the fundamental frequency scattered wave dictionary is:

[0081]

[0082] In the formula, This is the output of the fundamental frequency scattered wave dictionary. The spatial distance from the excitation point to the defect. The spatial distance from the defect to the receiving point;

[0083] The expression for the second harmonic dictionary is:

[0084]

[0085] In the formula, This is the output of the second harmonic dictionary. The center frequency of the excitation signal;

[0086] A sparse representation algorithm is used to reconstruct harmonic signals and extract features from time-domain signals by constructing three dictionaries.

[0087] In step S4, such as Figure 3 As shown, the multi-feature fusion neural network model is divided into one branch that extracts features from the reconstructed micro-damage scattering signal through a convolutional layer, reduces the feature dimension through a pooling layer, and finally integrates the features of the first branch through a global average pooling layer.

[0088] The second branch is divided into two branches. The features of the two-dimensional spatial domain signal are extracted through the convolutional layer, the feature dimension is reduced through the pooling layer, and finally the features of the second branch are integrated through the global average pooling layer.

[0089] The features from the first and second branches are merged and integrated, and overfitting is prevented by a Dropout layer. Finally, the features are passed through a fully connected layer and a softmax activation function to complete the classification and imaging tasks.

[0090] Specifically, in this embodiment, the multi-feature fusion neural network model mainly consists of a data input layer, a convolutional layer, a pooling layer, a global average pooling layer, and an output layer.

[0091] Two convolutional layers were designed to extract the input features from the reconstructed micro-damage scattering signal, two pooling layers were designed to reduce the feature dimensionality, and finally the features were integrated through a global average pooling layer.

[0092] By transforming the time-domain signal into a two-dimensional spatial domain, two convolutional layers were designed to extract the input features, two pooling layers were designed to reduce the feature dimensionality, and finally, the features were integrated through a global average pooling layer.

[0093] The features obtained from the two parts are integrated and input into the Dropout layer to prevent overfitting. Finally, the features are passed through a fully connected layer and the softmax function is used to complete the classification and imaging tasks.

[0094] During the research process, relevant parameters were repeatedly tested and adjusted. Furthermore, the ReLU activation function was used to achieve non-linear feature mapping. The RMSProp optimization algorithm was selected during network training, with a learning rate of 0.001, a loss function of MSE, and 300 iterations.

[0095] After the multi-feature fusion neural network model was trained, it was used to predict the rectangular defect model test set data. The shape and position of the result were consistent with the experimental settings, and the prediction results were accurate.

[0096] Example 2

[0097] This embodiment provides an electrical sleeve defect imaging system that implements the deep learning-based laser ultrasonic electrical sleeve micro-defect imaging method as described in Embodiment 1, comprising:

[0098] A high-energy laser excitation hardware platform is used to apply laser ultrasonic bulk waves to electrical bushings;

[0099] The visualization and analysis module is used to acquire signals from electrical bushings subjected to laser ultrasonic body waves. It performs time-domain and frequency-domain analysis on the acquired signals, extracts feature parameters of the damage scattering signals, and constructs a micro-damage characterization vector that integrates multi-dimensional feature parameters. A sparse representation algorithm is used to reconstruct harmonic signals and extract features from the micro-damage characterization vector using a pre-built direct-wave dictionary, fundamental frequency scattering wave dictionary, and second harmonic dictionary, resulting in a reconstructed micro-damage scattering signal. The reconstructed micro-damage scattering signal is then transformed from the time domain to the two-dimensional spatial domain. Both the reconstructed micro-damage scattering signal and the transformed two-dimensional spatial domain signal are input into a pre-trained multi-feature fusion neural network model for feature extraction. The features obtained from both parts are then integrated to obtain defect classification and imaging results.

[0100] In other words, this embodiment requires the design and development of a hardware platform with high-energy laser excitation and rapid scanning capabilities; and the development of visualization analysis software for signal acquisition, processing, and damage imaging during the micro-damage detection process of electrical conduit equipment.

[0101] Ultimately, we developed an integrated system for micro-damage detection, imaging, and evaluation of electrical conduit equipment, combining high-energy laser excitation, rapid scanning, and signal analysis functions.

[0102] like Figure 4 As shown, the hardware experimental platform for the electrical bushing under test includes emitting excitation light onto the electrical bushing under test through a laser exciter and a focusing objective lens; collecting the probe light of the electrical bushing under test through a dual-wave mixing interferometer; and collecting the probe light through a photodetector and analyzing it through a computer.

[0103] It should be noted that the specific details and beneficial effects of the device in this application can be found in the above-described method embodiments, and will not be repeated here.

[0104] This embodiment also provides a computer-readable storage medium on which a computer program is stored, and the computer program is executed by a processor as described in Embodiment 1.

[0105] The computer program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This computer program code can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the computer program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The computer program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0106] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium can be a machine-readable signal medium or a machine-readable storage medium. A computer-readable storage medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0107] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A deep learning-based laser ultrasonic electrical conduit micro-defect imaging method, characterized in that, Includes the following steps: Laser ultrasonic waves are applied to the electrical bushing, and the signal is acquired. The acquired signals were analyzed in the time and frequency domains to extract the characteristic parameters of the damage scattering signals and construct a micro-damage characterization vector that integrates multi-dimensional characteristic parameters. A sparse representation algorithm is used to reconstruct harmonic signals and extract features from the micro-damage characterization vector by using a pre-constructed direct wave dictionary, fundamental frequency scattered wave dictionary and second harmonic dictionary, so as to obtain the reconstructed micro-damage scattered signal. The reconstructed micro-damage scattering signal is transformed from the time domain to the two-dimensional spatial domain. The reconstructed micro-damage scattering signal and the transformed two-dimensional spatial domain signal are input into a pre-trained multi-feature fusion neural network model for feature extraction. The features obtained from the two parts are then integrated to obtain the defect classification and imaging results.

2. The method for imaging minute defects in laser ultrasonic electrical conduits based on deep learning according to claim 1, characterized in that, The specific steps for performing time-domain and frequency-domain analysis on the acquired signals are as follows: If the acquired signal has stationary signal characteristics, then the time domain characteristics of the signal are extracted through time domain analysis, and then frequency domain analysis is performed through fast Fourier transform to obtain frequency domain characteristics, thereby obtaining the characteristic parameters of the damage scattering signal. If the acquired signal has non-stationary characteristics, short-time Fourier transform is used for time-frequency joint analysis to capture the dynamic migration characteristics of signal energy in the time-frequency plane and obtain the characteristic parameters of the damage scattering signal.

3. The method for imaging minute defects in laser ultrasonic electrical conduits based on deep learning according to claim 1, characterized in that, The expression for the direct-pass wave dictionary is: In the formula, This is the output of the direct-pass dictionary. This is the spectral function of the original signal. For wave number, The spatial distance from the laser excitation point to the signal receiving point. j The imaginary unit, Angular frequency, is the base of the natural logarithm. Pi t For time; The expression for the fundamental frequency scattered wave dictionary is: In the formula, This is the output of the fundamental frequency scattered wave dictionary. The spatial distance from the excitation point to the defect. The spatial distance from the defect to the receiving point; The expression for the second harmonic dictionary is: In the formula, This is the output of the second harmonic dictionary. The center frequency of the excitation signal.

4. The method for imaging minute defects in laser ultrasonic electrical bushings based on deep learning according to claim 3, characterized in that, The expression for the reconstructed micro-damage scattering signal is: In the formula, This is the reconstructed micro-damage scattering signal.

5. The method for imaging minute defects in laser ultrasonic electrical conduits based on deep learning according to claim 1, characterized in that, The multi-feature fusion neural network model is divided into one branch that extracts features from the reconstructed micro-damage scattering signal through a convolutional layer, reduces the feature dimension through a pooling layer, and finally integrates the features of the first branch through a global average pooling layer. The second branch is divided into two branches. The features of the two-dimensional spatial domain signal are extracted through the convolutional layer, the feature dimension is reduced through the pooling layer, and finally the features of the second branch are integrated through the global average pooling layer. The features of the first and second branches are integrated, and overfitting is prevented by a Dropout layer. Finally, the classification and imaging tasks are completed by a fully connected layer.

6. The method for imaging minute defects in laser ultrasonic electrical bushings based on deep learning according to claim 1, characterized in that, The training process of the multi-feature fusion neural network model includes: A finite element simulation model of electrical bushing equipment with typical defects was established using simulation software. Excitation array elements were set up, and laser sources were placed at the positions of the excitation array elements. The interaction signals between the laser ultrasound and the defects of the finite element simulation model of the electrical bushing equipment were collected to construct a training dataset for training a multi-feature fusion neural network model.

7. The method for imaging minute defects in laser ultrasonic electrical bushings based on deep learning according to claim 6, characterized in that, During the training process, different acquisition signals are obtained by constructing different defect morphologies, different defect depths, and different distances between the laser source and the defect to build a training dataset.

8. The method for imaging minute defects in laser ultrasonic electrical conduits based on deep learning according to claim 6, characterized in that, The typical defects include cracks, porosity, and inclusions.

9. The method for imaging minute defects in laser ultrasonic electrical conduits based on deep learning according to claim 1, characterized in that, The acquired signal is the micro-damage scattering signal of the transmitted surface wave of the electrical bushing.

10. The method for imaging minute defects in laser ultrasonic electrical bushings based on deep learning according to claim 1, characterized in that, The method is implemented by a system consisting of a laser high-energy excitation hardware platform and a visualization analysis module. The laser high-energy excitation hardware platform is used to apply laser ultrasonic body waves to the electrical bushing; The visualization analysis module is used to acquire signals from electrical bushings subjected to laser ultrasonic body waves, perform time-domain and frequency-domain analysis on the acquired signals, extract feature parameters of the damage scattering signals, and construct a micro-damage characterization vector that integrates multi-dimensional feature parameters. A sparse representation algorithm is used to reconstruct harmonic signals and extract features from the micro-damage characterization vector using a pre-constructed direct-wave dictionary, fundamental frequency scattering wave dictionary, and second harmonic dictionary, resulting in a reconstructed micro-damage scattering signal. The reconstructed micro-damage scattering signal is then transformed from the time domain to the two-dimensional spatial domain. Both the reconstructed micro-damage scattering signal and the transformed two-dimensional spatial domain signal are input into a pre-trained multi-feature fusion neural network model for feature extraction. The features obtained from the two parts are then integrated to obtain defect classification and imaging results.

Images