CN116698978A - Method and device for ultrasonic recognition of transformer defects by convolutional neural network - Google Patents

Abstract

The embodiment of the invention discloses a method and a device for identifying defects of a transformer by using ultrasonic waves of a convolutional neural network, wherein the method comprises the following steps: detecting the current transformer by using an ultrasonic detection device, and collecting waveform data of ultrasonic waves corresponding to different defects; extracting spectrum information of the waveform data by applying Fourier transformation; generating characteristic images corresponding to different defect categories by using a Graham angle field algorithm; dividing the generated characteristic image into a training set and a testing set, and inputting the training set into a convolutional neural network model for training; inputting the test set into a trained convolutional neural network model, extracting and classifying defect features in the feature images, and determining the ultrasonic defect type.

Description

Method and device for ultrasonic recognition of transformer defects by convolutional neural network

technical field

The present invention relates to the technical field of defect identification of current transformers, and more specifically, to a method and device for ultrasonically identifying defects of transformers using a convolutional neural network.

Background technique

In order to perform rapid and full-coverage pre-screening of metering transformers, select metering transformers with a high probability of inferior quality for further investigation through full performance tests, improve the defect detection rate of metering transformers and measure transformers to be put into operation The quality level of the transformer is tested by ultrasonic, a fast non-destructive testing method, in this project.

Basic principles of ultrasonic testing Signals usually appear in the form of waves, such as sound waves and electromagnetic waves. Ultrasonic waves are sound waves, and their frequency is usually greater than 20KHz. It is a mechanical vibration in the form of waves caused by the propagation of ultrasonic vibration in the medium. Ultrasound has the characteristics of short wavelength, high frequency and directional emission, which can produce reflection, refraction and wave mode transformation on the interface. Ultrasonic energy is proportional to the square of its frequency, so the energy density is high, the propagation distance is large, and the penetration ability is strong.

The reflection and transmission of ultrasonic waves on the interface composed of different material media are closely related to the acoustic impedance of the material. Because the elastic modulus and density of different material media at the interface are different, the acoustic impedance of ultrasonic waves propagating in this area is also different. The strength of the reflected wave depends critically on the acoustic impedance Z1 and Z2 of the material medium on both sides of the interface, so defects with the same size but different properties have different defect echo intensities. The acoustic impedance of air is much smaller than that of steel. For gas-containing defects such as pores and cracks in solid materials, it can be approximated that sound waves are totally reflected on the surface of the defect. Ultrasonic testing is to judge whether there is a defect in the device based on the echo of the ultrasonic wave in the device.

General ultrasonic testing research is carried out on devices with relatively simple structures and relatively single materials, and the waveforms are relatively regular and easy to analyze. The figure below is a schematic diagram of the current transformer section, in which the gray ring represents the iron core, and the black, yellow, brown and orange areas corresponding to the iron core are the core plastic shell, secondary coil, buffer layer and epoxy resin, and the purple area is the ultrasonic emission and receiving device. It can be seen that the internal structure of the current transformer is complex, there are many types of materials, and the echo waveform is extremely complex, so the general analysis method is no longer applicable.

Contents of the invention

Current transformers may have various defects due to various uncertain factors in the manufacturing process, and related defects can be detected by non-contact and non-destructive ultrasonic defect detection methods. Different defects correspond to different ultrasonic waveforms, so defects can be identified based on ultrasonic waveforms. However, manual identification is not only low in accuracy but also very inefficient. In view of this problem, the present invention is proposed. Embodiments of the present invention provide a convolutional neural network ultrasonic recognition method and device for transformer defects.

According to an aspect of an embodiment of the present invention, a convolutional neural network ultrasonic defect recognition method is provided, including:

Use the ultrasonic testing device to detect the current transformer, and collect the waveform data of ultrasonic waves corresponding to different defects;

Apply the Fourier transform to extract the spectral information of the waveform data;

Using the Graham angle field algorithm to generate spectral information into feature images corresponding to different defect categories;

Divide the generated feature images into a training set and a test set, and input the training set into the convolutional neural network model for training;

Input the test set into the trained convolutional neural network model, extract and classify the defect features in the feature image, and determine the ultrasonic defect type.

Optionally, the defect types of the current transformer include: air bubbles and cracks.

Optionally, the feature image generated using the Graham angle field algorithm is a two-dimensional color image.

Optionally, the trained convolutional neural network model includes 3 convolutional layers, 2 pooling layers, and 2 fully connected layers; the convolution kernels of the 3 convolutional layers are all 2×2, and the convolution The number of cores is 16, 32, and 48 respectively; the pooling window size of the two pooling layers is 2×2, and the stride is 2; the number of neurons in the two fully connected layers is 240 and 120; output through the softmax classifier 10 classification results.

According to another aspect of the embodiments of the present invention, a convolutional neural network ultrasonic defect recognition device is provided, including:

The waveform data acquisition module is used to detect the current transformer by using the ultrasonic detection device, and collect the waveform data corresponding to the ultrasonic waves under different defects;

Spectrum information extraction module, for applying Fourier transform to extract the spectrum information of waveform data;

A feature image generation module, configured to use the Graham angle field algorithm to generate feature images corresponding to different defect categories from spectral information;

A model training module, used to divide the generated feature images into a training set and a test set, and input the training set to a convolutional neural network model for training;

The defect type determination module is used to input the test set into the trained convolutional neural network model, extract and classify the defect features in the feature image, and determine the ultrasonic defect type.

Optionally, the defect types of the current transformer include: air bubbles and cracks.

Optionally, the feature image generated using the Graham angle field algorithm is a two-dimensional color image.

Optionally, the trained convolutional neural network model includes 3 convolutional layers, 2 pooling layers, and 2 fully connected layers; the convolution kernels of the 3 convolutional layers are all 2×2, and the convolution The number of cores is 16, 32, and 48 respectively; the pooling window size of the two pooling layers is 2×2, and the stride is 2; the number of neurons in the two fully connected layers is 240 and 120; output through the softmax classifier 10 classification results.

According to another aspect of the embodiments of the present invention, there is also provided a computer-readable storage medium, wherein the storage medium stores a computer program, and the computer program is used to execute the above method.

According to another aspect of the embodiments of the present invention, an electronic device is also provided, and the electronic device includes:

processor;

memory for storing said processor-executable instructions;

The processor is configured to read the executable instruction from the memory, and execute the instruction to implement the above method.

The ultrasonic defect recognition method of the convolutional neural network proposed by the present invention can improve the recognition accuracy and further reduce the time required for defect type recognition by taking advantage of the image classification algorithm in machine vision. In the present invention, fast Fourier transform and Graham angle field algorithm are used to process the acquired feature data, and feature pictures corresponding to different defect types of current transformers are acquired. Then the convolutional neural network is applied to classify the feature images to realize defect type identification. Compared with previous methods, the present invention can apply mature image classification algorithms among artificial intelligence algorithms to defect identification, so as to realize accurate and rapid identification of defect types.

The technical solutions of the present invention will be described in further detail below with reference to the accompanying drawings and embodiments.

Description of drawings

The above and other objects, features and advantages of the present invention will become more apparent by describing the embodiments of the present invention in more detail with reference to the accompanying drawings. The accompanying drawings are used to provide a further understanding of the embodiments of the present invention, and constitute a part of the specification, and are used together with the embodiments of the present invention to explain the present invention, and do not constitute limitations to the present invention. In the drawings, the same reference numerals generally represent the same components or steps.

Fig. 1 is a schematic flow chart of an ultrasonic defect recognition method of a convolutional neural network provided by an exemplary embodiment of the present invention;

Fig. 2 is a schematic diagram of a three-dimensional model of a low-voltage electromagnetic current transformer provided by an exemplary embodiment of the present invention;

Fig. 3 is a flowchart of a defect identification algorithm provided by an exemplary embodiment of the present invention;

Fig. 4 is a schematic diagram of the accuracy curve of the CNN training process provided by an exemplary embodiment of the present invention;

Fig. 5 is a schematic diagram of the CNN training process loss curve provided by an exemplary embodiment of the present invention;

Fig. 6 is a schematic structural diagram of a convolutional neural network ultrasonic defect recognition device provided by an exemplary embodiment of the present invention.

Detailed ways

Hereinafter, exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Apparently, the described embodiments are only some embodiments of the present invention, rather than all embodiments of the present invention, and it should be understood that the present invention is not limited by the exemplary embodiments described here.

It should be noted that the relative arrangements of components and steps, numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present invention unless specifically stated otherwise.

Those skilled in the art can understand that terms such as "first" and "second" in the embodiments of the present invention are only used to distinguish different steps, devices or modules, etc. necessary logical sequence.

It should also be understood that in the embodiments of the present invention, "plurality" may refer to two or more than two, and "at least one" may refer to one, two or more than two.

It should also be understood that for any component, data or structure mentioned in the embodiments of the present invention, it can generally be understood as one or more unless there is a clear limitation or a contrary suggestion is given in the context.

In addition, the term "and/or" in the present invention is only an association relationship describing associated objects, indicating that there may be three relationships, for example, A and/or B may indicate: A exists alone, and A and B exist at the same time , there are three cases of B alone. In addition, the character "/" in the present invention generally indicates that the contextual objects are an "or" relationship.

It should also be understood that the description of the various embodiments of the present invention emphasizes the differences between the various embodiments, and the same or similar points can be referred to each other, and for the sake of brevity, details are not repeated one by one.

At the same time, it should be understood that, for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

The following description of at least one exemplary embodiment is merely illustrative in nature and in no way taken as limiting the invention, its application or uses.

Techniques, methods and devices known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, techniques, methods and devices should be considered part of the description.

It should be noted that like numerals and letters denote like items in the following figures, therefore, once an item is defined in one figure, it does not require further discussion in subsequent figures.

Embodiments of the present invention can be applied to electronic equipment such as terminal equipment, computer systems, servers, etc., which can operate with many other general-purpose or special-purpose computing system environments or configurations. Examples of well known terminal devices, computing systems, environments and/or configurations suitable for use with electronic devices such as terminal devices, computer systems, servers include, but are not limited to: personal computer systems, server computer systems, thin clients, thick client Computers, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, networked personal computers, minicomputer systems, mainframe computer systems, and distributed cloud computing technology environments including any of the foregoing, etc.

Electronic devices such as terminal devices, computer systems, servers, etc. may be described in the general context of computer system-executable instructions, such as program modules, being executed by the computer system. Generally, program modules may include routines, programs, objects, components, logic, data structures, etc., that perform particular tasks or implement particular abstract data types. The computer system/server can be practiced in distributed cloud computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed cloud computing environment, program modules may be located in both local and remote computing system storage media including storage devices.

exemplary method

Fig. 1 is a schematic flowchart of a convolutional neural network ultrasonic defect recognition method provided by an exemplary embodiment of the present invention. As shown in Figure 1, the ultrasonic defect recognition method of convolutional neural network includes:

S1: Use the ultrasonic testing device to detect the current transformer, and collect the waveform data of ultrasonic waves corresponding to different defects.

Optionally, the defect types of the current transformer include: air bubbles and cracks.

S2: Apply Fourier transform to extract the spectrum information of the waveform data.

In the embodiment of the present invention, Fourier transform (FFT) is a commonly used signal transformation method, which can transform the signal in the time domain into the frequency domain for analysis, and extract the frequency domain information in the original signal. The transformation process is to use different sinusoidal signals to analyze the different components contained in the original signal. By applying Fourier transform, any signal can be decomposed into the sum of countless sinusoidal signals of different frequencies.

For any periodic voltage or current signal f(t), it can be expressed as a function of period T:

f(t)=f(t+kT)(k=0,1,2,...)(1)

The Fourier series of equation (1) can be expressed as:

In equation (2), a ₀ is the DC component, ω is the fundamental angular frequency, n is the harmonic order, a _n and b _n are the sine phase coefficient and cosine term coefficient of the nth harmonic, respectively.

According to Euler's formula:

Formula (2) can be transformed into formula (4):

In the above formula

For a continuous function f(t) whose fundamental angular frequency is ω, when it satisfies the Dirichlet condition and is absolutely integrable, its Fourier transform can be expressed as:

In mathematical analysis, the signals analyzed by Fourier transform are all continuous. However, discrete signals are stored in modern signal processing systems, and discrete Fourier transform is required to analyze discrete signals. For a discrete signal f(n) stored in a modern signal processing system, when the number of sampling points is N, the discrete Fourier transform formula is:

In formula (6), f(n) is the original discrete signal, N is the number of sampling points of the discrete signal, and F(k) is the spectrum of the original signal after discrete Fourier transform.

Obtain the original fault signal and apply FFT to generate corresponding spectrum information.

S3: Use the Graham angle field algorithm to generate feature images corresponding to different defect categories from the spectral information.

Optionally, the feature image generated using the Graham angle field algorithm is a two-dimensional color image.

In the embodiment of the present invention, after obtaining the original fault signal and applying FFT to generate corresponding spectrum information, the Graham Angle Field (GAF) algorithm can be used to generate pictures corresponding to different fault types from the spectrum information. The specific process of GAF generating pictures is as follows: for a given one-dimensional sequence X=(x ₁ ,x ₂ ,…,x _n ) composed of actual observations, in this sequence x _i (i=1,2,…,n ) corresponds to time t _i , and the time interval is 1/n. Use the maximum and minimum values in the sequence to scale it to [-1,1]. The specific scaling formula is:

The scaled sequence The value of ψ i is mapped to the angle ψ _i , and its corresponding time stamp t _i is mapped to the radius r, so that the scaled time series can be re-expressed in the polar coordinate system, as shown in formula (8).

In the formula, N is a constant factor for adjusting the radial span of polar coordinates.

After mapping a 1D signal to a polar coordinate system, we can easily take advantage of the angular perspective to identify temporal correlations in different time intervals by considering the difference in trigonometric functions between each point. That is, applying trigonometric functions to generate the GAF matrix:

Since the value range of the elements in the GAF matrix is [-1, 1], it is necessary to scale the value of each element in the GAF matrix to between 0 and 255 through formula (10) to make it correspond to the pixel data of the image, thus obtaining 2D image.

I(j,k)=int(127.5(G(j,k)+1))(10)

In the formula, I(j,k) is the pixel value of the jth (j=1,2,…,n)th row and the kth (k=1,2,…,n)th column in the generated image. int is the rounding function, and G(j,k) is the value of the jth row and kth column element of the GAF matrix.

Therefore, using FFT to obtain spectral information and using GAF to generate two-dimensional color images corresponding to different defect categories can effectively improve the classification accuracy of CNN.

S4: Divide the generated feature images into a training set and a test set, and input the training set into the convolutional neural network model for training.

Optionally, the trained convolutional neural network model includes 3 convolutional layers, 2 pooling layers, and 2 fully connected layers; the convolution kernels of the 3 convolutional layers are all 2×2, and the convolution The number of cores is 16, 32, and 48 respectively; the pooling window size of the two pooling layers is 2×2, and the stride is 2; the number of neurons in the two fully connected layers is 240 and 120; output through the softmax classifier 10 classification results.

In the embodiment of the present invention, a typical convolutional neural network (CNN) generally includes a convolutional layer, a pooling layer, a fully connected layer and an output layer. When classifying a large number of pictures, CNN can perform feature extraction mapping and dimensionality reduction on the input pictures through the convolutional layer and the pooling layer, and finally output the classification results through the fully connected layer and the output layer.

The CNN used in the present invention is expanded on the basis of LeNet-5, and a convolutional layer is added to the basic LeNet-5 network. The final network has 3 convolutional layers, 2 pooling layers, and 2 fully connected layers. The convolution kernel size of the three convolutional layers is 2×2, and the number of convolution kernels is 16, 32, and 48 respectively; the pooling window size of the two pooling layers is 2×2, and the stride is 2; 2 The number of neurons in each fully connected layer is 240 and 120; finally, 10 classification results are output through softmax. The above CNN structure is shown in Table 1.

Table 1 CNN structure

S5: Input the test set into the trained convolutional neural network model, extract and classify the defect features in the feature image, and determine the ultrasonic defect type.

The technical solutions in the embodiments of the present invention will be clearly and completely described below in combination with the best embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

This patent takes the low-voltage electromagnetic current transformer as an example for research, and establishes a three-dimensional geometric model as shown in Figure 2.

Figure 2 is a schematic diagram of a cut section of a current transformer, in which the gray ring represents the iron core, and the black, yellow, brown, and orange areas centered on the iron core are respectively the iron core plastic shell, secondary coil, buffer layer, and epoxy resin, and the purple area is the ultrasonic emission and receiving device. It can be seen that the internal structure of the current transformer is complex, there are many types of materials, and the echo waveform is extremely complex, so the general analysis method is no longer applicable.

As shown in Figure 3, the scheme includes the following steps:

Step 1. Obtain ultrasonic waveforms under different defects of the current transformer by experiment or simulation.

Step 2. Apply fast Fourier transform to the ultrasonic waveform obtained in step 1 to obtain spectrum information.

Step 3, using the GAF to generate two-dimensional color images corresponding to different defect categories from the spectral information generated in step 2.

Step 4, divide the image sample of step 3 into a training set and a test set, and train the CNN model of the training set, complete the network training and save the network model;

Step 5. Input the test set in step 4 into the trained CNN model for testing, and the CNN model performs adaptive extraction and classification of defect features, thereby performing defect discrimination;

Step 6. Collect ultrasonic waveforms under different defect conditions, and repeat the above steps 1-5.

In order to ensure that the trained CNN can perform accurate classification, the training data must contain ultrasonic data of different defects in the line. The training data can be obtained through simulation or experiment. When CNN is used to realize multi-classification, the output layer adopts softmax function, and the final output result is a one-dimensional vector composed of 0 and 1, as shown in Table 2.

Table 2 Classification results corresponding to different defects

CNN is used to classify and train the images generated by GAF for the spectral information in the ultrasonic data acquisition window. A total of 4 rounds of training are set up, and a test set is verified every 30 iterations, so there are 120 training iterations in total. The specific classification accuracy and loss curves during the training process are shown in Figure 4 and Figure 5.

It can be seen from Figures 4 and 5 that the convergence speed of CNN is very fast during the training process. At the 10th iteration, the classification accuracy of the training set is close to 100%, and the loss is also reduced to about 0.1; after the 40th iteration, training The set classification accuracy has stabilized to 100%, and the loss is also close to 0. For a test set verification every 30 iterations, the accuracy of the classification results is maintained at 100%, and the loss is also close to 0. And it can be seen from Figure 5 that the defect type recognition model of FFT-GAF-CNN maintains good classification performance for different defect classification effects of current transformers.

Therefore, the present invention can improve the identification accuracy and further reduce the time required for defect type identification by taking advantage of the image classification algorithm in machine vision. In the present invention, fast Fourier transform and Graham angle field algorithm are used to process the acquired feature data, and feature pictures corresponding to different defect types of current transformers are acquired. Then the convolutional neural network is applied to classify the feature images to realize defect type identification. Compared with previous methods, the present invention can apply mature image classification algorithms among artificial intelligence algorithms to defect identification, so as to realize accurate and rapid identification of defect types.

Exemplary device

Fig. 6 is a schematic structural diagram of a convolutional neural network ultrasonic defect recognition device provided by an exemplary embodiment of the present invention. As shown in Figure 6, the ultrasonic defect recognition device of the convolutional neural network proposed in this embodiment includes:

The waveform data acquisition module is used to detect the current transformer by using the ultrasonic detection device, and collect the waveform data corresponding to the ultrasonic waves under different defects;

Spectrum information extraction module, for applying Fourier transform to extract the spectrum information of waveform data;

A feature image generation module, configured to use the Graham angle field algorithm to generate feature images corresponding to different defect categories from spectral information;

The model training module is used to divide the generated feature images into a training set and a test set, and input the training set into the convolutional neural network model for training;

The defect type determination module is used to input the test set into the trained convolutional neural network model, extract and classify the defect features in the feature image, and determine the ultrasonic defect type.

Optionally, the defect types of the current transformer include: air bubbles and cracks.

Optionally, the feature image generated using the Graham angle field algorithm is a two-dimensional color image.

Optionally, the trained convolutional neural network model includes 3 convolutional layers, 2 pooling layers, and 2 fully connected layers; the convolution kernels of the 3 convolutional layers are all 2×2, and the convolution The number of cores is 16, 32, and 48 respectively; the pooling window size of the two pooling layers is 2×2, and the stride is 2; the number of neurons in the two fully connected layers is 240 and 120; output through the softmax classifier 10 classification results.

The ultrasonic defect recognition device of the convolutional neural network in the embodiment of the present invention corresponds to the ultrasonic defect recognition method of the convolutional neural network in another embodiment of the present invention, and will not be repeated here.

The basic principles of the present disclosure have been described above in conjunction with specific embodiments, but it should be pointed out that the advantages, advantages, effects, etc. mentioned in the present disclosure are only examples rather than limitations, and these advantages, advantages, effects, etc. Various embodiments of the present disclosure must have. In addition, the specific details disclosed above are only for the purpose of illustration and understanding, rather than limitation, and the above details do not limit the present disclosure to be implemented by using the above specific details.

Each embodiment in this specification is described in a progressive manner, each embodiment focuses on the difference from other embodiments, and the same or similar parts of each embodiment can be referred to each other. As for the system embodiment, since it basically corresponds to the method embodiment, the description is relatively simple, and for the related parts, please refer to the part of the description of the method embodiment.

The block diagrams of devices, devices, devices, and systems involved in the present disclosure are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagrams. As will be appreciated by those skilled in the art, these devices, devices, devices, systems may be connected, arranged, configured in any manner. Words such as "including", "comprising", "having" and the like are open-ended words meaning "including but not limited to" and may be used interchangeably therewith. As used herein, the words "or" and "and" refer to the word "and/or" and are used interchangeably therewith, unless the context clearly dictates otherwise. As used herein, the word "such as" refers to the phrase "such as but not limited to" and can be used interchangeably therewith.

The methods and apparatus of the present disclosure may be implemented in many ways. For example, the methods and apparatuses of the present disclosure may be implemented by software, hardware, firmware or any combination of software, hardware, and firmware. The above sequence of steps for the method is for illustration only, and the steps of the method of the present disclosure are not limited to the sequence specifically described above unless specifically stated otherwise. Furthermore, in some embodiments, the present disclosure can also be implemented as programs recorded in recording media, the programs including machine-readable instructions for realizing the method according to the present disclosure. Thus, the present disclosure also covers a recording medium storing a program for executing the method according to the present disclosure.

It should also be pointed out that in the systems, devices and methods of the present disclosure, each component or each step can be decomposed and/or reassembled. These decompositions and/or recombinations should be considered equivalents of the present disclosure. The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the present disclosure. Thus, the present disclosure is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

The foregoing description has been presented for purposes of illustration and description. Furthermore, this description is not intended to limit the disclosed embodiments to the forms disclosed herein. Although a number of example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims (10)

1. an ultrasonic defect recognition method of convolutional neural network, is characterized in that, comprises: Use the ultrasonic testing device to detect the current transformer, and collect the waveform data of ultrasonic waves corresponding to different defects; Apply the Fourier transform to extract the spectral information of the waveform data; Using the Graham angle field algorithm to generate spectral information into feature images corresponding to different defect categories; Divide the generated feature images into a training set and a test set, and input the training set into the convolutional neural network model for training; Input the test set into the trained convolutional neural network model, extract and classify the defect features in the feature image, and determine the ultrasonic defect type. 2. The method according to claim 1, wherein the defect types of the current transformer include: air bubbles and cracks. 3. The method according to claim 1, wherein the feature image generated using the Graham angle field algorithm is a two-dimensional color image. 4. The method according to claim 1, wherein the trained convolutional neural network model includes 3 convolutional layers, 2 pooling layers and 2 fully connected layers; wherein the volume of the 3 convolutional layers The size of the product kernel is 2×2, and the number of convolution kernels is 16, 32, and 48 respectively; the pooling window size of the 2 pooling layers is 2×2, and the stride is 2; the neurons of the 2 fully connected layers The numbers are 240 and 120; 10 kinds of classification results are output through the softmax classifier. 5. An ultrasonic defect recognition device of a convolutional neural network, characterized in that it comprises: The waveform data acquisition module is used to detect the current transformer by using the ultrasonic detection device, and collect the waveform data corresponding to the ultrasonic waves under different defects; Spectrum information extraction module, for applying Fourier transform to extract the spectrum information of waveform data; A feature image generation module, configured to use the Graham angle field algorithm to generate feature images corresponding to different defect categories from spectral information; A model training module, used to divide the generated feature images into a training set and a test set, and input the training set to a convolutional neural network model for training; The defect type determination module is used to input the test set into the trained convolutional neural network model, extract and classify the defect features in the feature image, and determine the ultrasonic defect type. 6. The device according to claim 5, wherein the defect types of the current transformer include: air bubbles and cracks. 7. The device according to claim 5, wherein the feature image generated using the Graham angle field algorithm is a two-dimensional color image. 8. The device according to claim 5, wherein the trained convolutional neural network model includes 3 convolutional layers, 2 pooling layers and 2 fully connected layers; wherein the volume of the 3 convolutional layers The size of the product kernel is 2×2, and the number of convolution kernels is 16, 32, and 48 respectively; the pooling window size of the 2 pooling layers is 2×2, and the stride is 2; the neurons of the 2 fully connected layers The numbers are 240 and 120; 10 kinds of classification results are output through the softmax classifier. 9. A computer-readable storage medium, wherein the storage medium stores a computer program, and the computer program is used to execute the method according to any one of claims 1-4. 10. An electronic device, characterized in that the electronic device comprises: processor; memory for storing said processor-executable instructions; The processor is configured to read the executable instruction from the memory, and execute the instruction to implement the method described in any one of claims 1-4.