Online detection method for small and special motor winding defects based on pulsed eddy current and deep learning

Abstract

The invention discloses an online detection method for small and special motor winding defects based on pulsed eddy current and deep learning, and relates to the technical field of nondestructive testing, original pulsed eddy current detection data containing time domain signals and frequency domain features are collected through a multi-channel eddy current sensor array, and a Maxwell equation constraint network is used for denoising and feature enhancement of the data, so that the detection accuracy is improved. Generating a defect feature candidate set in combination with a multi-scale defect generative adversarial network to obtain target eddy current data, applying a chaos frequency modulation pulse sequence to the target eddy current data as an excitation signal, and collecting a response signal by using a phase synchronization frequency locking technology to obtain enhanced eddy current data; the method comprises the following steps: extracting a physical feature of a sensor, extracting a corresponding electromagnetic feature vector containing the physical feature and a deep learning feature, comparing the electromagnetic feature vector with a preset defect feature library, and determining a defect detection result by combining dynamic excitation response difference. And constructing a dynamically updated defect feature library and combining with a fine tuning network.

Description

Technical Field

[0001] The present invention relates to the technical field of nondestructive testing, and more specifically, to an online detection method for micro-special motor winding defects based on pulsed eddy current and deep learning. Background Art

[0002] Due to its advantages such as small volume and low power, micro-special motors are widely used in many fields. The detection of their winding defects is crucial. Existing detection methods have many drawbacks and are difficult to meet the actual needs. With the development of technology, people have begun to explore more advanced detection means. The pulsed eddy current detection technology, due to its unique principle, has brought new means for the detection of motor winding defects. At the same time, deep learning technology has demonstrated powerful capabilities in fields such as image recognition and signal processing. Combining the pulsed eddy current technology with deep learning for the online detection of micro-special motor winding defects has become a current research hotspot. However, this technology still faces many challenges in practical applications. The original pulsed eddy current detection data is greatly affected by noise. How to effectively denoise and enhance useful features is a major problem. The extraction and recognition of defect features are not accurate enough and it is difficult to comprehensively and accurately reflect the true situation of winding defects. In addition, the adaptability of the detection model is poor. Facing different motor models and operating conditions, the detection effect will decrease significantly, and the reliability of the detection results also needs to be improved, and there are easy cases of misjudgment and missed judgment.

[0003] Existing methods for detecting micro-special motor winding defects have problems such as inaccurate detection, strong noise interference, and difficult feature extraction. Summary of the Invention

[0004] In order to overcome the problems of inaccurate detection, strong noise interference, and difficult feature extraction existing in the existing methods for detecting micro-special motor winding defects, the present invention discloses an online detection method for micro-special motor winding defects based on pulsed eddy current and deep learning, which can effectively solve the above technical problems.

[0005] To solve the above technical problems, the technical solution of the present invention is as follows:

[0006] An online detection method for micro-special motor winding defects based on pulsed eddy current and deep learning, comprising the following steps:

[0007] Obtain the original pulsed eddy current detection data during the operation of the micro-special motor. The original pulsed eddy current detection data is collected by a multi-channel eddy current sensor array and contains multi-dimensional data of time-domain signals and frequency-domain features;

[0008] Based on a preset Maxwell equation constrained network, denoise and enhance the features of the original pulsed eddy current detection data, and combine a multi-scale defect generative adversarial network to generate a candidate set of defect features to obtain target eddy current data;

[0009] Apply a chaotic frequency modulation pulse sequence to the target eddy current data as an excitation signal, and use the phase synchronization frequency locking technology to collect the response signal to obtain enhanced eddy current data containing dynamic characteristics;

[0010] Extract the electromagnetic feature vector corresponding to the enhanced eddy current data, and the electromagnetic feature vector includes physical features constrained by Maxwell's equations and multi-dimensional features extracted by deep learning;

[0011] Compare the electromagnetic feature vector with a preset defect feature library, and combine the dynamic excitation response differences to determine the defect detection result of the micro-special motor winding.

[0012] Preferably, the method for obtaining the target eddy current data by denoising and feature enhancement of the original pulsed eddy current detection data based on a preset Maxwell equation constraint network and generating a defect feature candidate set by combining a multi-scale defect generative adversarial network specifically includes:

[0013] Input the original pulsed eddy current detection data into the Maxwell equation constraint network, and optimize the network output through a loss function embedded with electromagnetic equations to achieve data denoising and physical feature extraction;

[0014] Input the denoised data into a multi-scale defect generative adversarial network to generate defect feature candidate sets at different scales;

[0015] Based on a preset feature screening rule, screen out the target eddy current data from the candidate sets.

[0016] Preferably, the method for applying a chaotic frequency modulation pulse sequence to the target eddy current data as an excitation signal and using the phase synchronization frequency locking technology to collect the response signal to obtain enhanced eddy current data containing dynamic characteristics specifically includes:

[0017] Generate a chaotic frequency modulation pulse sequence with ergodicity and non-periodicity, whose frequency, amplitude and pulse width change dynamically with the chaotic mapping;

[0018] Load the chaotic frequency modulation pulse sequence to the excitation end of the eddy current sensor, and at the same time use the phase synchronization frequency locking technology to lock the phase relationship between the excitation and response signals;

[0019] Collect and record the eddy current response signal after excitation, and construct enhanced eddy current data containing dynamic characteristics.

[0020] Preferably, the method for extracting the electromagnetic feature vector corresponding to the enhanced eddy current data, and the electromagnetic feature vector includes physical features constrained by Maxwell's equations and multi-dimensional features extracted by deep learning specifically includes:

[0021] Calculate the physical features of the magnetic field intensity and eddy current density of the enhanced eddy current data through the Maxwell equation constraint network;

[0022] Input the enhanced eddy current data into a convolutional neural network to extract multi-dimensional deep learning features of time-domain waveforms and frequency-domain spectral lines;

[0023] Fuse physical features and deep learning features to form an electromagnetic feature vector.

[0024] Preferably, the method of comparing the electromagnetic feature vector with a preset defect feature library and combining the dynamic excitation response difference to determine the defect detection result of the micro-special motor winding specifically includes:

[0025] Calculate the cosine similarity between the electromagnetic feature vector and the standard feature in the defect feature library to obtain a static matching score;

[0026] Analyze the response signal difference under chaotic dynamic excitation and calculate the dynamic feature deviation;

[0027] Based on the weighted sum of the static matching score and the dynamic feature deviation, compare it with a preset threshold to determine whether there is a defect and the type of defect in the micro-special motor winding.

[0028] Preferably, the multi-channel eddy current sensor array is arranged axially and circumferentially on the micro-special motor winding, and there is partial overlap in the detection areas of each sensor. The method further includes:

[0029] Determine the detection weight of each eddy current sensor, and the detection weight is calculated based on the electromagnetic coupling degree between the sensor position and the winding structure;

[0030] Perform weighted fusion on the data collected by each sensor to generate global eddy current detection data.

[0031] Preferably, the method further includes:

[0032] Construct a dynamically updated defect feature library based on historical detection data and defect types, and the defect feature library is optimized in real time through an incremental learning algorithm;

[0033] After every N detections, use the new detection data to jointly fine-tune the Maxwell equation constraint network and the multi-scale defect generative adversarial network.

[0034] Preferably, an electronic device includes: a memory and at least one processor. Instructions are stored in the memory, and at least one of the processors calls the instructions in the memory so that the device executes each step of the detection method as described above.

[0035] Preferably, a computer-readable storage medium stores instructions thereon, and when the instructions are executed by a processor, each step of the detection method as described above is implemented.

[0036] Preferably, a computer program product includes instructions that, when executed, cause the steps of the detection method described above to be performed.

[0037] Compared with the prior art, the beneficial effects of the present invention are as follows: It is difficult for existing methods to accurately detect the defects of the micro-special motor windings. However, the present invention combines pulsed eddy current technology with deep learning to more accurately identify defects. The multi-channel eddy current sensor array collects multi-dimensional data, including time-domain and frequency-domain information. After denoising and feature enhancement by the Maxwell equation-constrained network, the noise interference is effectively removed, and the useful features are highlighted. The multi-scale defect generative adversarial network generates candidate sets of defect features at different scales, filters out the target eddy current data, making the defect features more comprehensive and accurate, thus improving the detection accuracy. The original pulsed eddy current detection data is vulnerable to noise. The present invention optimizes the network output through a loss function embedded with electromagnetic equations to achieve data denoising and physical feature extraction, which can effectively suppress noise interference, ensure the data quality and the accuracy of feature extraction, and ensure the stability and reliability of the subsequent detection process. On the one hand, the Maxwell equation-constrained network is used to calculate physical features such as magnetic field strength and eddy current density. Based on electromagnetic theory, it can accurately reflect the physical essence of the winding defects. On the other hand, the data is input into a convolutional neural network to extract multi-dimensional deep learning features of the time-domain waveform and frequency-domain spectrum line, mine the potential laws of the data, and form an electromagnetic feature vector by fusing physical features and deep learning features, which more comprehensively characterizes the characteristics of the winding defects and provides a basis for accurately detecting the defects. The present invention considers the detection weights of different sensors, determines the weights according to the electromagnetic coupling degree between the sensor position and the winding structure, and performs data fusion to generate global eddy current detection data, making the detection result more representative. In addition, a dynamically updated defect feature library is constructed and optimized in real time through an incremental learning algorithm to adapt to different motor models and operating conditions and maintain good detection performance. When comparing the electromagnetic feature vector with the defect feature library, not only the static matching score is calculated, but also the dynamic feature deviation degree is analyzed, and the defect is determined by combining their weighted sum. Considering the differences in static features and dynamic responses comprehensively, the false positive and false negative rates are reduced, and the reliability of the detection result is ensured. The phase synchronization frequency-locking technology is used to collect response signals, improving the signal collection efficiency and accuracy, ensuring the rapid progress of the detection process. At the same time, the entire detection process has a high degree of automation and is realized through electronic devices, computer-readable storage media, and computer program products, improving the detection efficiency and meeting the rapid detection requirements of practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only exemplary, and for those of ordinary skill in the art, other implementation drawings can be obtained according to the provided drawings without creative efforts.

[0039] Figure 1 It is a step diagram of an on-line detection method for micro-special motor winding defects based on pulsed eddy current and deep learning. Specific implementation manners

[0040] The accompanying drawings are only for illustrative purposes and should not be construed as a limitation of this patent;

[0041] To better illustrate this embodiment, some components in the accompanying drawings will be omitted, enlarged or reduced, which do not represent the dimensions of the actual product;

[0042] For those skilled in the art, it is understandable that some well-known structures and their descriptions in the accompanying drawings may be omitted.

[0043] The technical solutions of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0044] Embodiment

[0045] An on-line detection method for micro-special motor winding defects based on pulsed eddy current and deep learning includes the following steps:

[0046] Obtain the original pulsed eddy current detection data during the operation of the micro-special motor. The original pulsed eddy current detection data is collected by a multi-channel eddy current sensor array and contains multi-dimensional data of time-domain signals and frequency-domain features;

[0047] Based on a preset Maxwell equation constraint network, denoise and feature enhancement are performed on the original pulsed eddy current detection data, and a defect feature candidate set is generated by combining a multi-scale defect generative adversarial network to obtain target eddy current data;

[0048] Apply a chaotic frequency modulation pulse sequence to the target eddy current data as an excitation signal, and use the phase synchronous frequency locking technology to collect the response signal to obtain enhanced eddy current data containing dynamic features;

[0049] Extract the electromagnetic feature vectors corresponding to the enhanced eddy current data. The electromagnetic feature vectors include physical features based on Maxwell equation constraints and multi-dimensional features extracted by deep learning;

[0050] Compare the electromagnetic feature vectors with a preset defect feature library, and combine the dynamic excitation response differences to determine the defect detection results of the micro-special motor winding.

[0051] The specific steps of performing denoise and feature enhancement on the original pulsed eddy current detection data based on a preset Maxwell equation constraint network, and generating a defect feature candidate set by combining a multi-scale defect generative adversarial network to obtain target eddy current data include:

[0052] Input the original pulsed eddy current detection data into the Maxwell equation-constrained network, and optimize the network output through the loss function embedded with electromagnetic equations to achieve data denoising and physical feature extraction;

[0053] Input the denoised data into the multi-scale defect generative adversarial network to generate a candidate set of defect features at different scales;

[0054] Based on the preset feature screening rules, screen out the target eddy current data from the candidate set.

[0055] Applying the chaotic frequency modulation pulse sequence to the target eddy current data as the excitation signal, and using the phase synchronization frequency locking technology to collect the response signal to obtain the enhanced eddy current data containing dynamic features specifically includes:

[0056] Generate a chaotic frequency modulation pulse sequence with ergodicity and non-periodicity, whose frequency, amplitude and pulse width change dynamically with the chaotic mapping;

[0057] Load the chaotic frequency modulation pulse sequence to the excitation end of the eddy current sensor, and at the same time use the phase synchronization frequency locking technology to lock the phase relationship between the excitation and response signals;

[0058] Collect and record the eddy current response signal after excitation, and construct the enhanced eddy current data containing dynamic features.

[0059] Extracting the electromagnetic feature vector corresponding to the enhanced eddy current data, where the electromagnetic feature vector includes physical features based on Maxwell equation constraints and multi-dimensional features extracted by deep learning specifically includes:

[0060] Calculate the physical features of the magnetic field intensity and eddy current density of the enhanced eddy current data through the Maxwell equation-constrained network;

[0061] Input the enhanced eddy current data into the convolutional neural network to extract multi-dimensional deep learning features of the time-domain waveform and frequency-domain spectrum line;

[0062] Fuse the physical features and deep learning features to form an electromagnetic feature vector.

[0063] Comparing the electromagnetic feature vector with the preset defect feature library, and combining the dynamic excitation response differences to determine the defect detection result of the micro-special motor winding specifically includes:

[0064] Calculate the cosine similarity between the electromagnetic feature vector and the standard features in the defect feature library to obtain the static matching score;

[0065] Analyze the response signal differences under chaotic dynamic excitation, and calculate the dynamic feature deviation;

[0066] Based on the weighted sum of the static matching score and the dynamic feature deviation, and comparing it with a preset threshold, it is determined whether there are defects in the micro-special motor winding and the type of defects.

[0067] The multi-channel eddy current sensor array is arranged axially and circumferentially on the micro-special motor winding, and there is partial overlap in the detection areas of each sensor. The method further includes:

[0068] Determine the detection weight of each eddy current sensor, and the detection weight is calculated based on the electromagnetic coupling degree between the sensor position and the winding structure;

[0069] Perform weighted fusion on the data collected by each sensor to generate global eddy current detection data.

[0070] The method further includes:

[0071] Construct a dynamically updated defect feature library based on historical detection data and defect types, and the defect feature library is optimized in real time through an incremental learning algorithm;

[0072] After every N detections are completed, use the new detection data to jointly fine-tune the Maxwell equation constraint network and the multi-scale defect generative adversarial network.

[0073] An electronic device includes: a memory and at least one processor. Instructions are stored in the memory, and at least one of the processors calls the instructions in the memory so that the device executes each step of the detection method as described above.

[0074] A computer-readable storage medium has instructions stored thereon, and when the instructions are executed by a processor, each step of the detection method as described above is implemented.

[0075] A computer program product includes instructions, and when the instructions are run, the steps of the detection method as described above are executed.

[0076] In a specific implementation, please refer to Figure 1 On the test bench where the micro-special motor is running, arrange each detection device and device to ensure that the micro-special motor is in a normal operating state, its rotational speed is stable at the set value, and arrange the multi-channel eddy current sensor array axially and circumferentially on the micro-special motor winding. There is partial overlap in the detection area of each sensor to ensure the comprehensiveness of the winding detection. For example, select an 8-channel eddy current sensor array, and each sensor is evenly distributed at different positions in the axial and circumferential directions of the winding.

[0077] Through the above multi-channel eddy current sensor array, the original pulsed eddy current detection data during the operation of the special micro-motor is collected in real time. These data include multi-dimensional data of time-domain signals and frequency-domain characteristics. The sampling rate of the time-domain signal is 1 MHz, and the frequency range of the frequency-domain characteristics is from 1 kHz to 100 kHz. The collected original data is preliminarily stored, and the data storage format can adopt common binary formats or ASCII code formats, etc.

[0078] Input the original pulsed eddy current detection data into a network constrained by the preset Maxwell equations. The structure of this network can adopt the CNN (Convolutional Neural Network) architecture, and a loss function embedded with electromagnetic equations, such as the mean square error loss function combined with the Maxwell equation terms.

[0079] Train the network. Using an optimization algorithm, such as the Adam optimizer, continuously adjust the network parameters to enable the network output to achieve data denoising and physical feature extraction, and obtain the denoised data. For example, set the number of training iterations to 1000 times and the learning rate to 0.001.

[0080] Input the denoised data into a multi-scale defect generative adversarial network (GAN). This GAN includes a generator and a discriminator. The generator is used to generate a candidate set of defect features at different scales, and the discriminator is used to distinguish real defect features from the generated defect features.

[0081] By training the GAN, enable the candidate set of defect features generated by the generator to simulate as realistically as possible the winding defect features at different scales, such as defect features like broken wires, short circuits, and inter-turn short circuits, as well as defect features of different sizes, shapes, and positions. During the training process, the network structures of both the generator and the discriminator can be set to the multi-layer perceptron (MLP) structure, and the number of training iterations is about 2000 times.

[0082] According to the preset feature screening rules, screen out the target eddy current data from the generated candidate set of defect features. The feature screening rules can be based on indicators such as the significance, stability, and correlation of the defect features. For example, set the significance threshold to 0.8, the stability threshold to 0.9, and the correlation threshold to 0.7. The candidate set of defect features that meet these threshold conditions is used as the target eddy current data.

[0083] Generate a chaotic frequency-modulated pulse sequence with ergodicity and non-periodicity. Its frequency, amplitude, and pulse width change dynamically with the chaotic mapping. For example, use the Logistic chaotic mapping to generate a chaotic sequence and map it to the frequency, amplitude, and pulse width parameters of the pulse sequence. The frequency range is from 5 kHz to 50 kHz, the amplitude range is from 1 V to 5 V, and the pulse width range is from 50 ns to 500 ns.

[0084] Load the chaotic frequency modulation pulse sequence to the excitation end of the eddy current sensor. At the same time, use the phase synchronization frequency locking technology to lock the phase relationship between the excitation and response signals. Through the phase synchronization frequency locking technology, it can be ensured that the collected response signal has the same phase as the excitation signal, improving the accuracy and stability of signal acquisition; collect and record the eddy current response signal after excitation, construct enhanced eddy current data containing dynamic characteristics, the acquisition time can be set from 10 ms to 100 ms, and the sampling rate of the acquired data is the same as that of the original data.

[0085] Calculate the physical characteristics of the magnetic field strength and eddy current density of the enhanced eddy current data through the Maxwell equation constrained network. Use Ampere's circuital law, Faraday's law of electromagnetic induction, etc. in the Maxwell equations to analyze and calculate the enhanced eddy current data, and obtain the physical characteristic values of the magnetic field strength and eddy current density.

[0086] Input the enhanced eddy current data into a convolutional neural network to extract multi-dimensional deep learning features of the time-domain waveform and frequency-domain spectral line. For example, the structure of the convolutional neural network can include an input layer, several convolutional layers, pooling layers, fully connected layers and an output layer. Through the trained network model, extract features from the enhanced eddy current data to obtain time-domain waveform features such as the amplitude, width, rise time, fall time, etc. of the waveform and frequency-domain spectral line features such as frequency components, amplitude spectra, phase spectra, etc.

[0087] Fuse the physical characteristics and deep learning features to form an electromagnetic feature vector. Combine the physical characteristics of the magnetic field strength and eddy current density calculated above and the deep learning features such as the extracted time-domain waveform and frequency-domain spectral line to form a complete electromagnetic feature vector for subsequent defect detection comparison.

[0088] Compare the extracted electromagnetic feature vector with a preset defect feature library. The defect feature library can be constructed based on historical detection data and defect types, and contains electromagnetic feature vector samples of different types and degrees of winding defects.

[0089] Calculate the cosine similarity between the electromagnetic feature vector and the standard feature in the defect feature library to obtain a static matching score. The calculation formula of the cosine similarity is: cosθ = (A·B) / (|A|·|B|), where A and B are the electromagnetic feature vector and the standard feature vector respectively, and θ is the angle between them. The range of the static matching score is between -1 and 1, and the larger the value, the higher the similarity.

[0090] Analyze the difference in the response signal under chaotic dynamic excitation, and calculate the dynamic feature deviation. By comparing the differences in dynamic characteristics between the current response signal and the response signal in the normal state, such as frequency response characteristics, amplitude change characteristics, phase change characteristics, etc., calculate the dynamic feature deviation, which can reflect the degree of change in the dynamic characteristics of the winding defect.

[0091] Based on the weighted sum of the static matching score and the dynamic feature deviation degree, compare it with a preset threshold to determine whether there are defects in the winding of the special micro-motor and the type of defects. For example, set the weight of the static matching score to 0.6, the weight of the dynamic feature deviation degree to 0.4, and the preset threshold to 0.7. If the weighted sum is greater than or equal to 0.7, it is determined that there are defects, and the type of defects is determined according to the comparison result with the features of different defect types in the defect feature library; if the weighted sum is less than 0.7, it is determined that the winding is normal.

[0092] Display the detection results (whether there are defects and the type of defects) on the human-machine interface of the electronic device so that the operator can timely understand the operating state of the winding of the special micro-motor. At the same time, store the detection results in the memory, and the storage format can be the same as the original data storage format.

[0093] The above detection method can be implemented by a specially designed electronic device, which includes a memory and at least one processor. Instructions are stored in the memory, and the processor calls the instructions in the memory to execute each step of the above detection method.

[0094] The detection method can also be written as a computer program product and stored on a computer-readable storage medium, such as a hard disk, an optical disc, a USB flash drive, etc. When the program is run, the computer is made to execute the steps of the above detection method.

[0095] The same or similar reference numerals correspond to the same or similar components;

[0096] The terms used to describe the positional relationship in the drawings are only for illustrative purposes and should not be construed as a limitation of this patent;

[0097] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, rather than limitations on the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to enumerate all the implementation manners here. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. An on-line detection method for winding defects of small and special motors based on pulsed eddy current and deep learning, characterized in that, Including the following steps: Obtain the original pulsed eddy current detection data during the operation of the special micro-motor. The original pulsed eddy current detection data is collected by a multi-channel eddy current sensor array and includes multi-dimensional data of time-domain signals and frequency-domain characteristics; Based on a preset Maxwell equation-constrained network, denoise and enhance the features of the original pulsed eddy current detection data, and combine a multi-scale defect generative adversarial network to generate a candidate set of defect features to obtain target eddy current data; Apply a chaotic frequency-modulated pulse sequence to the target eddy current data as an excitation signal, and use the phase synchronization frequency-locking technology to collect the response signal to obtain enhanced eddy current data containing dynamic features; Extract the electromagnetic feature vectors corresponding to the enhanced eddy current data. The electromagnetic feature vectors include physical features based on the constraints of Maxwell's equations and multi-dimensional features extracted by deep learning; Compare the electromagnetic feature vectors with a preset defect feature library, and combine the dynamic excitation response differences to determine the defect detection results of the special micro-motor winding.

2. The detection method according to claim 1, wherein The specific steps of denoising and enhancing the features of the original pulsed eddy current detection data based on a preset Maxwell equation-constrained network, and combining a multi-scale defect generative adversarial network to generate a candidate set of defect features to obtain target eddy current data include: Input the original pulsed eddy current detection data into the Maxwell equation-constrained network, and optimize the network output through a loss function embedded with electromagnetic equations to achieve data denoising and physical feature extraction; Input the denoised data into the multi-scale defect generative adversarial network to generate a candidate set of defect features at different scales; Based on a preset feature screening rule, screen out the target eddy current data from the candidate set.

3. The detection method according to claim 1, wherein The specific steps of applying a chaotic frequency-modulated pulse sequence to the target eddy current data as an excitation signal, and using the phase synchronization frequency-locking technology to collect the response signal to obtain enhanced eddy current data containing dynamic features include: Generate a chaotic frequency-modulated pulse sequence with ergodicity and non-periodicity, whose frequency, amplitude, and pulse width change dynamically with the chaotic mapping; Load the chaotic frequency-modulated pulse sequence to the excitation end of the eddy current sensor, and at the same time use the phase synchronization frequency-locking technology to lock the phase relationship between the excitation and response signals; Collect and record the eddy current response signal after excitation, and construct enhanced eddy current data containing dynamic features.

4. The detection method according to claim 1, wherein The specific steps of extracting the electromagnetic feature vectors corresponding to the enhanced eddy current data. The electromagnetic feature vectors include physical features based on the constraints of Maxwell's equations and multi-dimensional features extracted by deep learning include: Calculate the physical features of the magnetic field strength and eddy current density of the enhanced eddy current data through the Maxwell equation-constrained network; Input the enhanced eddy current data into a convolutional neural network to extract multi-dimensional deep learning features of the time-domain waveform and frequency-domain spectrum line; Fuse the physical features and deep learning features to form electromagnetic feature vectors.

5. The detection method according to claim 1, wherein The specific steps of comparing the electromagnetic feature vectors with a preset defect feature library, and combining the dynamic excitation response differences to determine the defect detection results of the special micro-motor winding include: Calculate the cosine similarity between the electromagnetic feature vectors and the standard features in the defect feature library to obtain a static matching score; Analyze the response signal differences under chaotic dynamic excitation and calculate the dynamic feature deviation; Based on the weighted sum of the static matching score and the dynamic feature deviation degree, compare it with a preset threshold to determine whether there are defects in the winding of the special micro-motor and the type of defects.

6. The detection method according to claim 1, wherein The multi-channel eddy current sensor array is arranged axially and circumferentially on the winding of the special micro-motor, and there is partial overlap in the detection areas of each sensor. The method further includes: Determine the detection weight of each eddy current sensor, and the detection weight is calculated based on the electromagnetic coupling degree between the sensor position and the winding structure; Perform weighted fusion on the data collected by each sensor to generate global eddy current detection data.

7. The detection method according to claim 1, wherein The method further includes: Construct a dynamically updated defect feature library based on historical detection data and defect types, and the defect feature library is optimized in real time through an incremental learning algorithm; After every N detections are completed, use the new detection data to jointly fine-tune the Maxwell equation constraint network and the multi-scale defect generative adversarial network.

8. An electronic device, characterized in that, Includes: A memory and at least one processor. Instructions are stored in the memory, and at least one of the processors calls the instructions in the memory so that the device executes each step of the detection method described in any one of claims 1-7.

9. A computer-readable storage medium having instructions stored thereon, characterized in that, When the instructions are executed by the processor, each step of the detection method described in any one of claims 1-7 is implemented.

10. A computer program product, characterized in that, The computer program product includes instructions, and when the instructions are run, the steps of the detection method described in any one of claims 1-7 are executed.

Images