Cable defect positioning and intelligent identification system and method based on broadband impedance spectroscopy

Abstract

The invention belongs to the technical field of buildings, and particularly relates to a cable defect positioning and intelligent identification system and method based on a broadband impedance spectrum. The system comprises a vector network analyzer which can inject a sweep frequency excitation signal into a target cable to be detected, a high-precision receiver which is used for collecting signal reflection / transmission response so as to obtain full-band complex impedance spectrum data, and computer equipment which receives test data. The positioning method based on Nuttall-Kai ser mixed window and cepstrum coupling does not need to depend on an intact cable parameter database, is high in anti-interference capability, can accurately identify defect positions, and is high in positioning precision; the one-dimensional impedance spectrum is converted into the two-dimensional image through the GAF, the automatic identification of the defect type is realized by combining the residual neural network (ResNet), the manual intervention is reduced, and the automation degree is high; the deep feature learning capability of ResNet is utilized, and a CBAM attention module is combined, so that the capture of a fine defect mode is enhanced, and the recognition accuracy is improved.

Description

Technical Field

[0001] This invention belongs to the field of printing wastewater technology and relates to a cable defect location and intelligent identification system and method based on broadband impedance spectrum. Background Technology

[0002] High-voltage and medium-voltage power cables, as key transmission and distribution carriers in modern power networks, directly impact grid security and power supply quality due to their operational stability and reliability. As cables age, various defects such as insulation aging, mechanical damage, joint deterioration, and water treeing / electrical treeing become unavoidable. Accurate identification and location of these potential defects are crucial for preventing faults and reducing power outage losses. Traditional methods based on time-domain reflectometry (TDR) or partial discharge detection sometimes suffer from insufficient sensitivity, limited location accuracy, or weak anti-interference capabilities. In recent years, broadband impedance spectroscopy analysis has emerged as a superior tool for cable condition diagnosis and defect detection. It provides a rich fingerprint reflecting impedance changes along the cable line, potential defect types (such as insulation damage, partial discharge sources, moisture intrusion, and geometric deformation), and their location characteristics, enabling the acquisition of impedance information between the cable conductor and shielding layer over a wide frequency range.

[0003] However, broadband impedance spectroscopy, as a one-dimensional complex sequence, contains highly complex modes and exhibits nonlinear coupling between features. Current methods for cable defect detection and location typically rely on manual judgment of collected data by operators, requiring significant engineering experience and lacking automation. Existing cable defect location methods based on frequency domain impedance spectroscopy have several limitations, including poor location accuracy, the need to pre-measure characteristic parameters of intact cables to establish a database, and susceptibility to interference affecting location accuracy. Furthermore, traditional identification methods based on manually designed features or shallow models often struggle to fully mine and utilize deep-level defect pattern information when processing WBIS data, exhibiting limited ability to distinguish subtle defects, necessitating the adoption of more powerful feature learning and classification techniques. Therefore, there is an urgent need to design a cable defect location and intelligent identification system and method based on broadband impedance spectroscopy that can overcome the above shortcomings.

[0004] To overcome the shortcomings of existing technologies, people have continuously explored and proposed various solutions. For example, a Chinese patent discloses an online method for locating defects in communication cables [Application No.: 202411109709.5]. This method includes the following steps: collecting relevant communication cable information within a jurisdictional area; obtaining dust data of relevant cable areas through a dust detector to establish a communication cable information database; setting a dust concentration threshold for partial discharge detection of cables; determining the dust concentration of the jurisdictional area in the communication cable information database; simultaneously collecting meteorological information of relevant areas; correlating the meteorological information and dust data judgment results within a time period to obtain the detection time point of the relevant numbered communication cable; obtaining the reference reflection intensity of the surface of the relevant communication cable through a drone and recording it into the communication cable information database; and determining the detection time point of the relevant numbered communication cable. However, this solution still has shortcomings in the measurement process, such as poor positioning accuracy, the need to measure the characteristic parameters of intact cables in advance to establish a database, and susceptibility to interference affecting positioning accuracy. Summary of the Invention

[0005] The purpose of this invention is to address the above-mentioned problems by providing a cable defect location and intelligent identification system and method based on broadband impedance spectrum.

[0006] To achieve the above objectives, the present invention adopts the following technical solutions:

[0007] A cable defect location and intelligent identification system based on broadband impedance spectrum includes a vector network analyzer that can inject a sweep frequency excitation signal into the target cable to be tested, a high-precision receiver for acquiring signal reflection / transmission response to obtain full-band complex impedance spectrum data, and a computer device for receiving test data. The vector network analyzer and the computer device are connected through a standard communication interface.

[0008] A method for cable defect location and intelligent identification system based on broadband impedance spectrum includes the following steps:

[0009] S1: The vector network analyzer tests the impedance spectrum of the target cable based on a preset test frequency range and sends the test results to a computer device;

[0010] S2: The computer equipment receives the impedance spectrum and determines the positioning function corresponding to the target cable through the preset Nuttall window;

[0011] S3: The computer equipment determines the positioning curve based on the positioning function, and determines the location of the defect in the target cable within the positioning curve;

[0012] S4: Next, the impedance spectrum will be calculated, and the Gram matrix (dot product or angle difference) will be calculated (or a short-time Fourier transform or wavelet transform time-frequency graph will be used to construct a two-dimensional image);

[0013] S5: Input the generated two-dimensional image into ResNet to identify the defect type;

[0014] S6: Complete the evaluation of defect location and identification of the target cable.

[0015] In the above-mentioned method for cable defect location and intelligent identification system based on broadband impedance spectrum, in step S1, the vector network analyzer injects a sweep frequency excitation signal into the target cable according to the preset broadband test frequency range of 1MHz–30MHz.

[0016] In the above-mentioned method for cable defect location and intelligent identification system based on broadband impedance spectrum, in step S1, after injecting a frequency sweep excitation signal into the target cable, its reflection / transmission response is synchronously acquired by a high-precision receiver to obtain full-band complex impedance spectrum data (including amplitude and phase information).

[0017] In the above-mentioned method for cable defect location and intelligent identification system based on broadband impedance spectrum, in step S1, after acquiring full-band complex impedance spectrum data, the vector network analyzer transmits the test dataset containing impedance spectrum characteristic parameters to the computer device in real time through a standard communication interface.

[0018] In the above-mentioned method for cable defect location and intelligent identification system based on broadband impedance spectrum, in step S2, the computer device receives full-band complex impedance spectrum data (including real and imaginary matrices) from the vector network analyzer, and performs sidelobe suppression and spectral leakage optimization on the impedance spectrum through a preset Nuttall-Kaiser hybrid window and windowing function.

[0019] In the above-mentioned method for cable defect location and intelligent identification system based on broadband impedance spectrum, in step S3, the computer device determines the location curve according to the location function, and identifies the defect of the target cable in the location curve. The location curve graph generated by the computer device has the horizontal axis representing the distance from the beginning and the vertical axis representing the amplitude. Therefore, the location curve graph can accurately determine the corresponding position of each peak point on the target cable, thereby determining the end position of the target cable and the location of the defect on the target cable.

[0020] In the above-mentioned cable defect location and intelligent identification system and method based on broadband impedance spectrum, in step S4, Gramian Angular Field (GAF) transformation is used to convert one-dimensional time series or spectral data into a two-dimensional image suitable for deep learning model processing. This method encodes the frequency sequence into an image structure by preserving the time dependency.

[0021] In the above-mentioned method for cable defect location and intelligent identification system based on broadband impedance spectrum, the specific process of encoding the frequency sequence into an image structure is as follows:

[0022] 1) Normalizing the data: The broadband impedance spectrum data Z = {z1, z2, ..., zn} (real part, imaginary part, or amplitude) first needs to be normalized to the interval [-1, 1]:

[0023] After normalization, each data point zi is located in the interval [-1,1]. This step ensures that the data have the same scale, which is convenient for subsequent processing.

[0024] 2) Polar coordinate transformation: Transforming the normalized data... To convert to polar coordinates, the conversion method is as follows: Angle: Use the inverse cosine function to map each normalized value to an angle:

[0025] because Therefore φ i ∈[0, π], thus each data point is converted into an angle, which reflects the position of the data point in the normalized sequence. Radius: The radius is scaled proportionally to the frequency index (or time index) to ensure that the radius increases with increasing frequency.

[0026] Among them, f i This is the frequency index (i.e., the sequence number of the data point), N is the total number of samples, therefore, r i ∈[0,1], here, the introduction of the radius can reflect the time order (or frequency order) of the data points, because the radius is monotonically increasing; through polar coordinate transformation, the one-dimensional time series is converted into two-dimensional polar coordinate points with angle and radius, so that each point in the original sequence can be represented on the unit circle (because ri is normalized to [0,1], so all points are inside the unit circle).

[0027] After representing the data in polar coordinates, we can use the Gramian matrix to capture the temporal (or frequency) correlations between points. The Gramian matrix has two forms: GASF and GADF. (a) Gramian Angular Summation Field (GASF): GASF constructs the matrix by calculating the cosine of the sum of the angles between two points. Its formula is:

[0028]

[0029] The expanded element expression is: GASF i，j =cos(φ i +φj )

[0030] Using trigonometric identities, the above equation can be written as:

[0031] GADF i，j =sin(φ) i cos(φ) j )-cos(φ i sin(φ) j )

[0032] Similarly, this is used here. The GADF matrix captures the differences between different frequency points by using angle differences;

[0033] Normalized data is mapped to angle space using an inverse cosine function, converting the scale information (normalized value) of the original sequence into angles. At the same time, the time (or frequency) order is represented by the radius. By calculating the sum of angles (GASF) or the difference of angles (GADF) between any two points, an n×n matrix is ​​formed. This matrix can be understood as the Gramian matrix image generated by the correlation between any two time points (or frequency points) in the original sequence. It has a clear geometric meaning, while other positions reflect the correlation between different points.

[0034] In the aforementioned method for cable defect localization and intelligent identification system based on broadband impedance spectrum, in step S5, during the image normalization stage, the input Gram matrix or time-frequency graph is first uniformly scaled to a standard size of 224×224 pixels, and a normalization operation is performed (e.g., subtracting the mean and dividing by the standard deviation). To adapt to the limited resource constraints of industrial embedded devices, ResNet-18 is used as the backbone network. This network is initialized using ImageNet pre-trained weights and contains four residual stages (Stage 1 to Stage 4), each stage consisting of two residual blocks, for a total of eight blocks. A CBAM (Convolutional Block Attention Module) attention module is embedded after the last residual block of Stage 3 and Stage 4.

[0035]

[0036] Simultaneously, data augmentation optimization was adopted to introduce two new strategies to address the characteristics of the Gram matrix: channel shuffling (randomly exchanging the real and imaginary parts of the impedance spectrum) to simulate channel interference, and frequency domain random masking (blocking 5%-15% of the frequency band) to simulate measurement noise. Meanwhile, the original enhancement method, Gaussian noise injection (simulating industrial environment interference), was maintained.

[0037] Compared with existing technologies, the advantages of this invention are:

[0038] 1. This invention is a cable defect location method based on broadband impedance spectrum, Nuttall window and cepstral coupling. By performing windowing processing on the impedance spectrum of the target cable and performing cepstral analysis, the location curve is determined, and the defect location on the target cable is identified, thereby solving the relevant shortcomings of existing cable defect location methods.

[0039] 2. This invention constructs a two-dimensional image by mapping sequence data to a polar coordinate system (encoding numerical information in terms of angle) and calculating the Gram matrix (dot product or angle difference). This transformation cleverly encodes the temporal dependence and numerical relationship (amplitude and phase) of the one-dimensional time series into the pixel space structure of the image, generating a visually interpretable pseudo-color image that retains the main time-frequency domain features of the original sequence. After combining GAF image processing to perform mode conversion on the BIS signal, it is input into a customized residual neural network (ResNet) architecture for deep feature learning and classification recognition, ultimately achieving automated and high-precision classification of cable defect types.

[0040] 3. This invention utilizes a Nuttall-Kaiser hybrid window and cepstral coupling positioning method, which does not rely on a database of intact cable parameters, exhibits strong anti-interference capabilities, and can accurately identify defect locations with high positioning accuracy. It converts the one-dimensional impedance spectrum into a two-dimensional image using Gram angle field (GAF) and combines it with a residual neural network (ResNet) to achieve automated defect type identification, reducing manual intervention and achieving a high degree of automation. By leveraging ResNet's deep feature learning capabilities and combining it with a CBAM attention module, it enhances the capture of subtle defect patterns, improving recognition accuracy. The invention employs a lightweight ResNet-18 architecture and data augmentation strategies, adapting to the resource constraints (low power consumption, small memory) of industrial embedded devices, balancing accuracy and practicality.

[0041] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of broadband impedance spectrum measurement.

[0043] Figure 2 A schematic diagram of the Nuttall-Kaiser hybrid window in the time and frequency domains;

[0044] Figure 3 This is a comparison chart of positioning curves based on measured cable data;

[0045] Figure 4 yes Figure 3 Enlarged diagram of point A in the middle.

[0046] Figure 5 yes Figure 3 Enlarged diagram of point B in the middle.

[0047] Figure 6 This is a schematic diagram of converting an impedance spectrum sequence into a two-dimensional Gram angle field image.

[0048] Figure 7 This is a schematic diagram of a residual neural network combined with an attention mechanism.

[0049] In the diagram: 1. Vector network analyzer; 2. High-precision receiver; 3. Computer equipment; 4. Target cable 100. Detailed Implementation

[0050] The present invention will be further described below with reference to the accompanying drawings.

[0051] Example 1

[0052] like Figure 1 As shown, a cable defect location and intelligent identification system based on broadband impedance spectrum includes a vector network analyzer 1 that can inject a sweep frequency excitation signal into the target cable 100 to be tested, a high-precision receiver 2 for acquiring signal reflection / transmission response to obtain full-band complex impedance spectrum data, and a computer device 3 for receiving test data. The vector network analyzer 1 and the computer device 3 are connected through a standard communication interface.

[0053] In this embodiment, the system consists of a vector network analyzer 1, a high-precision receiver 2, and a computer device 3. The three work together through a standard communication interface (such as GPIB, LAN, or USB). The vector network analyzer 1 acts as a signal source, injecting a 1MHz–30MHz sweep excitation signal into the target cable 100 to cover a wide frequency band and capture rich defect features. The high-precision receiver 2 works synchronously with the vector network analyzer, acquiring the cable's reflection / transmission response and extracting full-band complex impedance spectrum data (including amplitude, phase, real part, and imaginary part) to ensure data integrity. The computer device 3 receives and stores data, executes core algorithms such as signal processing (positioning), image conversion (GAF), and deep learning recognition, and is the "brain" of the system.

[0054] The vector network analyzer and receiver ensure consistent signal transmission and reception timing through a hardware synchronization mechanism. The computer equipment acquires data in real time through the communication interface, forming a closed loop of "excitation-acquisition-processing". The integrated design realizes full-process automation from signal injection to defect identification, solving the problems of traditional equipment being scattered and data transmission delay.

[0055] Example 2

[0056] Combination Figure 1-7As shown, a method for cable defect location and intelligent identification system based on broadband impedance spectrum includes the following steps:

[0057] S1: Vector network analyzer 1 tests the impedance spectrum of target cable 100 based on a preset test frequency range and sends the test results to computer device 3;

[0058] S2: Computer device 3 receives the impedance spectrum and determines the positioning function corresponding to the target cable through the preset Nuttall window;

[0059] S3: Computer device 3 determines the positioning curve based on the positioning function, and determines the defect location of the target cable in the positioning curve;

[0060] S4: Next, the impedance spectrum will be calculated, and the Gram matrix (dot product or angle difference) will be calculated (or a short-time Fourier transform or wavelet transform time-frequency graph will be used to construct a two-dimensional image);

[0061] S5: Input the generated two-dimensional image into ResNet to identify the defect type;

[0062] S6: Complete the evaluation of defect location and identification of target cable 100.

[0063] The Vector Network Analyzer (VNA) injects a sweep excitation signal into the target cable 100 according to a preset wideband test frequency range (e.g., 1MHz–30MHz), and synchronously acquires its reflection / transmission response through a high-precision receiver 2, thereby obtaining full-band complex impedance spectrum data (including amplitude and phase information); subsequently, the instrument transmits the test dataset containing impedance spectrum characteristic parameters to the computer device 3 in real time through a standard communication interface (e.g., GPIB, LAN, or USB);

[0064] Computer device 3 receives full-band complex impedance spectrum data (including real and imaginary matrices) from vector network analyzer 1 (VNA), and performs sidelobe suppression and spectral leakage optimization on the impedance spectrum through a preset Nuttall-Kaiser hybrid window and windowing function.

[0065] The Nuttall window is renowned for its extremely high sidelobe attenuation performance. Unlike the Blackman window, the Nuttall window uses more cosine series to achieve better spectral characteristics. Its design goal is to balance main lobe width and sidelobe attenuation, with a particular emphasis on extremely low sidelobe levels. Its symmetrical window form is defined as follows:

[0066]

[0067] Nuttall windows have several coefficient versions, but the most common is the four-coefficient combination form (sometimes referred to as the standard form of the Nuttall window).

[0068]

[0069] The different parameter coefficients represent different meanings and physical significance. a0 represents the fundamental frequency energy ratio control, a1 is the main lobe sharpening coefficient, a2 is the side lobe suppression factor, and a3 is the high-frequency leakage suppression term. The Nuttall window is a window function that sacrifices frequency resolution for excellent anti-spectral leakage performance (extremely low sidelobes), while the Kaiser window achieves precise control of spectral characteristics through a zero-order modified Bessel function. By adjusting a single parameter β, a smooth trade-off can be made between the main lobe width and the side lobe attenuation level. Its core formula is:

[0070]

[0071] I0 is the Modified Bessel Function of the First Kind (order zero), which is the core mathematical function defining the shape of the Kaiser window. The main lobe control parameter is β. By introducing the parameter β and a centrally symmetric expression, the Kaiser window cleverly utilizes the Modified Bessel Function I0 to provide the ability to make continuous and flexible trade-offs between the main lobe width (affecting the transition band) and the side lobe attenuation (affecting the stopband suppression).

[0072] Combining the advantages of both window functions, a Nuttall-Kaiser hybrid window was adopted to process the impedance.

[0073]

[0074] When analyzing spectra containing weak signal components, or when weak signals exist alongside strong signals, the Nuttall-Kaiser window can significantly reduce the "drowning" effect (spectral leakage) of weak signals by strong signal sidelobes, allowing for clearer differentiation of weak signals. While maintaining sidelobe suppression performance, it provides continuously adjustable degrees of freedom, allowing users to fine-tune the window near its designed maximum sidelobe suppression level. This optimizes the balance between frequency resolution (main lobe width) and sidelobe suppression (leakage control) to meet the specific needs of particular application scenarios. This makes it extremely valuable in precision spectrum analysis applications requiring extremely low spectral leakage and highly flexible control.

[0075] The computer equipment determines the positioning curve based on the positioning function, and identifies the defects in the target cable within the positioning curve. The positioning curve graph generated by the computer equipment has the horizontal axis representing the distance from the beginning to the end and the vertical axis representing the amplitude. Therefore, the positioning curve graph can accurately determine the corresponding position of each peak point on the target cable, thereby determining the end position of the target cable and the location of the defects on the target cable.

[0076] To transform one-dimensional time series or spectral data into two-dimensional images suitable for deep learning models, the Gramian Angular Field (GAF) transform is used. This method encodes the frequency sequence into an image structure by preserving time dependencies. The specific process is as follows:

[0077] 1) Normalizing the data: The broadband impedance spectrum data Z = {z1, z2, ..., zn} (real part, imaginary part, or amplitude) first needs to be normalized to the interval [-1, 1]:

[0078] After normalization, each data point zi is located in the interval [-1,1]. This step ensures that the data have the same scale, which is convenient for subsequent processing.

[0079] 2) Polar coordinate transformation: Transforming the normalized data... To convert to polar coordinates, the conversion method is as follows: Angle: Use the inverse cosine function to map each normalized value to an angle:

[0080] because Therefore φ i ∈[0, π], thus each data point is converted into an angle, which reflects the position of the data point in the normalized sequence. Radius: The radius is scaled proportionally to the frequency index (or time index) to ensure that the radius increases with increasing frequency.

[0081] Among them, f i This is the frequency index (i.e., the sequence number of the data point), N is the total number of samples, therefore, r i ∈[0,1], here, the introduction of the radius can reflect the time order (or frequency order) of the data points, because the radius is monotonically increasing; through polar coordinate transformation, the one-dimensional time series is converted into two-dimensional polar coordinate points with angle and radius, so that each point in the original sequence can be represented on the unit circle (because ri is normalized to [0,1], so all points are inside the unit circle).

[0082] After representing the data in polar coordinates, we can use the Gramian matrix to capture the temporal (or frequency) correlations between points. The Gramian matrix has two forms: GASF and GADF. (a) Gramian Angular Summation Field (GASF): GASF constructs the matrix by calculating the cosine of the sum of the angles between two points. Its formula is:

[0083]

[0084] The expanded element expression is: GASF i，j =cos(φ i +φ j )

[0085] Using trigonometric identities, the above equation can be written as:

[0086] GADF i,j =sin(φ) i cos(φ) j )-cos(φ i sin(φ) j )

[0087] Similarly, this is used here. The GADF matrix captures the differences between different frequency points by using angle differences;

[0088] Normalized data is mapped to angle space using an inverse cosine function, converting the scale information (normalized values) of the original sequence into angles. The time (or frequency) order is represented by the radius. An n×n matrix is ​​formed by calculating the sum of angles (GASF) or the difference of angles (GADF) between any two points. This matrix can be understood as the Gramian matrix generated from the correlation between any two time points (or frequency points) in the original sequence. The image has a clear geometric meaning; for example, the main diagonal elements of the GASF matrix are cos(2φi), while other positions reflect the correlation between different points. This representation method converts a one-dimensional time series into a two-dimensional image, preserving the complete information of the original sequence and possessing advantages such as translation invariance.

[0089] The generated two-dimensional image is input into ResNet to achieve defect type identification;

[0090] In the image normalization stage, the input Gram matrix or time-frequency image is first uniformly scaled to a standard size of 224×224 pixels, and a normalization operation (e.g., subtracting the mean and dividing by the standard deviation) is performed to ensure that all image data have consistent size and distribution characteristics before entering the network. This step effectively eliminates input noise and size differences, thereby improving the stability of subsequent processing. To adapt to the limited resource constraints of industrial embedded devices, ResNet-18 is used as the backbone network. This network is initialized using ImageNet pre-trained weights and contains four residual stages (Stage 1 to Stage 4), each stage consisting of two residual blocks, for a total of eight blocks. Its lightweight design (only 5.2M parameters) significantly reduces computational overhead while maintaining high feature extraction capabilities, meeting the low power consumption and small memory requirements of embedded deployments. To enhance feature representation capabilities, CBAM (Convolutional Block Attention Module) attention modules are embedded after the last residual block of Stage 3 and Stage 4.

[0091]

[0092] Simultaneously, data augmentation optimization introduces two new strategies tailored to the characteristics of the Gram matrix: channel shuffling (randomly swapping the real and imaginary parts of the impedance spectrum) to simulate channel interference, and frequency domain random masking (blocking 5%-15% of the frequency band) to simulate measurement noise. Meanwhile, the original augmentation method, Gaussian noise injection (simulating industrial environment interference), is maintained. Combined with the aforementioned attention simplification, this approach effectively adapts to the resource constraints of industrial embedded scenarios while maintaining high accuracy, achieving the advantage of lightweight deployment.

[0093] The working principle of this invention is:

[0094] The core working principle of this invention is the collaborative process of "wideband signal excitation - hybrid window processing positioning - image conversion - deep learning recognition":

[0095] 1. Signal excitation and acquisition: A wideband signal is injected into the vector network analyzer, and the receiver synchronously acquires the impedance spectrum to provide basic data for subsequent analysis;

[0096] 2. Defect location: The Nuttall-Kaiser hybrid window is used to suppress spectral leakage, and cepstral analysis is used to generate a location curve. The peak coordinates are then used to determine the defect location.

[0097] 3. Feature transformation and recognition: GAF transformation converts one-dimensional impedance spectrum into two-dimensional image, preserving deep features; ResNet-18 combined with CBAM attention module extracts image features, and data augmentation enhances anti-interference ability, ultimately achieving intelligent classification of defect types; the whole process requires no manual intervention, has a high degree of automation, and is adapted to industrial scenarios through lightweight design, solving the problems of low accuracy, reliance on experience, and difficulty in deployment of traditional methods.

[0098] The specific embodiments described herein are merely illustrative examples of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention.

[0099] Although this document frequently uses terms such as vector network analyzer 1, high-precision receiver 2, computer equipment 3, and target cable 100, the possibility of using other terms is not excluded. These terms are used merely for the convenience of describing and explaining the essence of this invention, and interpreting them as any additional limitation would contradict the spirit of this invention.

Claims

1. A cable defect location and intelligent identification system based on broadband impedance spectrum, characterized in that, The device includes a vector network analyzer (1) that can inject a sweep excitation signal into the target cable (100) to be tested, a high-precision receiver (2) for acquiring signal reflection / transmission response to obtain full-band complex impedance spectrum data, and a computer device (3) for receiving test data. The vector network analyzer (1) and the computer device (3) are connected through a standard communication interface.

2. A method for cable defect location and intelligent identification system based on broadband impedance spectrum, characterized in that, Includes the following steps: S1: The vector network analyzer (1) tests the impedance spectrum of the target cable (100) based on the preset test frequency range and sends the test results to the computer device (3); S2: The computer device (3) receives the impedance spectrum and determines the positioning function corresponding to the target cable (100) through the preset Nuttall window; S3: Computer equipment (3) determines the positioning curve according to the positioning function, and determines the defect location of the target cable (100) in the positioning curve; S4: Next, the impedance spectrum will be calculated, and the Gram matrix (dot product or angle difference) will be calculated (or a short-time Fourier transform or wavelet transform time-frequency graph will be used to construct a two-dimensional image); S5: Input the generated two-dimensional image into ResNet to identify the defect type; S6: Complete the evaluation of defect location and identification of the target cable (100).

3. The method for cable defect location and intelligent identification system based on broadband impedance spectrum according to claim 2, characterized in that, In step S1 above, the vector network analyzer (1) injects a sweep frequency excitation signal into the target cable according to the preset wideband test frequency range of 1MHz–30MHz.

4. The method for cable defect location and intelligent identification system based on broadband impedance spectrum according to claim 3, characterized in that, In step S1 above, after injecting a sweep excitation signal into the target cable, its reflection / transmission response is synchronously acquired by a high-precision receiver (2) to obtain full-band complex impedance spectrum data (including amplitude and phase information).

5. The method for cable defect location and intelligent identification system based on broadband impedance spectrum according to claim 4, characterized in that, In step S1 above, after acquiring the full-band complex impedance spectrum data, the vector network analyzer (1) transmits the test dataset containing impedance spectrum characteristic parameters to the computer device (3) in real time through the standard communication interface.

6. A method for cable defect location and intelligent identification system based on broadband impedance spectrum according to claim 4 or 5, characterized in that, In step S2 above, the computer device (3) receives full-band complex impedance spectrum data (including real and imaginary matrices) from the vector network analyzer (1) and performs sidelobe suppression and spectral leakage optimization on the impedance spectrum through a preset Nuttall-Kaiser hybrid window and window function.

7. The method for cable defect location and intelligent identification system based on broadband impedance spectrum according to claim 2, characterized in that, In step S3 above, the computer device (3) determines the positioning curve according to the positioning function, and determines the defect of the target cable in the positioning curve. The positioning curve diagram made by the computer device (3) has the horizontal axis representing the distance from the beginning and the vertical axis representing the amplitude. Therefore, the corresponding position of each peak point on the target cable can be accurately determined through the positioning curve diagram, thereby determining the end position of the target cable and the location of the defect on the target cable.

8. The method for cable defect location and intelligent identification system based on broadband impedance spectrum according to claim 2, characterized in that, In step S4 above, the Gramian Angular Field (GAF) transform is used to convert one-dimensional time series or spectral data into a two-dimensional image suitable for deep learning models. This method encodes the frequency sequence into an image structure by preserving the time dependencies.

9. The method for cable defect location and intelligent identification system based on broadband impedance spectrum according to claim 8, characterized in that, The specific process of encoding frequency sequences into image structures is as follows: 1) Normalizing the data: The broadband impedance spectrum data Z = {z1, z2, ..., zn} (real part, imaginary part, or amplitude) first needs to be normalized to the interval [-1, 1]: After normalization, each data point zi is located in the interval [-1,1]. This step ensures that the data have the same scale, which is convenient for subsequent processing. 2) Polar coordinate transformation: Transforming the normalized data... To convert to polar coordinates, the conversion method is as follows: Angle: Use the inverse cosine function to map each normalized value to an angle: because Therefore φ i ∈[0, π], thus each data point is converted into an angle, which reflects the position of the data point in the normalized sequence. Radius: The radius is scaled proportionally to the frequency index (or time index) to ensure that the radius increases with increasing frequency. Among them, f i This is the frequency index (i.e., the sequence number of the data point), N is the total number of samples, therefore, r i ∈[0,1], here, the introduction of the radius can reflect the time order (or frequency order) of the data points, because the radius is monotonically increasing; through polar coordinate transformation, the one-dimensional time series is converted into two-dimensional polar coordinate points with angle and radius, so that each point in the original sequence can be represented on the unit circle (because ri is normalized to [0,1], so all points are inside the unit circle). After representing the data in polar coordinates, we can use the Gramian matrix to capture the temporal (or frequency) correlations between points. The Gramian matrix has two forms: GASF and GADF. (a) Gramian Angular Summation Field (GASF): GASF constructs the matrix by calculating the cosine of the sum of the angles between two points. Its formula is: The expanded element expression is: GASF i，j =cos(φ i +φ j ) Using trigonometric identities, the above equation can be written as: GADF i，j =sin(φ i )cos(φ j )-cos(φ i )sin(φ j ) Similarly, this is used here. The GADF matrix captures the differences between different frequency points by using angle differences; Normalized data is mapped to angle space using an inverse cosine function, converting the scale information (normalized value) of the original sequence into angles. At the same time, the time (or frequency) order is represented by the radius. By calculating the sum of angles (GASF) or the difference of angles (GADF) between any two points, an n×n matrix is ​​formed. This matrix can be understood as the Gramian matrix image generated by the correlation between any two time points (or frequency points) in the original sequence. It has a clear geometric meaning, while other positions reflect the correlation between different points.

10. The method for cable defect location and intelligent identification system based on broadband impedance spectrum according to claim 2, characterized in that, In step S5 above, during the image normalization stage, the input Gram matrix or time-frequency graph is first uniformly scaled to a standard size of 224×224 pixels, and a normalization operation is performed (e.g., subtracting the mean and dividing by the standard deviation). To adapt to the limited resource constraints of industrial embedded devices, ResNet-18 is used as the backbone network. This network is initialized using ImageNet pre-trained weights and contains four residual stages (Stage 1 to Stage 4), each stage consisting of two residual blocks, for a total of eight blocks. A CBAM (Convolutional Block Attention Module) is embedded after the last residual block in Stages 3 and 4. F′=F′ channel +F′ spatial Element-wise multiplication, σ: Sigmoid function Simultaneously, data augmentation optimization was adopted to introduce two new strategies to address the characteristics of the Gram matrix: channel shuffling (randomly exchanging the real and imaginary parts of the impedance spectrum) to simulate channel interference, and frequency domain random masking (blocking 5%-15% of the frequency band) to simulate measurement noise. Meanwhile, the original enhancement method, Gaussian noise injection (simulating industrial environment interference), was maintained.

Images