Quantitative detection method of internal corrosion of cables based on terahertz

Abstract

The present invention relates to the field of cable detection technology, and in particular to a quantitative detection method for internal corrosion of cables based on terahertz. The method comprises the following steps: scanning the cable surface through a terahertz sensor and receiving a reflected signal to obtain terahertz detection signal data; performing terahertz signal preprocessing on the terahertz detection signal data to obtain terahertz preprocessing data; obtaining cable corrosion degree data, and integrating the cable corrosion degree data and the terahertz preprocessing data to obtain corrosion degree terahertz data of different corrosion degrees; constructing a deep learning model for internal corrosion detection of cables, training the model using corrosion degree terahertz data to obtain a corrosion degree terahertz recognition model; and using the corrosion degree terahertz recognition model to identify and obtain cable internal corrosion degree data. The present invention realizes end-to-end refined identification of the corrosion grade of steel wires inside cables divided according to specifications.

Description

Technical Field

[0001] The present invention relates to the technical field of cable detection, and in particular to a terahertz-based quantitative detection method for internal corrosion of cables. Background Art

[0002] Cable-supported bridges are widely used in long-span bridges due to their beautiful architectural shape and superior spanning capacity. At present, there are more than 450 cable-supported bridges with a span of more than 400 meters in the world. As one of the most important load-bearing components of such bridges, the health of the cable is crucial to the safe operation and scientific management of the bridge. However, under the coupling of adverse factors such as harsh service environment, complex loads, and improper maintenance, the cable is inevitably damaged, resulting in premature replacement of the cable. The main reason for cable breakage and premature retirement is the corrosion fatigue disease inside the cable. That is, after the cable sheath is damaged and cracked, moisture and air enter the cable, and the internal steel wire is corroded. Under the action of complex alternating loads, fatigue cracks form on the surface of the steel wire and continue to expand inward, resulting in brittle failure of the steel wire. Corrosion and wire breakage of the steel wires inside the cable are highly concealed and difficult to detect, and the damage caused by such defects is sudden and unpredictable. They are still the pain points and difficulties in bridge maintenance. Therefore, how to accurately identify and precisely locate internal defects has become a major problem that needs to be solved urgently. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a terahertz-based quantitative detection method for internal corrosion of cables to solve at least one of the above technical problems.

[0004] The present application provides a terahertz-based quantitative detection method for internal corrosion of cables, the method comprising:

[0005] S1, scanning the cable surface through a terahertz sensor and receiving a reflected signal to obtain terahertz detection signal data;

[0006] S2, performing terahertz signal preprocessing on the terahertz detection signal data to obtain terahertz preprocessing data;

[0007] S3, obtaining cable corrosion degree data, and integrating the cable corrosion degree data and terahertz preprocessing data to obtain corrosion degree terahertz data of different corrosion degrees;

[0008] S4. Construct a deep learning model based on the terahertz data of corrosion degree to obtain a terahertz recognition model of corrosion degree;

[0009] S5. Acquire new terahertz detection signal data, and use the corrosion degree terahertz recognition model to identify the new terahertz detection signal data and visualize them by color coding to obtain the internal corrosion degree data of the cable.

[0010] In the present invention, without peeling off the sheath, according to the different time domain characteristics of the THz signal collected in different corrosion areas, the different absorption peaks (different corrosion products have different absorption peaks), the recognition model of different corrosion levels is established according to the deep learning algorithm, and the rapid non-destructive detection of the corrosion level is realized. By scanning the surface of the cable with a terahertz sensor, the corrosion of the cable can be detected efficiently and quickly without the need for cumbersome manual inspection and disassembly operations. Through the deep learning model, the collected terahertz detection signal is integrated with the corrosion degree data to realize the quantitative analysis of the corrosion degree, so that the corrosion degree can be more accurately measured. The new terahertz detection signal can be input into the corrosion degree terahertz identification model in real time for real-time monitoring, and the color coding visualization results are used to make the operation more intuitive, so that the internal corrosion of the cable can be intuitively presented. The use of terahertz technology for detection does not require destructive detection methods for the cable, and the integrity of the cable can be maintained.

[0011] Optionally, the terahertz detection signal data includes first terahertz detection signal data and second terahertz detection signal data, and S1 is specifically:

[0012] Controlling the terahertz sensor to scan the cable surface through the preset first terahertz threshold angle data and receive the reflected signal to obtain the first terahertz detection signal data;

[0013] The terahertz sensor is controlled to scan the cable surface through the preset second terahertz threshold angle data and receive the reflected signal to obtain the second terahertz detection signal data, wherein the first terahertz threshold angle data is smaller than the second terahertz threshold angle data.

[0014] In the present invention, by controlling the terahertz sensor to scan using the first terahertz threshold angle and the second terahertz threshold angle respectively, two sets of terahertz detection signal data at different angles are obtained, which helps to comprehensively consider the signal characteristics at different angles and improve the accuracy and reliability of detection. The first terahertz detection signal data and the second terahertz detection signal data can be regarded as different signal modes, and the signal modes at these two different angles are fused to comprehensively utilize their information, so as to more comprehensively understand the situation of the cable. The terahertz signals at different angles will show different characteristics at the corrosion site. Through multi-angle detection and signal fusion, the degree of corrosion can be better detected and quantified, and the ability of corrosion detection is improved. Due to the use of multiple angles, the present invention helps to improve the detection sensitivity of the internal corrosion of the cable, and can more easily identify and locate the corrosion site. By comparing the terahertz detection signals at different angles, the possibility of false alarms can be reduced, the reliability of the system is enhanced, and the misjudgment rate is reduced.

[0015] Optionally, the step of generating the first terahertz threshold angle data includes the following steps:

[0016] Acquire corrosion product characteristic data, wherein the corrosion product characteristic data includes cable steel wire corrosion product refractive index data, cable steel wire corrosion product absorption coefficient data and cable steel wire corrosion product scattering characteristic data;

[0017] Acquire cable shape data, and construct a three-dimensional cable model according to the cable shape data to obtain cable three-dimensional model data;

[0018] The three-dimensional model data of the cable is annotated using the characteristic data of the corrosion products to obtain the annotated three-dimensional model data of the cable;

[0019] Generate random incident angle data, and perform terahertz incident simulation on the cable three-dimensional model annotation data according to the random incident angle to obtain terahertz incident simulation data;

[0020] Calculate the reflectivity according to the terahertz incident simulation data to obtain incident simulation reflectivity data;

[0021] The random incident angle data are sorted and screened according to the incident simulation reflectivity data to obtain the first terahertz threshold angle data.

[0022] In the present invention, by obtaining the corrosion product characteristic data and the cable shape data, combined with the terahertz incident simulation, the first terahertz threshold angle data can be quantitatively generated, which not only reflects the physical characteristics of the cable, but also is related to the propagation characteristics of the terahertz signal. The method of the present invention allows the generation of random incident angle data, and sorts and screens the incident simulation reflectivity data to obtain the most suitable first terahertz threshold angle data. According to the specific cable material and shape characteristics, the best terahertz signal incident angle is personalized to maximize the information about corrosion. By quantitatively generating the first terahertz threshold angle data, it is helpful to improve the accuracy of corrosion detection. Selecting a suitable incident angle can enhance the sensitivity of the signal and make it easier to detect the corrosion inside the cable. Through personalized angle setting, the misjudgment rate can be reduced, unnecessary false alarms or omissions can be reduced, and the reliability of the system can be improved.

[0023] Optionally, the step of generating random incidence angle data comprises the following steps:

[0024] Generate random angles through a preset random number generator to obtain random angle data;

[0025] The polarization state of the random angle data is randomly selected to obtain random incident angle data.

[0026] In the present invention, random angle data is generated by using a preset random number generator, which helps to introduce diversity and randomness, and different incident angles can be obtained in multiple tests, reducing deviations and improving the comprehensiveness and reliability of the test. Randomly generating incident angle data eliminates human preferences or presets, making angle selection more objective and fair, helping to avoid subjective influences, and making test results more objective and accurate. The introduction of random angle selection increases the robustness of the system, making it more adaptable to different working conditions and environmental changes, and helping the system maintain stability and reliability in complex environments. The generation of random incident angle data can help the system optimize signal acquisition, making it better adaptable to the characteristics of terahertz signals in different situations, thereby improving signal quality and analysis accuracy.

[0027] Optionally, S2 is specifically:

[0028] Performing integrated empirical mode decomposition on terahertz detection signal data to obtain terahertz eigenmode data;

[0029] Perform sample entropy calculation on the terahertz eigenmode data to obtain eigenmode sample entropy data;

[0030] The terahertz intrinsic mode data is subjected to noise filtering according to the eigenmode sample entropy data to obtain the terahertz intrinsic mode noise reduction data;

[0031] Reconstruct the terahertz eigenmode noise reduction data to obtain terahertz reconstructed signal data;

[0032] Baseline correction is performed on the terahertz reconstruction signal data to obtain terahertz preprocessed data.

[0033] By integrating empirical mode decomposition and intrinsic mode sample entropy calculation in the present invention, the present invention can effectively reduce the noise in the terahertz detection signal, which helps to improve the signal-to-noise ratio of the signal. The generation of terahertz intrinsic mode noise reduction data allows the noise in the original signal to be filtered out while retaining important information about cable corrosion. Through signal reconstruction, the morphology of the original signal can be restored to make it more consistent with actual physical properties. By performing baseline correction on the terahertz reconstructed signal data, the present invention helps to make the signal have a consistent baseline, further improving the accuracy and analyzability of the signal. The terahertz preprocessing method of the present invention helps to improve signal quality, reduce signal distortion caused by environmental noise or interference, and help to obtain more accurate terahertz signals, thereby more accurately identifying and evaluating cable corrosion.

[0034] Optionally, sample entropy calculation is performed on the terahertz eigenmode data to obtain eigenmode sample entropy data, including:

[0035] Perform periodic window extraction on terahertz intrinsic mode data to obtain periodic window data;

[0036] The sample entropy of the terahertz eigenmode data is calculated according to the periodic window data and the preset similarity threshold data to obtain the eigenmode sample entropy data.

[0037] In the present invention, by performing periodic window extraction and sample entropy calculation on the terahertz intrinsic mode data, the present invention can detect the periodic features in the signal, which helps to discover the periodic changes in the signal caused by internal corrosion of the cable, thereby making it easier to locate and identify the corrosion site. The present invention uses preset similarity threshold data to calculate the intrinsic mode sample entropy data, and the threshold can be adaptively adjusted according to the characteristics and requirements of the signal, which helps to adapt to signal changes in different situations and improves the sensitivity and accuracy of detection. Through sample entropy calculation, the present invention helps to filter out noise and random fluctuations in the signal, thereby improving the clarity and analyzability of the signal, and helping to accurately identify features in the signal, such as the degree of corrosion. The sample entropy calculation method of the present invention helps to improve the reliability and stability of detection, and can extract important information about the degree of corrosion from the signal, making corrosion detection more reliable.

[0038] Optionally, performing signal reconstruction on the terahertz eigenmode noise reduction data to obtain terahertz reconstructed signal data includes:

[0039] Performing linear combination processing on the terahertz eigenmode noise reduction data to obtain first terahertz reconstruction signal data;

[0040] Performing wavelet reconstruction on the terahertz eigenmode denoising data to obtain second terahertz reconstructed signal data;

[0041] Performing signal-to-noise ratio calculation based on the terahertz detection signal data and the first terahertz reconstruction signal data to obtain first signal-to-noise ratio data, and performing signal-to-noise ratio calculation based on the terahertz detection signal data and the second terahertz reconstruction signal data to obtain second signal-to-noise ratio data;

[0042] The first terahertz reconstruction signal data and the second terahertz reconstruction signal data are dynamically weighted calculated according to the first signal-to-noise ratio data and the second signal-to-noise ratio data to obtain the terahertz reconstruction signal data.

[0043] In the present invention, by performing linear combination and wavelet reconstruction processing on the terahertz intrinsic mode noise reduction data, the present invention can reconstruct the terahertz signal, thereby improving the quality and clarity of the signal, and helping to more accurately identify cable corrosion and locate the problem area. The present invention calculates the signal-to-noise ratio of the terahertz detection signal data and the reconstruction signal data to obtain the first and second signal-to-noise ratio data, and adopts dynamic weighted calculation to optimize the combination of the two reconstruction signal data according to the signal-to-noise ratio to retain the useful signal components to the greatest extent and reduce the noise, which helps to improve the accuracy and reliability of the corrosion degree. The present invention not only provides a reconstructed signal, but also provides two different reconstruction signal data, each of which can be calculated and dynamically weighted with the original detection signal data for signal-to-noise ratio, so that the most suitable reconstruction signal can be selected from different angles and trade-offs to more comprehensively analyze and understand the internal corrosion of the cable. The dynamic weighted calculation in the present invention allows adaptive adjustment according to specific application requirements and signal characteristics, so that the present invention is suitable for different types of cables and corrosion conditions.

[0044] Optionally, S4 is specifically:

[0045] Perform convolution calculation according to the terahertz data of the corrosion degree to obtain the terahertz convolution layer data of the corrosion degree;

[0046] Perform self-attention calculation on the terahertz convolution layer data of the corrosion degree to obtain terahertz self-attention data of the corrosion degree;

[0047] Performing pooling layer calculation on the terahertz self-attention data of the corrosion degree to obtain the terahertz pooling layer data of the corrosion degree;

[0048] Performing full connection calculation on the corrosion degree terahertz pooling layer data to obtain the corrosion degree terahertz fully connected layer data;

[0049] Perform activation layer calculation on the terahertz fully connected layer data of the corrosion degree to obtain the terahertz index data of the corrosion degree;

[0050] The terahertz index data of the corrosion degree is modeled according to the cable corrosion degree data corresponding to the terahertz data of the corrosion degree, and a terahertz identification model of the corrosion degree is obtained.

[0051] By using a convolutional neural network (CNN) combined with a self-attention mechanism, the present invention can extract rich feature information from the corrosion degree terahertz data, achieve highly accurate corrosion degree identification, and help to detect and evaluate the corrosion situation inside the cable early so that appropriate maintenance and repair measures can be taken. By performing convolution, self-attention, pooling and full-connection calculations on the corrosion degree terahertz data, the present invention constructs an efficient corrosion degree terahertz identification model. This model has a strong learning ability and can automatically learn and capture key features in terahertz data, improving the accuracy and robustness of the corrosion degree. The corrosion degree terahertz identification model of the present invention can be used to monitor and identify the corrosion situation inside the cable in real time without manual intervention, which helps to improve the efficiency of maintenance and monitoring and reduce labor costs and error rates.

[0052] Optionally, the corrosion degree terahertz convolution layer data includes first corrosion degree terahertz convolution layer data, second corrosion degree terahertz convolution layer data and third corrosion degree terahertz convolution layer data, and convolution calculation is performed according to the corrosion degree terahertz data to obtain the corrosion degree terahertz convolution layer data, including:

[0053] According to the terahertz data of the degree of corrosion, waveform shape feature extraction, spectrum feature extraction and time-frequency feature extraction are performed to obtain terahertz waveform shape feature data, terahertz spectrum feature data and terahertz time-frequency feature data;

[0054] Performing one-dimensional convolution layer calculation on the terahertz waveform shape characteristic data to obtain terahertz convolution layer data of the first corrosion degree;

[0055] Performing two-dimensional convolution layer calculation on the terahertz spectrum feature data to obtain terahertz convolution layer data of the second corrosion degree;

[0056] Perform multi-channel input convolution calculation on the terahertz time-frequency characteristic data to obtain the terahertz convolution layer data of the third corrosion degree.

[0057] In the present invention, by using different convolution layers to process terahertz waveform shape feature data, terahertz spectrum feature data and terahertz time-frequency feature data, the present invention can capture important information in the terahertz data of corrosion degree from multiple angles, which helps to improve the performance of the recognition model and enable it to better understand and distinguish different degrees of corrosion. By adopting multi-channel input convolution calculation, the present invention fully utilizes the multi-dimensional information in the terahertz time-frequency feature data, further enhances the ability of the recognition model, and helps to more accurately identify and classify the degree of corrosion. Taking the waveform shape, spectrum and time-frequency characteristics into consideration, the terahertz convolution layer data of the degree of corrosion can be made more informative, which helps to improve the sensitivity of the model to different degrees of corrosion, thereby improving the accuracy and reliability of the degree of corrosion.

[0058] Optionally, S5 is specifically:

[0059] Acquire new terahertz detection signal data;

[0060] The new terahertz detection signal data is identified by using the terahertz corrosion degree identification model to obtain corrosion degree identification data;

[0061] The corrosion degree identification data is mapped through a preset corrosion grade coding table to obtain the internal corrosion degree data of the cable.

[0062] The present invention uses a terahertz corrosion degree recognition model to identify new terahertz detection signal data, so that the corrosion degree information inside the cable can be obtained in real time, which helps to discover corrosion problems in time, take necessary maintenance measures, and reduce potential safety risks. By mapping the corrosion degree recognition data to a preset rust grade coding table, the present invention provides quantitative cable internal corrosion degree data. By mapping the corrosion degree recognition data, the corrosion degree is presented in a color-coded manner, so that the user can intuitively understand the corrosion situation inside the cable, which helps maintenance personnel quickly identify problem areas and improve work efficiency.

[0063] The purpose of the present invention is to use a terahertz sensor to scan the surface of the cable, and to achieve highly accurate corrosion degree detection through deep learning model construction. The deep learning model can learn corrosion characteristics from a large amount of data, improve the accuracy of detection, help to timely discover corrosion problems, and prevent accidents. Through the terahertz recognition model of corrosion degree, the new terahertz detection signal data can be recognized in real time and color-coded and visualized. When the degree of corrosion inside the cable changes, the system can respond immediately, provide real-time monitoring and early warning, and help maintenance personnel take timely maintenance measures. The present invention can not only detect the existence of corrosion problems, but also provide quantitative corrosion degree data, which helps to determine the severity of the problem, formulate maintenance plans in a targeted manner, avoid unnecessary maintenance costs, and extend the service life of the cable. The deep learning model and automated process in the present invention reduce the need for manual intervention, improve the efficiency of detection and identification, and reduce the risk of human error. By discovering and dealing with the corrosion problem of the cable in advance, the present invention helps to improve the safety of the cable system, help prevent potential accidents and damage, and ensure the safety of equipment and personnel. The present invention does not need to peel off the sheath. According to the different time domain characteristics and absorption peaks of THz signals collected in different corrosion areas (different corrosion products have different absorption peaks), identification models of different corrosion levels are established according to the deep learning algorithm to achieve rapid non-destructive detection of corrosion levels. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting implementations made with reference to the following drawings:

[0065] Figure 1 A flowchart showing the steps of a quantitative detection method for internal corrosion of a cable according to an embodiment;

[0066] Figure 2 A flowchart showing the steps of a method for generating first terahertz threshold angle data according to an embodiment is shown;

[0067] Figure 3 A flowchart showing a method for preprocessing a terahertz signal according to an embodiment of the present invention is shown;

[0068] Figure 4 A flowchart showing a method for reconstructing a terahertz signal according to an embodiment of the present invention is shown;

[0069] Figure 5 A diagram showing the effect of a terahertz signal reconstruction method according to an embodiment is shown;

[0070] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0071] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0072] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0073] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0074] See also Figures 1 to 5 The present application provides a terahertz-based quantitative detection method for internal corrosion of cables, the method comprising:

[0075] S1, scanning the cable surface through a terahertz sensor and receiving a reflected signal to obtain terahertz detection signal data;

[0076] Specifically, a terahertz sensor device having a terahertz radiation source and a receiver is used, the sensor device is placed on the surface of the cable, and a scanning process is started to collect reflected signal data.

[0077] Specifically, the reflection mode of the portable THz-TDS system is used to scan the cable surface and collect THz detection signals. The scanning angle and spacing can be adjusted as needed.

[0078] S2, performing terahertz signal preprocessing on the terahertz detection signal data to obtain terahertz preprocessing data;

[0079] Specifically, the preprocessing step may include noise removal, signal correction, signal enhancement, etc. For example, a digital filter may be used to remove noise, smooth the signal, or perform background correction to eliminate environmental influences.

[0080] Specifically, moisture and noise in the environment will introduce a non-flat baseline in the THz detection signal, thereby reducing the signal-to-noise ratio of the detection signal. However, it is often unrealistic to achieve low humidity and high signal-to-noise ratio detection conditions in the actual detection environment. Therefore, the collected THz original signal is decomposed and reconstructed based on the method of integrated empirical mode decomposition (EEMD) sample entropy to filter out the noise in the signal, remove the non-flat baseline, and improve the signal-to-noise ratio. The baseline, THz original signal, and THz reconstructed signal are offset by 1200 in a certain acquisition.

[0081] Construct a coordinate system with the horizontal axis being time (ps) and the vertical axis being the terahertz wave. The horizontal axis is divided into 0, 40, 80, 120, 160, 200 and 240, and the vertical axis is -600, 0, 600, 1200 and 1800. The original signal and the baseline are plotted and generated, and the reconstructed signal is offset by 1200.

[0082] S3, obtaining cable corrosion degree data, and integrating the cable corrosion degree data and terahertz preprocessing data to obtain corrosion degree terahertz data of different corrosion degrees;

[0083] Specifically, the corrosion degree data can come from different data sources, such as optical detection, chemical analysis, etc. These data are integrated with the pre-processed terahertz signal data for subsequent analysis and modeling.

[0084] S4. Construct a deep learning model based on the terahertz data of corrosion degree to obtain a terahertz recognition model of corrosion degree;

[0085] Specifically, using deep learning methods, a neural network model, such as a convolutional neural network (CNN) or a recurrent neural network (RNN), can be constructed to learn the association between terahertz signals and corrosion levels through the process of data labeling and model training. For example, according to the corrosion mechanism of cable steel wire, the corrosion substances (ZnSO4, Fe3O4, Fe2O3, Fe(OH)3, etc.) at different corrosion stages have different absorption coefficients for terahertz waves, which are reflected in the THz detection signal, showing the THz detection signal at different corrosion stages, with different amplitudes in the time domain, different reflection pulse intervals, and different peak frequencies in the frequency domain. Therefore, according to the preset steel wire corrosion degree classification, a cable THz detection signal database with different steel wire corrosion levels is established, and it is used as a data set for training signal classification models.

[0086] Rust grade: the steel wire is intact and has no rust, and the surface is shining with metallic luster; white galvanized casting powder appears on the surface of the steel wire, but the steel wire matrix is ​​not rusted; the galvanizing on the surface of the steel wire is partially exhausted, and the matrix rust causes yellow spots to appear. After cleaning these yellow spots with a wire brush and a rag, the surface of the steel wire is smooth again, and there are visible traces of rust; yellow spots appear on the surface of the steel wire, and they cannot be cleaned with a wire brush; the yellow rust spots on the surface of the steel wire become darker in color and increase in number. Some rust spots are connected into pieces, but the area covered by rust is smaller than that of zinc powder. Rust products that are easy to peel off appear in the center of the rust spots, and shallow pits can be seen after rust removal; rust spots appear in pieces on the surface of the steel wire, and the area covered by rust is larger than that of zinc powder. The rust products expand and bulge, and obvious pits appear on the surface of the steel wire after rust removal; the galvanizing is exhausted, the steel wire is severely rusted, and obvious cross-sectional losses can be found before casting removal; the cross-sectional loss of the steel wire is more than 80%, which can be regarded as a break.

[0087] S5. Acquire new terahertz detection signal data, and use the corrosion degree terahertz recognition model to identify the new terahertz detection signal data and visualize them by color coding to obtain the internal corrosion degree data of the cable.

[0088] Specifically, when new terahertz detection signal data is available, it is input into the trained terahertz corrosion recognition model. The model will analyze the signal and output a predicted value for the degree of corrosion. These predicted values ​​can be mapped to a color coding scheme to visualize the degree of corrosion inside the cable.

[0089] In the present invention, without peeling off the sheath, according to the different time domain characteristics of the THz signal collected in different corrosion areas, the different absorption peaks (different corrosion products have different absorption peaks), the recognition model of different corrosion levels is established according to the deep learning algorithm, and the rapid non-destructive detection of the corrosion level is realized. By scanning the surface of the cable with a terahertz sensor, the corrosion of the cable can be detected efficiently and quickly without the need for cumbersome manual inspection and disassembly operations. Through the deep learning model, the collected terahertz detection signal is integrated with the corrosion degree data to realize the quantitative analysis of the corrosion degree, so that the corrosion degree can be more accurately measured. The new terahertz detection signal can be input into the corrosion degree terahertz identification model in real time for real-time monitoring, and the color coding visualization results are used to make the operation more intuitive, so that the internal corrosion of the cable can be intuitively presented. The use of terahertz technology for detection does not require destructive detection methods for the cable, and the integrity of the cable can be maintained.

[0090] Optionally, the terahertz detection signal data includes first terahertz detection signal data and second terahertz detection signal data, and S1 is specifically:

[0091] S11, controlling the terahertz sensor to scan the cable surface through a preset first terahertz threshold angle data and receive a reflected signal to obtain first terahertz detection signal data;

[0092] Specifically, the scanning parameters of the terahertz sensor are set, including the scanning speed, scanning range and scanning angle. Then, the sensor is aligned with the surface of the cable and scanned at a preset first terahertz threshold angle. After the sensor receives the reflected signal, it records and processes the first terahertz detection signal data.

[0093] Scanning speed: 10 cm / s, scanning range: 0-180 degrees, scanning angle: 30 degrees, first terahertz threshold angle data: 30 degrees, the terahertz sensor is set to scan at a speed of 10 cm / s and at an angle of 30 degrees in the range of 0 to 180 degrees. The sensor first sets its angle to 30 degrees and then starts scanning the cable surface. When the sensor receives the reflected signal, it records and processes the first terahertz detection signal data.

[0094] S12, controlling the terahertz sensor to scan the cable surface through the preset second terahertz threshold angle data and receiving the reflected signal to obtain second terahertz detection signal data, wherein the first terahertz threshold angle data is smaller than the second terahertz threshold angle data.

[0095] Specifically, the scanning parameters of the terahertz sensor are set, but this time the preset second terahertz threshold angle is used for scanning. Ensure that the first terahertz threshold angle is smaller than the second terahertz threshold angle so that terahertz signals of different frequencies can be obtained at different angles. After the sensor receives the reflected signal, it records and processes the second terahertz detection signal data.

[0096] Scanning speed: 12 cm / s, scanning range: 0-180 degrees, scanning angle: 45 degrees, first terahertz threshold angle data: 30 degrees, second terahertz threshold angle data: 60 degrees, the terahertz sensor is set to scan at a speed of 12 cm / s at an angle of 45 degrees in the range of 0 to 180 degrees. First, the sensor sets its angle to 30 degrees, then starts scanning the surface of the cable, records and processes the first terahertz detection signal data. Next, the sensor sets the angle to 60 degrees, scans the surface of the cable again, and records and processes the second terahertz detection signal data. Here, the first terahertz threshold angle data is less than the second terahertz threshold angle data, ensuring that terahertz signals at different angles are obtained.

[0097] Specifically, in the subsequent model input process, the first terahertz detection signal data and the second terahertz detection signal data are simultaneously input into a fusion algorithm or a deep learning model. The former fuses two data with similar characteristics to reduce the interference of noise data, while the latter can calculate two data with different dimensions into the same dimension for combined calculation.

[0098] Specifically, the first terahertz detection signal data and the second terahertz detection signal data have different characteristics because they are acquired at different incident angles. The first terahertz detection signal data reflects the surface reflection characteristics at different angles and has more high-frequency components or surface detail information. The second terahertz detection signal data reflects different physical properties or depth information and is more sensitive to the structure or changes inside the material. For different input data, different weights or contributions are assigned for data fusion, and then model training is performed.

[0099] Specifically, the first terahertz detection signal data and the second terahertz detection signal data are regarded as measurement data with the same dimension but different values, the two data are connected in series or spliced ​​as the input of the model, or they are input into different models respectively, and then the output results of the models are fused.

[0100] In the present invention, by controlling the terahertz sensor to scan using the first terahertz threshold angle and the second terahertz threshold angle respectively, two sets of terahertz detection signal data at different angles are obtained, which helps to comprehensively consider the signal characteristics at different angles and improve the accuracy and reliability of detection. The first terahertz detection signal data and the second terahertz detection signal data can be regarded as different signal modes, and the signal modes at these two different angles are fused to comprehensively utilize their information, so as to more comprehensively understand the situation of the cable. The terahertz signals at different angles will show different characteristics at the corrosion site. Through multi-angle detection and signal fusion, the degree of corrosion can be better detected and quantified, and the ability of corrosion detection is improved. Due to the use of multiple angles, the present invention helps to improve the detection sensitivity of the internal corrosion of the cable, and can more easily identify and locate the corrosion site. By comparing the terahertz detection signals at different angles, the possibility of false alarms can be reduced, the reliability of the system is enhanced, and the misjudgment rate is reduced.

[0101] Optionally, the step of generating the first terahertz threshold angle data includes the following steps:

[0102] S111, obtaining corrosion product characteristic data, wherein the corrosion product characteristic data includes cable steel wire corrosion product refractive index data, cable steel wire corrosion product absorption coefficient data, and cable steel wire corrosion product scattering characteristic data;

[0103] Specifically, the corrosion product characteristic data of the corrosive substances on the cable wires are collected, such as the corrosion product characteristic data of the refractive index, absorption coefficient and scattering characteristic data, which are obtained by acquiring material data and querying through a material characteristic database.

[0104] Specifically, the refractive index of the cable wire surface is measured by a measuring sensor to understand the optical properties of the corrosive material, such as the refractive index data of 1.45. By scanning different areas of the cable and analyzing its absorption characteristics, the absorption coefficient data is obtained, such as the absorption coefficient of 0.005. By measuring the intensity of scattered light on the cable surface, the scattering characteristic data is obtained, such as the scattering characteristic is low scattering.

[0105] S112, acquiring cable shape data, and constructing a three-dimensional cable model according to the cable shape data to obtain three-dimensional cable model data;

[0106] Specifically, the shape data of the cable, including diameter, length, curvature, etc., is obtained using measuring instruments or imaging technology. This can be done by laser ranging, cameras or other measuring devices. Based on the obtained cable shape data, a three-dimensional model of the cable is constructed using computer-aided design (CAD) software or a three-dimensional modeling tool. This model should accurately reflect the actual cable shape.

[0107] Specifically, the diameter, length and curvature of the cable are measured, such as a cable diameter of 10 mm and a length of 50 m. The acquired data is input into computer software, such as CAD software or a 3D modeling tool, to generate an accurate 3D model of the cable.

[0108] S113, annotating the cable 3D model data using the corrosion product characteristic data to obtain the cable 3D model annotated data;

[0109] Specifically, the terahertz optical parameter library of corrosion product characteristic data is used to annotate the three-dimensional model of the cable, including associating the refractive index, absorption coefficient and scattering characteristic information with the corresponding parts in the model.

[0110] Specifically, the corrosion product distribution data is obtained, and the corrosion product distribution data and the cable three-dimensional model data are used to generate the corrosion product distribution to obtain the cable corrosion product distribution three-dimensional model, and the obtained corrosion product characteristic data is associated with the cable corrosion product distribution three-dimensional model, and annotated to obtain the cable three-dimensional model annotation data. For example, according to the location and type of the corrosive material, the corresponding corrosion area marked in the three-dimensional model is associated and marked. For example, two corrosion areas have been marked in the cable model, which are located in the middle and bottom of the cable respectively, and the corresponding corrosion product characteristic data are annotated.

[0111] S114, generating random incident angle data, and performing terahertz incident simulation on the cable three-dimensional model annotation data according to the random incident angle to obtain terahertz incident simulation data;

[0112] Specifically, a random number generator is used to generate a series of random incident angle data, which will be used to simulate the incident angle of the terahertz signal. Using the generated random incident angle data, these angles are applied to the annotated three-dimensional cable model to simulate the reflection of the terahertz signal at different incident angles.

[0113] Specifically, a random number generator is used, such as the random library or numpy library in Python, to generate a series of random numbers as incident angle data, such as randomly generating between 0 and 90 degrees to simulate different incident angles. Assume that 5 random incident angle data are generated: 15°, 30°, 45°, 60°, and 75°. The generated random incident angle data is applied to the annotated three-dimensional cable model to simulate the reflection of the terahertz signal at different incident angles. For each generated random incident angle, it is applied to the corresponding position on the three-dimensional model to simulate the incidence of the terahertz signal. Use terahertz simulation software, such as CSTMicrowave Studio, Lumerical FDTD Solutions, etc., or build a mathematical model based on the transmission line theory, electromagnetic wave propagation equations, etc. to calculate the propagation path and reflection situation, and simulate the propagation and reflection process of the terahertz signal at each incident angle. Assume that at each random incident angle, the corresponding terahertz incident simulation data is simulated, including the intensity, phase and other information of the incident signal.

[0114] S115, calculating reflectivity according to the terahertz incident simulation data to obtain incident simulation reflectivity data;

[0115] Specifically, the reflectivity at each random incident angle is calculated based on the simulated incident terahertz signal and the material property data in the model.

[0116] Emissivity calculation formula:

[0117]

[0118] R is the reflectivity data, E r is the electric field amplitude data of the reflected wave, E i is the electric field amplitude data of the incident wave; during the calculation process, the refractive index and absorption coefficient in the material property data are used to correct the reflectivity, and the absorption characteristics of the material are represented by the complex refractive index.

[0119] Reflectivity correction calculation formula:

[0120]

[0121]

[0122] in is the reflectivity correction data, n is the refractive index correction data, is the refractive index data, i is the imaginary data of the absorption coefficient, is the absorption coefficient data. For each random incident angle, the calculated reflectivity data constitutes the incident simulated reflectivity data set.

[0123] S116. Sort and filter the random incident angle data according to the incident simulation reflectivity data to obtain first terahertz threshold angle data.

[0124] Specifically, the random incident angle data are sorted according to the calculated reflectivity data, and the first terahertz threshold angle data are screened out, such as the angle with the highest or most significant reflectivity.

[0125] Specifically, the incident simulated reflectivity data is sorted to find the angle with the highest or most significant reflectivity. Based on the sorted reflectivity data, a threshold or a certain standard is selected to determine the terahertz threshold angle. This threshold is set according to the actual situation, for example, the percentile of the reflectivity, a fixed threshold, etc. In general, the selected threshold can separate the corroded and non-corroded areas. The first random incident angle that meets the selected threshold is found as the first terahertz threshold angle data.

[0126] import numpy as np #incident simulation reflectance data reflectance_data = np.array([0.1, 0.2, 0.5, 0.8, 0.6, 0.3]) #sort the reflectance data sorted_indices = np.argsort(reflectance_data)[::-1]; sorted_reflectance = reflectance_data[sorted_indices]; #select threshold threshold = 0.7 #screening threshold angle first_threshold_angle = None; for i, reflectance in enumerate(sorted_reflectance): if reflectance>=threshold: first_threshold_angle = sorted_indices[i]; break #output the first terahertz threshold angle data print("First terahertz threshold angle data:", first_thresho ld_angle).

[0127] In the present invention, by obtaining the corrosion product characteristic data and the cable shape data, combined with the terahertz incident simulation, the first terahertz threshold angle data can be quantitatively generated, which not only reflects the physical characteristics of the cable, but also is related to the propagation characteristics of the terahertz signal. The method of the present invention allows the generation of random incident angle data, and sorts and screens the incident simulation reflectivity data to obtain the most suitable first terahertz threshold angle data. According to the specific cable material and shape characteristics, the best terahertz signal incident angle is personalized to maximize the information about corrosion. By quantitatively generating the first terahertz threshold angle data, it is helpful to improve the accuracy of corrosion detection. Selecting a suitable incident angle can enhance the sensitivity of the signal and make it easier to detect the corrosion inside the cable. Through personalized angle setting, the misjudgment rate can be reduced, unnecessary false alarms or omissions can be reduced, and the reliability of the system can be improved.

[0128] Optionally, the step of generating random incidence angle data comprises the following steps:

[0129] Generate random angles through a preset random number generator to obtain random angle data;

[0130] Specifically, a set of random angle data is generated using a random number generation function of a computer programming language or software (such as the random library in Python). These angles can be random values ​​between 0 and 90 degrees, indicating the direction of the incident light.

[0131] The polarization state of the random angle data is randomly selected to obtain random incident angle data.

[0132] Specifically, for each generated random angle, another random number generator or conditional statement is used to randomly select a polarization state. Polarization states usually have linear polarization states in horizontal, vertical, or other specific directions. The randomly selected polarization state can be determined according to experimental requirements or simulation scenarios.

[0133] Specifically, for example, an embodiment of generating 10 random incident angle data is as follows: Generate random angle data by a random number generator: [45°, 120°, 280°, 175°, 60°, 315°, 90°, 200°, 30°, 150°]; Randomly select the polarization state for each random angle data: For a 45° angle, select the horizontal polarization state. For a 120° angle, select the vertical polarization state. For a 280° angle, select the horizontal polarization state. For a 175° angle, select the vertical polarization state. For a 60° angle, select the horizontal polarization state. For a 315° angle, select the vertical polarization state. For a 90° angle, select the horizontal polarization state. For a 200° angle, select the vertical polarization state. For a 30° angle, select the horizontal polarization state. For a 150° angle, select the vertical polarization state.

[0134] In the present invention, random angle data is generated by using a preset random number generator, which helps to introduce diversity and randomness, and different incident angles can be obtained in multiple tests, reducing deviations and improving the comprehensiveness and reliability of the test. Randomly generating incident angle data eliminates human preferences or presets, making angle selection more objective and fair, helping to avoid subjective influences, and making test results more objective and accurate. The introduction of random angle selection increases the robustness of the system, making it more adaptable to different working conditions and environmental changes, and helping the system maintain stability and reliability in complex environments. The generation of random incident angle data can help the system optimize signal acquisition, making it better adaptable to the characteristics of terahertz signals in different situations, thereby improving signal quality and analysis accuracy.

[0135] Optionally, S2 is specifically:

[0136] S21, performing integrated empirical mode decomposition on the terahertz detection signal data to obtain terahertz eigenmode data;

[0137] Specifically, the EEMD algorithm is used to decompose the terahertz detection signal data. EEMD decomposes the original signal into multiple intrinsic mode functions (IMFs), each of which represents a different frequency component and amplitude, to obtain terahertz intrinsic mode data.

[0138] S22, performing sample entropy calculation on the terahertz eigenmode data to obtain eigenmode sample entropy data;

[0139] Specifically, the sample entropy is calculated for each intrinsic mode function. Sample entropy is an indicator used to measure the complexity and irregularity of a signal, which can be obtained through a series of mathematical calculations. The sample entropy is calculated for each IMF, and the result is stored as the intrinsic mode sample entropy data.

[0140] S23, filtering the terahertz eigenmode data for noise according to the eigenmode sample entropy data to obtain terahertz eigenmode noise reduction data;

[0141] Specifically, the sample entropy data is used to identify and filter out those intrinsic mode functions that are considered to be noise or unimportant. According to the preset threshold or rule, specific IMFs can be selected to be retained or deleted to obtain terahertz intrinsic mode noise reduction data.

[0142] S24, reconstructing the terahertz eigenmode noise reduction data to obtain terahertz reconstructed signal data;

[0143] Specifically, a reconstruction operation is performed on the retained intrinsic mode functions, and they are added together to reconstruct the signal to obtain terahertz reconstructed signal data.

[0144] S25. Perform baseline correction on the terahertz reconstruction signal data to obtain terahertz preprocessing data.

[0145] Specifically, baseline correction is performed on the terahertz reconstructed signal data to eliminate the existing baseline drift or deviation, ensure the accuracy of the signal data, and obtain terahertz preprocessed data.

[0146] By integrating empirical mode decomposition and intrinsic mode sample entropy calculation in the present invention, the present invention can effectively reduce the noise in the terahertz detection signal, which helps to improve the signal-to-noise ratio of the signal. The generation of terahertz intrinsic mode noise reduction data allows the noise in the original signal to be filtered out while retaining important information about cable corrosion. Through signal reconstruction, the morphology of the original signal can be restored to make it more consistent with actual physical properties. By performing baseline correction on the terahertz reconstructed signal data, the present invention helps to make the signal have a consistent baseline, further improving the accuracy and analyzability of the signal. The terahertz preprocessing method of the present invention helps to improve signal quality, reduce signal distortion caused by environmental noise or interference, and help to obtain more accurate terahertz signals, thereby more accurately identifying and evaluating cable corrosion.

[0147] Optionally, sample entropy calculation is performed on the terahertz eigenmode data to obtain eigenmode sample entropy data, including:

[0148] Perform periodic window extraction on terahertz intrinsic mode data to obtain periodic window data;

[0149] Specifically, a window size (e.g., a window contains n data points) and a degree of overlap between windows are selected. Starting from the terahertz intrinsic mode data, the windows are sequentially slid to cover the entire data sequence using the selected window size and degree of overlap. Each window contains n consecutive intrinsic mode data points. For each window, the intrinsic mode data contained in the window is recorded, which will constitute a periodic window data set.

[0150] The sample entropy of the terahertz eigenmode data is calculated according to the periodic window data and the preset similarity threshold data to obtain the eigenmode sample entropy data.

[0151] Specifically, for each periodic window, the similarity between the data points within the window is calculated. Using the preset similarity threshold data, determine which windows are considered similar and which are considered different, which is done by calculating the distance or similarity measure between the windows. Sample entropy is calculated for each similar set of windows. Sample entropy is a statistical measure used to measure the irregularity and complexity of the data. The sample entropy of each set of windows is recorded as part of the intrinsic mode sample entropy data.

[0152] In the present invention, by performing periodic window extraction and sample entropy calculation on the terahertz intrinsic mode data, the present invention can detect the periodic features in the signal, which helps to discover the periodic changes in the signal caused by internal corrosion of the cable, thereby making it easier to locate and identify the corrosion site. The present invention uses preset similarity threshold data to calculate the intrinsic mode sample entropy data, and the threshold can be adaptively adjusted according to the characteristics and requirements of the signal, which helps to adapt to signal changes in different situations and improves the sensitivity and accuracy of detection. Through sample entropy calculation, the present invention helps to filter out noise and random fluctuations in the signal, thereby improving the clarity and analyzability of the signal, and helping to accurately identify features in the signal, such as the degree of corrosion. The sample entropy calculation method of the present invention helps to improve the reliability and stability of detection, and can extract important information about the degree of corrosion from the signal, making corrosion detection more reliable.

[0153] Optionally, performing signal reconstruction on the terahertz eigenmode noise reduction data to obtain terahertz reconstructed signal data includes:

[0154] S241, performing linear combination processing on the terahertz eigenmode noise reduction data to obtain first terahertz reconstruction signal data;

[0155] Specifically, weight coefficients for linear combination are determined. These weight coefficients can be adjusted according to the characteristics and importance of the signal, or can be predefined. The terahertz intrinsic mode noise reduction data is multiplied by the corresponding weight coefficients, and they are added to generate the first terahertz reconstruction signal data.

[0156] Weight coefficient 1: 0.6, weight coefficient 2: 0.4, terahertz intrinsic mode noise reduction data: [0.1, 0.2, 0.3, 0.4, 0.5], calculation of the first terahertz reconstruction signal data: first terahertz reconstruction signal data = (0.6*0.1) + (0.4*0.2) + (0.6*0.3) + (0.4*0.4) + (0.6*0.5) = 0.26.

[0157] S242, performing wavelet reconstruction on the terahertz eigenmode denoising data to obtain second terahertz reconstructed signal data;

[0158] Specifically, an appropriate wavelet basis function and decomposition level are selected, which depends on the characteristics and requirements of the signal. The terahertz intrinsic mode denoised data is subjected to wavelet transformation using the selected wavelet basis function and decomposition level. The signal is reconstructed according to the wavelet coefficients to generate the second terahertz reconstruction signal data.

[0159] Wavelet basis function: Daubechies-4, decomposition level: 3, terahertz eigenmode denoising data: [0.1, 0.2, 0.3, 0.4, 0.5], calculation of the second terahertz reconstruction signal data: second terahertz reconstruction signal data = inverse wavelet transform ([0.1, 0.2, 0.3, 0.4, 0.5], Daubechies-4, 3).

[0160] S243, performing signal-to-noise ratio calculation according to the terahertz detection signal data and the first terahertz reconstruction signal data to obtain first signal-to-noise ratio data, and performing signal-to-noise ratio calculation according to the terahertz detection signal data and the second terahertz reconstruction signal data to obtain second signal-to-noise ratio data;

[0161] Specifically, the energy of the signal is calculated from the terahertz detection signal data and the first terahertz reconstruction signal data. The first signal-to-noise ratio data is calculated, and the definition of signal-to-noise ratio (SNR) is usually used, that is, the ratio of signal energy to noise energy. The energy of the signal is calculated from the terahertz detection signal data and the second terahertz reconstruction signal data. The second signal-to-noise ratio data is calculated, using the same definition of signal-to-noise ratio.

[0162] Terahertz detection signal energy: 1000, first terahertz reconstruction signal energy: 200, calculation of first signal-to-noise ratio data: first signal-to-noise ratio data = 10*log10 (signal energy / noise energy) = 10*log10 (1000 / (1000-200)) = 7.96dB, terahertz detection signal energy: 1000, second terahertz reconstruction signal energy: 150, calculation of second signal-to-noise ratio data: second signal-to-noise ratio data = 10*log10 (signal energy / noise energy) = 10*log10 (1000 / (1000-150)) = 6.95dB.

[0163] S244 . Perform dynamic weighted calculation on the first terahertz reconstruction signal data and the second terahertz reconstruction signal data according to the first signal-to-noise ratio data and the second signal-to-noise ratio data to obtain terahertz reconstruction signal data.

[0164] Specifically, dynamic weighted calculation is performed according to the first signal-to-noise ratio data and the second signal-to-noise ratio data, as well as a weight function or an algorithm, to generate terahertz reconstruction signal data.

[0165] The first signal-to-noise ratio data: 7.96 dB, the second signal-to-noise ratio data: 6.95 dB, dynamic weighted calculation: terahertz reconstruction signal data = (7.96*first terahertz reconstruction signal data) + (6.95*second terahertz reconstruction signal data).

[0166] In the present invention, by performing linear combination and wavelet reconstruction processing on the terahertz intrinsic mode noise reduction data, the present invention can reconstruct the terahertz signal, thereby improving the quality and clarity of the signal, and helping to more accurately identify cable corrosion and locate the problem area. The present invention calculates the signal-to-noise ratio of the terahertz detection signal data and the reconstruction signal data to obtain the first and second signal-to-noise ratio data, and adopts dynamic weighted calculation to optimize the combination of the two reconstruction signal data according to the signal-to-noise ratio to retain the useful signal components to the greatest extent and reduce the noise, which helps to improve the accuracy and reliability of the corrosion degree. The present invention not only provides a reconstructed signal, but also provides two different reconstruction signal data, each of which can be calculated and dynamically weighted with the original detection signal data for signal-to-noise ratio, so that the most suitable reconstruction signal can be selected from different angles and trade-offs to more comprehensively analyze and understand the internal corrosion of the cable. The dynamic weighted calculation in the present invention allows adaptive adjustment according to specific application requirements and signal characteristics, so that the present invention is suitable for different types of cables and corrosion conditions.

[0167] Optionally, S4 is specifically:

[0168] Perform convolution calculation according to the terahertz data of the corrosion degree to obtain the terahertz convolution layer data of the corrosion degree;

[0169] Specifically, the size and number of convolution kernels are determined, and these parameters are determined according to task requirements and model architecture. The terahertz data of corrosion degree is convolved with the convolution kernel to generate a convolution feature map.

[0170] Perform self-attention calculation on the terahertz convolution layer data of the corrosion degree to obtain terahertz self-attention data of the corrosion degree;

[0171] Specifically, a self-attention mechanism is designed or selected, such as the attention mechanism in Transformer. The eroded terahertz convolution layer data is input into the self-attention module to obtain the self-attention feature.

[0172] Performing pooling layer calculation on the terahertz self-attention data of the corrosion degree to obtain the terahertz pooling layer data of the corrosion degree;

[0173] Specifically, a pooling operation, such as max pooling or average pooling, is selected to perform a pooling operation on the self-attention data to reduce the dimensionality and complexity of the data.

[0174] Performing full connection calculation on the corrosion degree terahertz pooling layer data to obtain the corrosion degree terahertz fully connected layer data;

[0175] Specifically, the structure of the fully connected layer is designed, including the number of neurons and the activation function. The data of the corrosion degree terahertz pooling layer is input into the fully connected layer for linear transformation and activation function processing.

[0176] Perform activation layer calculation on the terahertz fully connected layer data of the corrosion degree to obtain the terahertz index data of the corrosion degree;

[0177] Specifically, an appropriate activation function is selected, such as ReLU or Sigmoid, and the output of the fully connected layer is applied to the activation function to generate the final corrosion degree terahertz index data.

[0178] The terahertz index data of the corrosion degree is modeled according to the cable corrosion degree data corresponding to the terahertz data of the corrosion degree, and a terahertz identification model of the corrosion degree is obtained.

[0179] Specifically, the above process is repeated, and the corrosion degree terahertz data (or the training data after the corrosion degree terahertz data is divided) is used for model training, and the model parameters are optimized by minimizing the loss function. Different optimization algorithms can be used in the training process, such as stochastic gradient descent (SGD) or Adam.

[0180] By using a convolutional neural network (CNN) combined with a self-attention mechanism, the present invention can extract rich feature information from the corrosion degree terahertz data, achieve highly accurate corrosion degree identification, and help to detect and evaluate the corrosion situation inside the cable early so that appropriate maintenance and repair measures can be taken. By performing convolution, self-attention, pooling and full-connection calculations on the corrosion degree terahertz data, the present invention constructs an efficient corrosion degree terahertz identification model. This model has a strong learning ability and can automatically learn and capture key features in terahertz data, improving the accuracy and robustness of the corrosion degree. The corrosion degree terahertz identification model of the present invention can be used to monitor and identify the corrosion situation inside the cable in real time without manual intervention, which helps to improve the efficiency of maintenance and monitoring and reduce labor costs and error rates.

[0181] Optionally, the corrosion degree terahertz convolution layer data includes first corrosion degree terahertz convolution layer data, second corrosion degree terahertz convolution layer data and third corrosion degree terahertz convolution layer data, and convolution calculation is performed according to the corrosion degree terahertz data to obtain the corrosion degree terahertz convolution layer data, including:

[0182] According to the terahertz data of the degree of corrosion, waveform shape feature extraction, spectrum feature extraction and time-frequency feature extraction are performed to obtain terahertz waveform shape feature data, terahertz spectrum feature data and terahertz time-frequency feature data;

[0183] Specifically, the waveform shape feature extraction is performed on the corrosion degree terahertz data, such as extracting the signal amplitude, peak value, periodicity and other features. The spectrum feature extraction is performed on the corrosion degree terahertz data, such as applying Fourier transform to obtain spectrum information. The time-frequency feature extraction is performed on the corrosion degree terahertz data, and time-frequency analysis methods such as wavelet transform or time-frequency transform can be used to obtain time-frequency information.

[0184] Performing one-dimensional convolution layer calculation on the terahertz waveform shape characteristic data to obtain terahertz convolution layer data of the first corrosion degree;

[0185] Specifically, determine the size and number of convolution kernels of the one-dimensional convolution layer. Perform a convolution operation on the terahertz waveform shape feature data and the one-dimensional convolution kernel to generate the first corrosion degree terahertz convolution layer data. One-dimensional convolution calculation of terahertz waveform shape feature data: convolution kernel size: 3, number of convolution kernels: 32, input data shape: (time steps, feature dimensions), for example (100, 5), one-dimensional convolution operation: perform a one-dimensional convolution operation on each feature dimension, and each convolution kernel generates a convolution result of a channel. Output shape: (time steps, number of convolution kernels), for example (98, 32).

[0186] Performing two-dimensional convolution layer calculation on the terahertz spectrum feature data to obtain terahertz convolution layer data of the second corrosion degree;

[0187] Specifically, determine the size and number of convolution kernels of the two-dimensional convolution layer. Convolve the terahertz spectrum feature data with the two-dimensional convolution kernel to generate the second erosion degree terahertz convolution layer data. Two-dimensional convolution calculation of terahertz spectrum feature data: convolution kernel size: (3,3), number of convolution kernels: 64, input data shape: (spectrum dimension, time step), for example (10,100), two-dimensional convolution operation: perform two-dimensional convolution operation on the spectrum dimension and time step, and each convolution kernel generates a convolution result of one channel. Output shape: (new spectrum dimension, new time step, number of convolution kernels), for example (8,98,64).

[0188] Perform multi-channel input convolution calculation on the terahertz time-frequency characteristic data to obtain the terahertz convolution layer data of the third corrosion degree.

[0189] Specifically, select the number of channels and the size of the convolution kernel for the multi-channel input convolution. Use the terahertz time-frequency feature data as the multi-channel input, perform a convolution operation with the convolution kernel, and generate the third erosion degree terahertz convolution layer data. Multi-channel input convolution calculation of terahertz time-frequency feature data: number of input channels: 3, convolution kernel size: (3,3), number of convolution kernels: 128, input data shape: (spectrum dimension, number of time steps, number of channels), for example (10,100,3), multi-channel input convolution operation: perform a two-dimensional convolution operation on each channel, and add the convolution results of each channel. Output shape: (new spectrum dimension, new time step, number of convolution kernels), for example (8,98,128).

[0190] In the present invention, by using different convolution layers to process terahertz waveform shape feature data, terahertz spectrum feature data and terahertz time-frequency feature data, the present invention can capture important information in the terahertz data of corrosion degree from multiple angles, which helps to improve the performance of the recognition model and enable it to better understand and distinguish different degrees of corrosion. By adopting multi-channel input convolution calculation, the present invention fully utilizes the multi-dimensional information in the terahertz time-frequency feature data, further enhances the ability of the recognition model, and helps to more accurately identify and classify the degree of corrosion. Taking the waveform shape, spectrum and time-frequency characteristics into consideration, the terahertz convolution layer data of the degree of corrosion can be made more informative, which helps to improve the sensitivity of the model to different degrees of corrosion, thereby improving the accuracy and reliability of the degree of corrosion.

[0191] Optionally, S5 is specifically:

[0192] Acquire new terahertz detection signal data;

[0193] Specifically, new terahertz detection signal data is acquired from a terahertz sensor or other data source, including new terahertz waveforms, spectra or time-frequency features, etc.

[0194] The new terahertz detection signal data is identified by using the corrosion degree terahertz identification model to obtain corrosion degree identification data;

[0195] Specifically, using the already constructed terahertz corrosion degree recognition model, the new terahertz detection signal data is input into the model for recognition, which is completed through the forward propagation of the deep learning model. The recognition result is an estimated value or classification label of the corrosion degree, indicating the corrosion condition inside the cable.

[0196] The corrosion degree identification data is mapped through a preset corrosion grade coding table to obtain the internal corrosion degree data of the cable.

[0197] Specifically, based on the corrosion degree identification data, a preset coding table is used to map the identification results to the corresponding internal cable corrosion level. The coding table can define the correspondence between different corrosion levels and identification results. The mapped data can represent the corrosion situation inside the cable in a readable way, usually a quantitative or qualitative assessment.

[0198] The present invention uses a terahertz corrosion degree recognition model to identify new terahertz detection signal data, so that the corrosion degree information inside the cable can be obtained in real time, which helps to discover corrosion problems in time, take necessary maintenance measures, and reduce potential safety risks. By mapping the corrosion degree recognition data to a preset rust grade coding table, the present invention provides quantitative cable internal corrosion degree data. By mapping the corrosion degree recognition data, the corrosion degree is presented in a color-coded manner, so that the user can intuitively understand the corrosion situation inside the cable, which helps maintenance personnel quickly identify problem areas and improve work efficiency.

[0199] The purpose of the present invention is to use a terahertz sensor to scan the surface of the cable, and to achieve highly accurate corrosion degree detection through deep learning model construction. The deep learning model can learn corrosion characteristics from a large amount of data, improve the accuracy of detection, help to timely discover corrosion problems, and prevent accidents. Through the terahertz recognition model of corrosion degree, the new terahertz detection signal data can be recognized in real time and color-coded and visualized. When the degree of corrosion inside the cable changes, the system can respond immediately, provide real-time monitoring and early warning, and help maintenance personnel take timely maintenance measures. The present invention can not only detect the existence of corrosion problems, but also provide quantitative corrosion degree data, which helps to determine the severity of the problem, formulate maintenance plans in a targeted manner, avoid unnecessary maintenance costs, and extend the service life of the cable. The deep learning model and automated process in the present invention reduce the need for manual intervention, improve the efficiency of detection and identification, and reduce the risk of human error. By discovering and dealing with the corrosion problem of the cable in advance, the present invention helps to improve the safety of the cable system, help prevent potential accidents and damage, and ensure the safety of equipment and personnel. The present invention does not need to peel off the sheath. According to the different time domain characteristics and absorption peaks of THz signals collected in different corrosion areas (different corrosion products have different absorption peaks), identification models of different corrosion levels are established according to the deep learning algorithm to achieve rapid non-destructive detection of corrosion levels.

[0200] Therefore, from any point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the attached application documents rather than the above description, and it is intended that all changes falling within the meaning and scope of equivalent elements of the application documents are included in the present invention.

[0201] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A terahertz-based quantitative detection method for internal corrosion of cables, characterized in that: The method comprises: S1, scanning the cable surface through a terahertz sensor and receiving a reflected signal to obtain terahertz detection signal data; S2, performing terahertz signal preprocessing on the terahertz detection signal data to obtain terahertz preprocessing data; S3, obtaining cable corrosion degree data, and integrating the cable corrosion degree data and terahertz preprocessing data to obtain corrosion degree terahertz data of different corrosion degrees; S4. Construct a deep learning model based on the terahertz data of corrosion degree to obtain a terahertz recognition model of corrosion degree; S5, obtaining new terahertz detection signal data, and using the corrosion degree terahertz recognition model to identify and color-code the new terahertz detection signal data to obtain the internal corrosion degree data of the cable; S4 is specifically: Perform convolution calculation according to the terahertz data of the corrosion degree to obtain the terahertz convolution layer data of the corrosion degree; Perform self-attention calculation on the terahertz convolution layer data of the corrosion degree to obtain terahertz self-attention data of the corrosion degree; Performing pooling layer calculation on the terahertz self-attention data of the corrosion degree to obtain terahertz pooling layer data of the corrosion degree; Performing full connection calculation on the corrosion degree terahertz pooling layer data to obtain the corrosion degree terahertz fully connected layer data; Perform activation layer calculation on the terahertz fully connected layer data of the corrosion degree to obtain the terahertz index data of the corrosion degree; The terahertz index data of the corrosion degree is modeled according to the cable corrosion degree data corresponding to the terahertz data of the corrosion degree, and a terahertz identification model of the corrosion degree is obtained.

2. The method according to claim 1, characterized in that The terahertz detection signal data includes the first terahertz detection signal data and the second terahertz detection signal data, and S1 is specifically: Controlling the terahertz sensor to scan the cable surface through the preset first terahertz threshold angle data and receive the reflected signal to obtain the first terahertz detection signal data; The terahertz sensor is controlled to scan the cable surface through the preset second terahertz threshold angle data and receive the reflected signal to obtain the second terahertz detection signal data, wherein the first terahertz threshold angle data is smaller than the second terahertz threshold angle data.

3. The method according to claim 2, characterized in that The step of generating the first terahertz threshold angle data comprises the following steps: Acquire corrosion product characteristic data, wherein the corrosion product characteristic data includes cable steel wire corrosion product refractive index data, cable steel wire corrosion product absorption coefficient data and cable steel wire corrosion product scattering characteristic data; Acquire cable shape data, and construct a three-dimensional cable model according to the cable shape data to obtain cable three-dimensional model data; The three-dimensional model data of the cable is annotated using the characteristic data of the corrosion products to obtain the annotated three-dimensional model data of the cable; Generate random incident angle data, and perform terahertz incident simulation on the cable three-dimensional model annotation data according to the random incident angle to obtain terahertz incident simulation data; Calculate the reflectivity according to the terahertz incident simulation data to obtain incident simulation reflectivity data; The random incident angle data are sorted and screened according to the incident simulation reflectivity data to obtain the first terahertz threshold angle data.

4. The method according to claim 3, characterized in that The steps for generating random incidence angle data include the following steps: Generate random angles through a preset random number generator to obtain random angle data; The polarization state of the random angle data is randomly selected to obtain random incident angle data.

5. The method according to claim 1, characterized in that S2 is specifically: Performing integrated empirical mode decomposition on terahertz detection signal data to obtain terahertz eigenmode data; Perform sample entropy calculation on the terahertz eigenmode data to obtain eigenmode sample entropy data; The terahertz intrinsic mode data is subjected to noise filtering according to the eigenmode sample entropy data to obtain the terahertz intrinsic mode noise reduction data; Reconstruct the terahertz eigenmode noise reduction data to obtain terahertz reconstructed signal data; Baseline correction is performed on the terahertz reconstruction signal data to obtain terahertz preprocessed data.

6. The method according to claim 5, characterized in that The sample entropy of the terahertz eigenmode data is calculated to obtain the eigenmode sample entropy data, including: Perform periodic window extraction on terahertz intrinsic mode data to obtain periodic window data; The sample entropy of the terahertz eigenmode data is calculated according to the periodic window data and the preset similarity threshold data to obtain the eigenmode sample entropy data.

7. The method according to claim 5, characterized in that The terahertz eigenmode noise reduction data is reconstructed to obtain terahertz reconstructed signal data, including: Performing linear combination processing on the terahertz intrinsic mode noise reduction data to obtain first terahertz reconstruction signal data; Performing wavelet reconstruction on the terahertz eigenmode denoising data to obtain second terahertz reconstructed signal data; Performing signal-to-noise ratio calculation based on the terahertz detection signal data and the first terahertz reconstruction signal data to obtain first signal-to-noise ratio data, and performing signal-to-noise ratio calculation based on the terahertz detection signal data and the second terahertz reconstruction signal data to obtain second signal-to-noise ratio data; Dynamic weighted calculation is performed on the first terahertz reconstruction signal data and the second terahertz reconstruction signal data according to the first signal-to-noise ratio data and the second signal-to-noise ratio data to obtain terahertz reconstruction signal data.

8. The method according to claim 1, characterized in that The corrosion degree terahertz convolution layer data includes first corrosion degree terahertz convolution layer data, second corrosion degree terahertz convolution layer data and third corrosion degree terahertz convolution layer data. Convolution calculation is performed according to the corrosion degree terahertz data to obtain the corrosion degree terahertz convolution layer data, including: According to the terahertz data of the degree of corrosion, waveform shape feature extraction, spectrum feature extraction and time-frequency feature extraction are performed to obtain terahertz waveform shape feature data, terahertz spectrum feature data and terahertz time-frequency feature data; Performing one-dimensional convolution layer calculation on the terahertz waveform shape characteristic data to obtain terahertz convolution layer data of the first corrosion degree; Performing two-dimensional convolution layer calculation on the terahertz spectrum feature data to obtain terahertz convolution layer data of the second corrosion degree; Perform multi-channel input convolution calculation on the terahertz time-frequency characteristic data to obtain the terahertz convolution layer data of the third corrosion degree.

9. The method according to claim 1, characterized in that: S5 is as follows: Acquire new terahertz detection signal data; The new terahertz detection signal data is identified by using the corrosion degree terahertz identification model to obtain corrosion degree identification data; The corrosion degree identification data is mapped through a preset corrosion grade coding table to obtain the internal corrosion degree data of the cable.

Images