A cable defect detection system

Abstract

The present application relates to the technical field of cable nondestructive testing, and particularly relates to a cable defect detection system. The system mainly solves the pain points of insufficient sensitivity in identifying early small defects in the cable, high missed detection rate and inability to accurately classify in the prior art. The system synchronously collects signals through a multi-physical field composite sensing array, optimizes excitation parameters through a genetic algorithm, fuses to generate a three-dimensional defect probability graph, uses a double-flow Transformer model to deeply analyze features and automatically identify defect types, and finally guides maintenance through AR visualization, thereby realizing non-invasive online accurate diagnosis and intelligent operation and maintenance of cable defects.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cable non-destructive testing, in particular to a cable defect detection system. BACKGROUND

[0002] In the field of cable non-destructive testing, the existing mainstream detection methods have many inherent defects. Time domain reflectometry technology is not sensitive to cable impedance changes, making it difficult to effectively identify early-stage micron-level insulation defects, and its spatial resolution is limited, making it prone to misjudgment and missed detection of multiple defects or faults at cable joints.

[0003] The ultrasonic detection method relies heavily on the integrity and uniformity of the coupling agent. In actual field or complex laying environment, the coupling condition is difficult to guarantee, resulting in serious signal attenuation and poor stability. In addition, this method requires high technical experience of the operator, and the detection result is highly subjective, making it difficult to achieve automatic interpretation.

[0004] The infrared thermal imaging technology can only detect overheating faults on the surface or near the surface of the cable, and is completely ineffective for internal early-stage defects that have not yet caused significant temperature rise. It is also extremely susceptible to environmental temperature, light and radiation interference, with a high false positive rate. Partial discharge detection is extremely sensitive to environmental electromagnetic noise, making it extremely difficult to extract and separate effective discharge signals in a complex industrial site environment.

[0005] Therefore, the existing technology either lacks detection sensitivity or is subject to harsh detection conditions, or can only sense specific types of late-stage defects, and cannot meet the urgent need for high-sensitivity, high-reliability and intelligent identification of internal early-stage small defects in cables. Therefore, there is an urgent need to invent a cable defect detection system to overcome the above technical defects. SUMMARY

[0006] The present application provides a cable defect detection system, which aims to solve the problems of low detection sensitivity, poor anti-interference ability and inability to intelligently classify internal early-stage defects in existing technology. The system synchronously collects electromagnetic wave, sound wave and thermal field signals through a multi-physical field composite sensor array, dynamically optimizes the excitation parameters using a genetic algorithm to adapt to different cable defect characteristics, and generates a three-dimensional defect probability distribution map using multi-source data fusion technology. Finally, the system realizes accurate identification and classification of defects through a dual-flow Transformer deep learning model.

[0007] The system integrates a lightweight mobile platform to realize autonomous detection operations, and uses AR visualization technology to superimpose diagnosis results and maintenance guidelines on the real scene, thereby building a complete system from accurate perception, intelligent diagnosis to intuitive interaction.

[0008] The solution significantly improves the reliability, automation level and operation efficiency of cable defect detection.

[0009] The object of the application can be achieved by the following technical solutions:

[0010] A cable defect detection system, characterized in that the system comprises a multi-physical field composite sensing array module, an intelligent signal excitation module, a multi-source data fusion processing engine module, a deep defect diagnosis decision module, a lightweight mobile detection module, a collaborative positioning decision hub module and an AR visualization interaction module, wherein:

[0011] The multi-physical field composite sensing array module is configured to perform multi-physical field coupling scanning on the surface of the cable relying on a reconfigurable probe shell to obtain a multi-dimensional original signal set of the cable.

[0012] The intelligent signal excitation module is configured to call the multi-dimensional original signal set to perform genetic algorithm optimization on excitation waveform parameters to generate type-adaptive excitation instructions for the cable.

[0013] The multi-source data fusion processing engine module is configured to perform cross-domain feature fusion on multi-source sensing data using the type-adaptive excitation instructions to output a three-dimensional probability distribution map of the cable defects.

[0014] The deep defect diagnosis decision module is configured to analyze the three-dimensional probability distribution map to perform double-flow Transformer analysis on defect features to form a diagnosis decision report of the type of cable defects.

[0015] The lightweight mobile detection module is configured to control path planning of a folding mechanical arm according to the diagnosis decision report to complete an autonomous mobile detection trajectory of the cable.

[0016] The AR visualization interaction module is configured to match the autonomous mobile detection trajectory to LiDAR-AR calibration matching of a repair space to construct a visual repair guide model of the cable defects.

[0017] Preferably, the multi-physical field composite sensing array module comprises the following when performing multi-physical field coupling scanning on the surface of the cable relying on the reconfigurable probe shell:

[0018] The reconfigurable probe shell is configured with an annular slide rail mechanism to realize continuous circumferential scanning.

[0019] The microwave antenna array is driven to rotate and switch the polarization direction by a stepping motor to enhance the response of defects at different angles.

[0020] The liquid metal coupled ultrasonic transducer is built-in with a piezoelectric ceramic array to form a directional acoustic beam focus.

[0021] The wide-spectrum infrared thermal imager is integrated with a narrow-band filter set to suppress environmental light interference.

[0022] The optical axes of the microwave antenna array, ultrasonic transducer, and infrared thermal imager are aligned to ensure consistency of spatial coordinates.

[0023] Preferably, the liquid metal-coupled ultrasonic transducer, when performing a coupling scan, includes:

[0024] A gallium-indium alloy-filled cavity is connected to a micro air pump to achieve dynamic volume adjustment;

[0025] The air pressure regulating mechanism controls the liquid metal pressure in real time based on the cable curvature radius feedback.

[0026] An elastic silicone sealing membrane is installed at the bottom of the cavity to prevent liquid metal leakage;

[0027] Nanoparticles dispersed in liquid metal enhance acoustic impedance matching properties;

[0028] The ultrasonic transducer operates in a frequency range that covers multiple resonant frequency bands.

[0029] Preferably, when the intelligent signal excitation module performs genetic algorithm optimization on the excitation waveform parameters by calling the multi-dimensional original signal set, it includes:

[0030] The fitness function calculates the ratio of the defect echo amplitude to the energy density of the background noise.

[0031] The iterative optimization process constrains the microwave pulse width to the sub-microsecond to millisecond range.

[0032] The ultrasonic sweep frequency range is dynamically adjusted according to the dielectric properties of the cable insulation material.

[0033] The genetic algorithm employs a tournament selection strategy to retain dominant genes.

[0034] The optimized type of adaptive excitation instruction is output to the programmable signal generator.

[0035] Preferably, when the multi-source data fusion processing engine module executes cross-domain feature fusion of multi-source sensor data using type-adaptive excitation instructions, it includes:

[0036] Complex envelope features of microwave time-domain reflection coefficients are extracted using orthogonal demodulation.

[0037] The ultrasonic echo attenuation slope was calculated using piecewise linear regression.

[0038] The infrared temperature rise gradient uses a gradient operator to enhance the hot spot boundary;

[0039] The unit size of the unified 3D mesh is correlated and matched with the cable structure scale;

[0040] The three-dimensional probability distribution chart annotates the spatial distribution and morphological parameters of defects.

[0041] Preferably, the deep defect diagnosis decision module, when performing dual-stream Transformer parsing of defect features by analyzing the three-dimensional probability distribution map, includes:

[0042] First-class time-frequency domain signal sequence input multi-head self-attention coding layer;

[0043] The spatial topological features of the second flow are used to extract neighborhood associations via a graph neural network.

[0044] The cross-attention layer establishes a spatiotemporal feature correlation mapping matrix;

[0045] The fully connected layer outputs the probability distribution vector of the cable defect type;

[0046] The diagnostic decision report includes an analysis of the degradation mechanism of defective materials.

[0047] Preferably, when the lightweight motion detection module performs path planning control of the folding robotic arm based on the diagnostic decision report, it includes:

[0048] The cable CAD model is imported with parametric surface geometry.

[0049] Real-time point clouds are acquired using a depth vision sensor and registered to the world coordinate system;

[0050] A heuristic search algorithm generates an optimized detection path to avoid joints / brackets;

[0051] The joint motion trajectory of the robotic arm uses spline curves for smooth transition.

[0052] The collision avoidance monitoring unit interrupts abnormal motion trajectories in real time.

[0053] Preferably, the lightweight motion detection module further includes:

[0054] The loop coil of the electromagnetic induction energy harvesting unit is coaxially wrapped around the cable conductor;

[0055] The fundamental component of the power supply network with adaptive matching frequency of alternating magnetic field;

[0056] The rectifier and voltage regulator circuit converts the induced current into a stable DC voltage;

[0057] The distributed energy storage module buffers the supply current;

[0058] The dynamic power distribution unit adjusts the power output according to the system load.

[0059] Preferably, when the AR visualization interaction module performs LiDAR-AR calibration matching on the maintenance space by matching the autonomous mobile detection trajectory, it includes:

[0060] The three-dimensional coordinates of the defect are projected onto the display coordinate system via a rigid body transformation matrix;

[0061] The AR coordinate system uses the head-mounted device's inertial navigation system as its spatial reference.

[0062] Spatial anchor points are bound to physical locations through feature point cloud registration;

[0063] The visual rendering parameters of virtual identifiers are correlated with the physical properties of defects;

[0064] The LiDAR point cloud data stream is synchronously coupled with the AR display frame rate.

[0065] Preferably, when constructing the visual maintenance guidance model, the AR visualization interaction module includes:

[0066] The defect type database is associated with a fault mode impact analysis knowledge base.

[0067] The tool list is intelligently filtered and recommended based on electrical insulation level;

[0068] The operation steps are broken down into augmented reality animations for spatial registration;

[0069] The gesture recognition engine verifies the compliance of maintenance actions in real time;

[0070] The augmented reality overlay guides the integration of a natural language interaction interface.

[0071] This invention discloses a cable defect detection system. Its core lies in employing a multi-physics composite sensor array to simultaneously acquire electromagnetic waves, acoustic waves, and thermal field signals from the cable, and dynamically optimizing excitation parameters using a genetic algorithm to accurately adapt to different defect characteristics. The system utilizes multi-source data fusion technology to construct a three-dimensional defect probability distribution map, and achieves automatic identification and classification of defect types based on a dual-stream Transformer deep learning model. Finally, it generates immersive maintenance guidance through AR visualization technology.

[0072] This invention effectively solves the technical problems of low sensitivity, poor anti-interference ability and inability to intelligently identify early defects inside cables in existing technologies. It realizes the integration of the entire process from signal acquisition and intelligent diagnosis to maintenance guidance, and significantly improves the reliability, automation level and operation and maintenance efficiency of cable defect detection.

[0073] Compared with the prior art, the present invention has the following beneficial effects:

[0074] 1. This invention significantly improves the detection capability of early micro-defects inside cables through the synergistic effect of multi-physics field composite sensing and intelligent optimization algorithms. It can accurately identify hidden dangers such as insulation degradation, micro-cracks and partial discharge that are difficult to detect by traditional single detection methods, greatly reducing the risk of cable failure caused by missed defect detection, and providing key technical support for the safe and stable operation of power systems.

[0075] 2. The system relies on a deep learning model to automatically analyze and classify defect features, which completely changes the traditional mode of relying on human experience for signal interpretation. It greatly improves the objectivity and consistency of the detection results, while reducing the technical dependence of the detection operation on professional technicians, making the diagnosis of complex cable defects more efficient and reliable.

[0076] 3. This invention deeply integrates advanced detection technology with mobile and visualization platforms, and intuitively presents the location of defects and maintenance plans through an augmented reality interface. It realizes intelligent operation of the entire process from detection and diagnosis to maintenance, effectively solving practical problems such as difficult location, slow decision-making and low efficiency in on-site operation and maintenance, and comprehensively improving the quality and efficiency of cable operation and maintenance. Attached Figure Description

[0077] Figure 1 This is a schematic diagram of the structure of a cable defect detection system according to the present invention. Detailed Implementation

[0078] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0079] Example: Figure 1 A schematic diagram of a cable defect detection system according to the present invention is provided. The cable defect detection system includes the following modules:

[0080] Multi-physics composite sensor array module: Based on the reconfigurable probe housing, multi-physics coupled scanning is performed on the cable surface to obtain the multi-dimensional original signal set of the cable;

[0081] Intelligent signal excitation module: calls a multi-dimensional original signal set to optimize the excitation waveform parameters using a genetic algorithm, and generates an adaptive excitation command for the cable type;

[0082] Multi-source data fusion processing engine module: Uses type-adaptive excitation instructions to perform cross-domain feature fusion on multi-source sensor data and outputs a three-dimensional probability distribution map of cable defects;

[0083] Deep Defect Diagnosis Decision Module: Analyzes the 3D probability distribution map and performs dual-stream Transformer analysis on the defect features to generate a diagnostic decision report on the cable defect type;

[0084] Lightweight movement detection module: Based on the diagnostic decision report, it performs path planning and control of the folding robotic arm to complete the autonomous movement detection trajectory of the cable;

[0085] AR Visualization Interaction Module: Matches autonomous mobile detection trajectories to perform LiDAR-AR calibration matching in the maintenance space, constructing a visual maintenance guidance model for cable defects.

[0086] In this embodiment, when the multiphysics composite sensor array module performs multiphysics coupled scanning of the cable surface using a reconfigurable probe housing, it includes:

[0087] The implementation of the multiphysics composite sensor array module involves a highly integrated precision mechanical structure and sensing system. The core carrier of this module is a reconfigurable probe housing meticulously manufactured from lightweight ceramic composite materials. This material was chosen based on its superior comprehensive performance: it is not only lightweight to reduce the load on the entire mobile platform, but also possesses high strength, high hardness, high temperature resistance, corrosion resistance, and electromagnetic wave transparency, perfectly adapting to the harsh conditions that may exist in complex industrial environments, such as high temperatures, oil contamination, and electromagnetic interference. The internal structure of the housing is precisely designed, dividing it into three independent mounting positions and wiring channels for fixing three different core sensor components based on different principles. This ensures that the relative spatial positions of each sensor are accurately fixed, thus providing a stable spatial registration basis for subsequent data fusion.

[0088] The first component installed inside the housing is a microwave antenna array operating in the Ka-band. This array consists of multiple microstrip patch antenna elements arranged in a specific pattern, capable of generating a directional beam. Its extremely wide operating bandwidth allows for the transmission and reception of linear frequency modulated pulse signals with extremely high bandwidth. A significant advantage of this signal is that pulse compression technology can achieve extremely high distance resolution at the receiving end, thus enabling the sensitive detection of minute impedance changes caused by defects within the cable. Adjacent to the microwave array is a 64-element ultrasonic phased array probe with a center frequency of 5MHz. This probe is composed of 64 independent piezoelectric ceramic wafers arranged in a matrix. Each element can be independently controlled for its transmission and reception delays and phases. Electronic control enables beam deflection and focusing, allowing for precise scanning of different depths and circumferential positions within the cable without mechanical movement, greatly improving the flexibility and efficiency of the inspection. The third key sensor is a long-wavelength uncooled infrared focal plane detector. Its pixel size determines the clarity of temperature imaging. It has extremely high detection sensitivity for the slight temperature rise on the cable surface caused by partial discharge or increased dielectric loss due to internal defects.

[0089] To enable these sensors to perform comprehensive and thorough inspection of the cable surface, the housing is mounted on a precision motion mechanism. At the core of this mechanism are a ring-shaped servo rail located at the bottom of the housing and a high-precision linear motor. During inspection, the ring-shaped servo rail activates first, driving the entire probe housing to rotate smoothly and uniformly around the cable's axis, achieving a complete scan of the cable's circumference. Simultaneously, the linear motor, according to a preset step distance, drives the entire rotating mechanism to move a small distance along the cable's axis after each rotation cycle. The combination of rotational and linear motion allows the sensor at the probe tip to form a complete helical scanning trajectory relative to the cable surface, ensuring full coverage of the inspection area without any blind spots.

[0090] During the synchronized scanning motion, three sensors perform their respective data acquisition tasks based on their physical characteristics. Under the control of the FPGA, the microwave antenna array periodically emits nanosecond-level broadband linear frequency modulated pulse signals. These signals penetrate the cable insulation layer and propagate. When they encounter impedance discontinuities such as defects, water droplets, bubbles, or conductor deformation within the insulation layer, they generate backscattered signals, which are captured and recorded by the sensitive receiving channel. Almost simultaneously, under the control of another FPGA channel, the ultrasonic phased array excites each element to emit ultrasonic waves according to a pre-calculated delay rule. These sound waves propagate through the cable medium, encountering interfaces with acoustic impedance differences and generating echoes. These echo signals are received by each element and amplified and digitized. Meanwhile, the infrared focal plane detector continuously captures infrared radiation images of the cable surface at an extremely high frame rate, recording the subtle spatiotemporal changes in the cable surface temperature field during the scanning process. These three data streams are synchronized in time using a highly stable atomic clock and spatially correlated using encoder data from the motion mechanism.

[0091] All acquired raw data, including microwave I / Q raw data, ultrasonic radio frequency waveform data, infrared thermal image data, and corresponding spatial coordinates and timestamps, are packaged, compressed, and indexed in real time by a data management unit to form a structured, multi-dimensional raw signal set. This data package contains synchronous sensing information on the cable's positional status from three physical dimensions: electromagnetic, acoustic, and thermal, laying a solid data foundation for subsequent intelligent signal processing and fusion analysis.

[0092] In this embodiment, the liquid metal-coupled ultrasonic transducer, when performing a coupling scan, includes:

[0093] The core of this module lies in completely solving the most critical coupling problem in ultrasonic testing: how to establish an efficient, stable, and lossless low-impedance acoustic wave transmission channel between cables and transducers of different diameters and surface conditions. Traditionally used gel-based or grease-based solid coupling agents exhibit numerous insurmountable limitations in real-world field or complex industrial environments: they are volatile and prone to drying, leading to a sharp decline in coupling effectiveness; they easily attract dust and oil, forming new acoustic barriers; uniform thickness is difficult to guarantee when applied to curved surfaces, and air bubbles are easily trapped, introducing significant acoustic energy loss and signal interference; furthermore, their maintenance and cleaning are cumbersome, severely impacting testing efficiency. This design aims to solve these inherent problems once and for all through a novel active liquid metal coupling mechanism.

[0094] The key component is a sealed, flexible cavity directly fabricated on the front end of the ultrasonic transducer's sound wave emitter. The design of this cavity is crucial; it is shaped like a flat, pouch, with a flexible, pressure-bearing bottom that contacts the cable. This bottom surface is made of a biomedical-grade, high-strength silicone film, its thickness precisely calculated to ensure good flexibility while minimizing attenuation of specific frequency sound waves. The cavity's sides and top are supported and sealed by a more rigid engineering plastic frame, forming a closed container. This container is filled with a sufficient amount of gallium, indium, and tin ternary eutectic alloy liquid metal in a specific weight percentage ratio. This ratio was selected through extensive experimental validation to ensure the alloy remains liquid within the expected outdoor operating temperature range, while possessing low viscosity and high surface tension. Furthermore, its acoustic impedance characteristics are designed to fall between those of a piezoelectric ceramic wafer and typical cable insulation materials, acting as an ideal acoustic transition layer.

[0095] To transform this cavity from a passive container into an intelligent mechanism capable of actively adapting to the cable's shape, a high-precision, low-noise miniature stepping air pump is connected to the sealed cavity via a flexible Teflon tube. This air pump incorporates a precision pressure sensor, forming a closed-loop feedback control system. Its operation is as follows: when the transducer is driven by the robotic arm to the point of contact with the cable surface, the air pump first activates, injecting or extracting gas into the cavity, thereby dynamically and precisely regulating the air pressure within the cavity. This pressure change directly acts on the liquid metal within the cavity, causing it to undergo compliant deformation. The flexible bottom membrane, thus under the hydraulic pressure transmitted by the liquid metal, bulges outward or contracts inward, like a flexible "hand," actively wrapping and conforming to the current cable's curvature, regardless of its diameter or whether the surface has uneven markings or armor marks. This process is continuously fine-tuned until the system determines, through pressure readings, that the liquid metal has reached the preset optimal contact pressure state with the cable surface, thus physically eliminating any air gaps and creating near-perfect conditions for ultrasonic transmission.

[0096] Furthermore, to further improve the transmission efficiency of acoustic energy, namely to achieve better acoustic impedance matching and enhance ultrasonic wave transmission, the basic liquid metal medium underwent nanoscale functional enhancement. In a gallium indium tin alloy liquid metal, a specific mass fraction of nanoscale zinc oxide particles was uniformly dispersed through specific surfactant treatment, mechanical stirring, and ultrasonic dispersion processes. The introduction of these nanoparticles was not a simple physical mixing; their size, morphology, and concentration were strictly designed and controlled. These high-hardness nanoparticles formed a stable suspension system in the liquid metal. Their presence effectively altered the overall density and sound velocity of the composite medium, thereby finely controlling its acoustic impedance value, enabling it to better bridge the acoustic characteristic differences between the transducer wafer and the cable insulation material. More importantly, under the influence of the sound field, these dispersed nanoparticles can effectively reduce the scattering loss of sound waves during propagation, allowing acoustic energy to be injected into the tested cable more concentratedly and efficiently, thus significantly enhancing the signal-to-noise ratio of the echo signal and the detection system's ability to detect minute defects.

[0097] This module, through the synergistic effect of a flexible cavity structure, a functional liquid metal medium, an active pneumatic control system, and nanoparticle enhancement technology, constitutes an intelligent, adaptive, and high-performance ultrasonic coupling solution. It not only physically ensures absolute consistency of the coupling interface, eliminating the uncertainties of human operation and the inherent defects of traditional coupling agents, but also optimizes the energy transmission path from an acoustic physics perspective. This lays a solid foundation for obtaining high-quality, high-reliability ultrasonic testing data, making it an indispensable key component of the entire high-precision testing system.

[0098] In this embodiment, when the intelligent signal excitation module performs genetic algorithm optimization on the excitation waveform parameters by calling the multi-dimensional original signal set, it includes:

[0099] The core task of the intelligent signal excitation module is to achieve adaptive optimization of detection parameters, which is specifically implemented through a closed-loop feedback and decision-making process based on a genetic algorithm. This process begins with in-depth analysis of the initial scan data. The module's built-in high-speed digital signal processor performs real-time spectral analysis and feature extraction from the multi-dimensional raw signal set. It does not process all the massive amounts of raw data, but rather selectively calculates the width and amplitude envelope of the echo pulse from the microwave reflection signal, extracts the energy integral within a specific time window from the ultrasonic echo sequence, and calculates the standard deviation of the temperature difference in a specific region from the infrared image sequence. These initially refined time-domain, frequency-domain, and spatial-domain features together constitute the initial feature set for evaluating the effectiveness of the defect response under the current excitation parameters, providing a quantitative evaluation basis for subsequent optimization algorithms.

[0100] Based on this, the system constructs an optimization objective centered on the signal-to-noise ratio (SNR). Specifically, the algorithm establishes the ratio of the energy of the defect echo signal to the energy of the background noise as the unique and most important fitness function of the genetic algorithm. The calculation of this function is crucial: the defect echo energy is obtained by integrating the feature signal over a specific window in the time or frequency domain, representing the intensity of useful information; while the noise energy is taken from the "quiet zone" in the signal that is far from the defect echo and free from other obvious interference, representing the level of system background noise and random interference. Maximizing the ratio of the two directly means maximizing detection sensitivity and reliability. This highly quantified objective function, like a fair judge, provides an objective and unified standard for evaluating the merits of each set of parameter schemes.

[0101] Once the objective is established, the genetic algorithm officially begins its iterative evolution process. During initialization, the algorithm randomly generates a certain number of individuals within a predefined, broad but reasonable parameter space, forming an initial population. Each individual here is not a single numerical value, but a complete, multi-dimensional combination of parameters, uniquely encoding all the key adjustable excitation parameters for this detection task. These parameters typically include, but are not limited to: the carrier frequency and instantaneous bandwidth of the microwave pulse, which determine the depth and resolution of microwave detection; and the center frequency, pulse repetition frequency, and amplitude of the emitted sound wave from the ultrasonic probe, which collectively determine the penetration capability, detection speed, and signal-to-noise ratio. Each individual represents a unique, "trial" detection strategy.

[0102] Subsequently, the algorithm enters a cyclical iteration of "selection-crossover-mutation". In the selection phase, the algorithm sorts individuals based on the fitness function value calculated for each combination of parameters, employing strategies such as tournament selection to decisively eliminate inferior individuals with poor performance and signal-to-noise ratios, while retaining and replicating high-fitness individuals with excellent performance, ensuring the continuation of superior "genes". In the crossover phase, the retained superior individuals exchange some of their parameter codes with a certain probability; for example, exchanging the microwave bandwidth parameter encoded in one individual with the ultrasonic frequency parameter encoded in another, thus potentially generating a new generation of more promising parameter combinations that combine the advantages of both parents. In the mutation phase, some parameters of individual individuals in the population are randomly fine-tuned with a small probability, such as slightly changing the pulse repetition frequency. Although this step has a low probability, it is crucial, injecting new genetic diversity into the population and preventing the algorithm from prematurely converging to a local optimum rather than the global optimum. This cyclical process continues for dozens or even hundreds of generations.

[0103] After sufficient iterative evolution, the quality of individuals in the population continuously improves, eventually converging to one or a few optimal solutions with extremely superior performance. The algorithm selects the individual with the highest fitness and decodes its encoded parameter combination into a complete set of executable control commands. This final output "type-adaptive excitation command" is the optimal detection scheme tailored by the algorithm for the specific type and state of the cable currently being detected. This command set is immediately sent to the core hardware responsible for signal generation—the Field-Programmable Gate Array (FPGA)—via a high-speed communication bus. Based on the received optimal command, the FPGA precisely controls its internal direct digital frequency synthesis module and high-voltage pulse generation circuit to generate microwave and ultrasonic drive signals with perfectly optimized amplitude, frequency, and timing. These signals are then synchronously output to the front-end sensor, thereby initiating a new round of higher-performance, higher-precision detection scanning, ultimately achieving the most sensitive perception and capture of cable defects.

[0104] In this embodiment, when the multi-source data fusion processing engine module executes a type-adaptive excitation instruction to perform cross-domain feature fusion on multi-source sensor data, it includes:

[0105] The implementation of the multi-source data fusion processing engine module is a crucial process for transforming heterogeneous data into a unified understanding. The core task of this module is to receive instructions from the host computer, optimized using a genetic algorithm, and drive the sensing system to perform a completely new, parameter-optimized scan, thereby obtaining a set of high-quality, synchronously acquired multi-source sensor data. This data carries mapping information on the internal state of the cable from three different physical dimensions: electromagnetic, acoustic, and thermal. However, their original forms vary, their dimensions differ, and the physical phenomena they reflect are completely different, making direct comparison and comprehensive judgment impossible. Therefore, the primary task of the fusion engine is to perform independent but unified preprocessing and feature extraction on these three data streams, preparing a standardized and structured feature field for subsequent deep fusion.

[0106] Preprocessing is performed on the microwave channel data. The original acquired microwave signal is a time-domain waveform containing amplitude and phase. To more clearly reveal the impedance discontinuities caused by defects inside the cable, the module uses the Hilbert transform, a classic signal processing tool, to analyze it. This transform constructs an analytical signal of the original signal, thereby accurately stripping and extracting the instantaneous envelope and instantaneous phase information of the signal. The peak position of the extracted envelope directly corresponds to the axial distance of the defect, while the changes in the amplitude and phase of the envelope contain rich information about the change in the dielectric constant of the medium at that location. Through inverse calculations of this information, the system can initially reconstruct the dielectric constant distribution profile along the cable axis, thereby obtaining a characteristic field reflecting the changes in the electromagnetic properties of the insulating material.

[0107] Almost in parallel, the data from the ultrasonic channel also undergoes a series of complex corrections and transformations. Because ultrasonic waves inevitably experience dispersion and amplitude attenuation when propagating in viscoelastic media such as cables, this leads to echo signal distortion and reduced resolution. Therefore, the module uses pre-calibrated dispersion curves and attenuation coefficients for this type of cable to perform dispersion compensation and attenuation correction on the received raw radio frequency ultrasonic signal. Dispersion compensation aims to synchronize the arrival of sound wave components of different frequencies, thereby sharpening the echo waveform; attenuation correction aims to compensate for the energy loss of sound waves as propagation depth increases, making the echo amplitudes of defects of the same size but different depths comparable. After these precise compensations, the algorithm calculates the attenuation rate of the echo signal energy over propagation time, ultimately generating an attenuation slope field that intuitively reflects the acoustic loss characteristics of the material. This field is extremely sensitive to defects such as microcracks and pores.

[0108] For infrared thermal imaging data, the processing strategy focuses on capturing dynamic changes. The system receives a series of thermal image frames acquired over time. The module employs temporal difference processing technology, subtracting the grayscale value of each subsequent frame from that of a selected reference frame at the pixel level. This processing can significantly suppress static thermal noise interference from background, ambient radiation, and uneven emissivity of the cable surface, thereby highlighting small temperature rise areas newly generated or changing due to internal defect activity. Further calculations on these difference results can extract an anomalous temperature rise gradient field characterizing the spatial gradient of temperature change; this field provides direct evidence for locating active thermal defects.

[0109] After successfully extracting three fundamentally different feature fields—dielectric constant distribution, acoustic attenuation slope, and temperature rise gradient—from three independent data sources, the biggest challenge in fusion is how to make them interact on the same stage. This is achieved through a pre-defined three-dimensional spatial registration matrix. This matrix, obtained through precise measurement and calculation during the system calibration phase, establishes a strict mathematical mapping relationship between the physical coordinates of each sensor, the encoder readings of the scanning motion mechanism, and a virtual three-dimensional mesh space constructed based on a precise CAD model of the cable. Using this matrix, the algorithm can map the data of each feature point in the three feature fields to the corresponding voxel in the virtual three-dimensional mesh without deviation. Thus, data that were originally in different coordinate systems and reflecting different physical quantities are perfectly unified under the same spatial reference system.

[0110] Within this unified three-dimensional grid space, the DS evidence theory, an advanced information fusion algorithm, is employed to complete the final decision fusion. This theory is well-suited for handling multi-source information problems with inherent uncertainty. The system treats data from three different sensors within each voxel unit as three independent pieces of "evidence," each providing a certain probability of support for the proposition "this voxel belongs to a defect" or "it does not belong to a defect." The core combination rule of the DS theory is responsible for synthesizing these three pieces of evidence and calculating the combined probability supported by their joint support. By traversing all voxels in the entire three-dimensional space, this fusion process ultimately generates a comprehensive three-dimensional probability distribution map. This map is no longer merely a reflection of a single physical quantity, but a comprehensive judgment based on electromagnetic, acoustic, and thermal evidence. It intuitively reveals, in the form of probability values, the most likely location, approximate spatial range, and relative credibility of the defect within the cable, providing the most direct and reliable basis for the final intelligent diagnosis.

[0111] In this embodiment, the deep defect diagnosis decision module, when performing dual-stream Transformer parsing of defect features by analyzing the three-dimensional probability distribution map, includes:

[0112] The implementation of the deep defect diagnosis decision-making module embodies an advanced artificial intelligence analysis paradigm that integrates both spatiotemporal information. The core input to this module is a three-dimensional probability distribution map produced by the upstream fusion engine. Each voxel in this image contains a probability value fused using DS evidence theory, representing the overall confidence level of a defect at that spatial location. However, this is still a raw probability field that requires further interpretation and understanding. The module's core task is to act as a highly specialized "image diagnostician," accurately interpreting key information such as the type, extent, and location of defects from this three-dimensional image. The core architecture for achieving this goal is a meticulously designed two-stream neural network that analyzes data from both sequential and spatial patterns, simulating the comprehensive analytical thinking of human experts.

[0113] The first step is to initiate a time-series analysis pathway, inspired by the powerful modeling capabilities of natural language processing for sequential information. The system first slices the three-dimensional probabilistic volume data along the cable's axis into a series of extremely thin two-dimensional slices, much like slicing bread. These slices are arranged in spatial order into a strict time series, where each "frame" no longer represents a single instant in time, but rather the cross-sectional probability distribution at a specific axial location on the cable. This carefully constructed sequence is then fed into a Transformer-based encoder for deep analysis. The core of the Transformer model is its self-attention mechanism, which dynamically calculates the correlation weights between any two slices in the sequence. Through this mechanism, the model can keenly capture the distribution patterns, periodicities, and long-range dependencies of defect probabilities along the cable's length. For example, it can identify whether a defect exhibits a discontinuous chain-like distribution along the axial direction, or whether an anomaly at one location is correlated with an anomaly at another distant location, thus gaining a profound understanding of the extensional characteristics of defects in the axial dimension.

[0114] Working in parallel with the temporal flow is the spatial flow analysis pathway, which aims to deeply mine the geometric morphology and topological structure information of defects in three-dimensional space. In this pathway, the entire three-dimensional probability distribution map is reinterpreted as a huge graph structure. Each node in the graph is a voxel, and the features carried by the node are its probability value and its three-dimensional coordinates. The connection relationship between nodes is defined according to their spatial proximity, usually using the k-nearest neighbor method or fixed radius neighborhood method to connect adjacent voxel nodes in physical space. This constructed graph is fed into a multi-layer graph convolutional network for processing. The core of the graph convolution operation is "message passing", that is, each node aggregates the feature information of its direct neighbor nodes to update its own feature representation. Through multi-layer stacking, each node can eventually perceive the features of a larger area within a certain radius around it. This process enables the network to accurately extract the complex spatial topological features of the defect region, such as distinguishing whether the defect is a spherical hole, a sheet-like crack, or a dendritic fractal structure, as well as the smoothness and orientation of its boundary. These features are crucial for distinguishing defect types.

[0115] After each of the two pathways completes its feature extraction, it generates two high-dimensional feature representations with different forms and perspectives: the temporal stream outputs a feature vector containing global sequence context information, while the spatial stream outputs a set of node features containing detailed three-dimensional morphological information. At this point, a cross-attention module is activated, acting as a "translator" and "integrator" between the two information domains. This module uses the first-order features from one stream as the query vector and the features from the other stream as the key and value. It calculates a set of attention weights by measuring the similarity between the query and the key, and then uses these weights to perform a weighted summation of the values. Essentially, this process allows the model to autonomously learn "which axial sequence context information should be referenced when analyzing defects at a certain spatial location" and vice versa, thereby achieving a deep, dynamic, and non-linear fusion of temporal and spatial features, generating a unified feature representation that combines macroscopic distribution patterns with microscopic morphological details.

[0116] This deeply fused top-level feature representation is fed into an output network consisting of a fully connected classification layer. This network is essentially a powerful pattern classifier, with its output neurons corresponding to various predefined defect types. The network not only outputs the most probable defect type label but also provides the confidence probability of its classification. Furthermore, through the regression head, the network can simultaneously and accurately calculate the envelope size of the defect region in three-dimensional space and the precise three-dimensional coordinates of its centroid in the cable's global coordinate system. All this information—defect type, confidence level, three-dimensional size, and precise coordinates—is automatically assembled and formatted to generate a detailed, quantitative, and directly applicable final diagnostic decision report that can guide subsequent maintenance actions, thus completing the ultimate leap from raw sensor data to advanced cognitive decision-making.

[0117] In this embodiment, when the lightweight motion detection module performs path planning control of the folding robotic arm based on the diagnostic decision report, it includes:

[0118] The lightweight mobile detection module is the core actuator for achieving automated precision scanning. Its hardware foundation is a high-performance six-degree-of-freedom collaborative robotic arm. This robotic arm, specially selected and customized, is made of lightweight alloy materials to reduce overall weight, achieving industrial-grade repeatability and ensuring extremely high pose certainty for the sensor probes during complex spatial movements. The robotic arm is integrated onto the mobile platform via a rigid and shock-resistant mounting base. Its sixth-axis flange at the far end is specially designed to securely mount and connect to the aforementioned multi-physics composite sensor probe assembly, forming an end effector integrating sensing and motion. The entire robotic arm system is centrally controlled by an embedded industrial computer. This computer, with its built-in high-performance multi-core processor, is responsible for running complex path planning algorithms and real-time kinematic calculations, serving as the "brain" of the entire mobile detection module.

[0119] The path planning process begins with the interpretation of diagnostic conclusions. After receiving the structured diagnostic decision report from the upstream diagnostic module via a standard communication interface, the algorithm first parses it, precisely extracting the three-dimensional spatial coordinates of each identified defect. These coordinates are typically defined in a "cable coordinate system" with the cable itself as the reference. To drive the robotic arm's movement, these target point coordinates must be transformed to a "base coordinate system" that the robotic arm can understand. This transformation involves a crucial coordinate transformation, achieved through a homogeneous transformation matrix precisely obtained beforehand through hand-eye calibration. Hand-eye calibration involves fixing a calibration object to the end effector of the robotic arm and guiding it to multiple different poses. The relative positional relationships between the camera, the robotic arm base, and the cable are determined using vision sensors on the platform, resulting in this fixed transformation matrix. Applying this matrix, the coordinates of the defect points in the cable coordinate system are accurately mapped to the robotic arm's base coordinate system, providing accurate spatial targets for subsequent planning.

[0120] Once the absolute coordinates of the target point are obtained, the path planning algorithm begins searching for a safe and efficient movement path within its 3D workspace. This 3D space is a virtual environment model containing the robotic arm itself, the cable under test, and surrounding known obstacles. The module employs an improved Fast Random Exploration Tree (RRT) algorithm to accomplish this task. The RRT algorithm explores feasible paths by randomly sampling and expanding the tree structure in the configuration space. Its improvement lies in that it not only finds a feasible path but also optimizes the initial path to minimize its cost. The algorithm treats the robotic arm's links and probes as a whole, performing collision detection with the support of its kinematic model to ensure that every pose along the planned path does not interfere with the cable itself or any obstacles on it, thus achieving "collision-free" intelligent movement.

[0121] The output of the planning algorithm is typically a series of discrete waypoint sequences defined in Cartesian or joint space. If this sequence is directly sent to the controller, the robotic arm's movement will be stiff and jerky, with abrupt linear motions between points. This not only causes mechanical vibration and affects probe stability but may also generate unnecessary shocks. Therefore, the path must be smoothed. The module uses a fifth-order polynomial trajectory generator to accomplish this task. This generator interpolates a smooth motion curve for each joint of the robotic arm between adjacent waypoints, ensuring continuous and smooth time, position, velocity, acceleration, and even jerk. The fifth-order polynomial guarantees the continuity of acceleration, resulting in extremely smooth and compliant robotic arm movement, minimizing shocks and vibrations, and providing a stable motion foundation for high-precision detection.

[0122] The smoothed, time-parameterized joint spatial trajectory is converted into a real-time position command stream and sent to the robotic arm's underlying multi-axis servo controller via a high-speed fieldbus. Upon receiving these commands, the controller internally closes its current, velocity, and position loops, precisely driving the servo motors of each joint to move along a predetermined, smooth trajectory. Thus, the robotic arm can autonomously and smoothly move the end-effector sensors sequentially to the coordinates of each defect point requiring detailed re-inspection, maintaining optimal coupling with the cable surface, accurately completing the entire autonomous movement and detection trajectory, and translating the intelligent diagnostic conclusions into precise physical execution actions.

[0123] In this embodiment, the lightweight motion detection module further includes:

[0124] To address the long-term endurance issue of mobile inspection platforms in field environments or without external power, a highly efficient and adaptive energy harvesting unit—an electromagnetic induction energy harvesting unit—was designed and implemented. The core design concept of this unit is to harvest energy from the live cable being inspected, achieving true "taking from the cable, using for inspection," based on Faraday's law of electromagnetic induction. The entire unit is not a simple coil, but a complex system integrating advanced soft magnetic materials, precision mechanical structures, and efficient power electronic conversion technology. It faced a series of engineering challenges: how to efficiently extract energy from a power frequency magnetic field of varying intensity; how to adapt to cables of different thicknesses and current ratings; and how to convert the induced unstable AC power into a stable, clean DC power supply usable by the electronic system. Its successful implementation is a crucial prerequisite for ensuring the entire system's long-term autonomous operation.

[0125] The core of this unit's energy harvesting mechanism is a specially designed clamp-on current transformer. Its magnetic core is made of nanocrystalline alloy strip wound and annealed. This material possesses extremely high initial permeability and saturation magnetic induction, while exhibiting very low coercivity. This means that even under conditions of small cable load current and weak magnetic field, the core can be easily magnetized to generate sufficiently strong magnetic flux, while remaining resistant to magnetic saturation under sudden large currents, ensuring wide-range adaptability and safety in energy harvesting. The core is shaped into an open ring, encapsulated within a high-strength engineering plastic shell, forming a classic clamp-on structure. One side of the shell contains a precision hinge and spring mechanism, allowing the magnetic ring to open and close easily like pliers. This enables convenient, non-contact clamping onto the periphery of the energized cable without any electrical connection or damage to the cable's original insulation, achieving truly non-invasive installation. The high-conductivity magnetic ring efficiently concentrates the diffused magnetic field generated by the cable current into its own magnetic circuit.

[0126] Winded around the nanocrystalline magnetic core is a secondary coil made of high-strength enameled wire. The number of turns in the coil has been carefully calculated and experimentally verified to achieve an optimal balance between induced voltage and output current for maximum power output. When the clamp unit is clamped onto the cable, the alternating magnetic field generated by the power frequency AC current flowing in the cable is coupled to the secondary coil by the magnetic core. According to Faraday's law of electromagnetic induction, an AC voltage of the same frequency is induced across the coil. The amplitude of this voltage is proportional to the magnitude of the current in the cable and the number of turns in the coil. However, the electrical energy directly induced from the coil is crude and cannot be used directly; its voltage and frequency fluctuate drastically with the cable load current and must be carefully conditioned by subsequent power electronic circuitry.

[0127] Therefore, the induced AC current is then fed into a high-efficiency power processing circuit module. This module first passes through a full-bridge rectifier circuit composed of fast recovery diodes, converting the AC current into pulsating DC current. Next, a multi-stage filter network composed of large-capacity electrolytic capacitors and ceramic capacitors smooths the rectified DC current, filtering out power frequency ripple and high-frequency glitches, resulting in a relatively stable DC current whose voltage value still varies with the primary current. The most critical step is completed by a high-performance DC-DC buck-boost converter chip. This chip and its peripheral circuits form a closed-loop feedback system that can accept a wide range of input voltages and dynamically adjusts the energy conversion process through high-frequency PWM control of the internal switching MOSFETs. Ultimately, it precisely generates an extremely stable DC voltage with very low ripple at its output, which fully meets the power supply requirements of all subsystems, including the mobile platform's main control computer, robotic arm servo drivers, and sensors.

[0128] This highly integrated electromagnetic induction energy harvesting unit transforms the magnetic energy originally dispersed around the cable into a stable and reliable "miniature power station" through multi-level conversion via physics, circuitry, and control. It completely eliminates reliance on traditional battery life or on-site sockets, providing a near-unlimited continuous power supply for the entire mobile inspection platform. This enables the system to perform long-duration, large-scale automated inspection operations, greatly expanding its application scenarios and practical value, and serving as an indispensable cornerstone for the high degree of autonomy of the entire system. Its wide input range design ensures that the system can reliably harvest energy whether on lightly loaded lines during off-peak hours or heavily loaded lines during peak hours, thus truly achieving all-weather energy self-sufficiency.

[0129] In this embodiment, when the AR visualization interaction module performs LiDAR-AR calibration matching on the maintenance space by matching the autonomous mobile detection trajectory, it includes:

[0130] In this embodiment, the specific implementation of the AR visualization interaction module is a key technological bridge for realizing the overlay of abstract data into concrete space in maintenance guidance. Its core lies in solving the problem of high-precision, real-time, and stable spatial registration between virtual information and the physical world. The hardware carrier of this module is a high-performance mixed reality glasses. Unlike ordinary consumer-grade AR devices, to meet the high-precision requirements of industrial inspection, two core sensors are highly integrated on the temples: a high-precision area-array LiDAR scanning module and a high-resolution RGB color camera. The LiDAR module actively and quickly acquires depth information of the surrounding environment by emitting an invisible laser beam and measuring its return time, with an accuracy down to the millimeter level, and is unaffected by changes in ambient lighting. The RGB camera passively captures the color texture information of the environment. These two complement each other, providing a rich data source for spatial perception and forming the system's "eyes."

[0131] After system startup, the environmental perception and map building process is executed first. The LiDAR module is activated, emitting laser pulses at its inherent high frequency and performing a high-speed rotating scan of the entire maintenance scene within the current field of view, including cables, supports, the ground, and even walls. Millions of laser point ranging data are received and processed in real time. Each point carries its own three-dimensional coordinates (X, Y, Z) relative to the LiDAR sensor itself. This massive set of points constitutes a high-density three-dimensional point cloud map describing the geometry of the environment. This process is fast and automated, typically completing the initial scan of a local work area within seconds, establishing an initial, accurate digital twin framework of the physical environment. Simultaneously, the RGB camera continuously captures video streams at a high frame rate.

[0132] Next, the visual SLAM algorithm begins to work in parallel, its task being to solve the core problem of "the position and orientation of the glasses in space." The algorithm performs real-time feature extraction on the incoming continuous video frames, identifying stable features such as corners and edges in the image, and performing feature matching between consecutive frames. By calculating the pixel displacements of these feature points in the image and combining this with possible inertial measurement unit data, the algorithm can calculate the pose changes of the AR glasses in space in real time at an extremely high frequency: three translational components (movement along the X, Y, and Z axes) and three rotational components. This real-time pose estimation result allows the system to know "where I am and which direction I am looking," which is a prerequisite for placing virtual objects in the correct position.

[0133] With accurate physical environment point clouds and the glasses' real-time pose, the system needs to perform the most crucial step: precisely aligning the virtual cable model with the real cable entity. This is accomplished through a point cloud registration algorithm. The algorithm takes two inputs: first, a point cloud describing the real cable and its surrounding environment obtained from a LiDAR scan; and second, an absolutely accurate virtual cable model point cloud generated from a CAD model based on the upstream autonomous motion detection trajectory. The core task of the algorithm is to calculate an optimal rigid body transformation matrix. When this transformation matrix is ​​applied to the "source point cloud," it maximizes the spatial overlap between the virtual model point cloud and the real scene point cloud. This calculation process is iteratively optimized, and the final transformation matrix precisely defines the mathematical mapping relationship between the virtual coordinate system and the real-world coordinate system.

[0134] Using this precisely calculated transformation matrix, all the virtual information to be displayed—including 3D highlighted markers of defect locations, text annotations, arrow guides, and cross-sectional models—is first defined in the coordinate system of the virtual cable model. At the beginning of each frame's rendering cycle, the system, based on the real-time acquired glasses pose and applying the aforementioned transformation matrix, uniformly transforms the coordinates of these virtual elements to the current AR glasses display coordinate system. The graphics rendering engine then, based on the final screen coordinates, overlays and renders these virtual contents onto the real-world image seen through the camera, with correct perspective and occlusion relationships. Thus, what maintenance personnel see through the AR glasses is a realistic virtual twin of the cable, spatially locked to the real cable, with defect points clearly marked at their precise, real locations, and related diagnostic data and operational instructions floating around it. This achieves millimeter-level precision, stability, and seamless integration of virtual information and physical entities, transforming abstract inspection reports into intuitive and actionable on-site maintenance guides.

[0135] In this embodiment, when the AR visualization interaction module constructs the visual maintenance guidance model, it includes:

[0136] The implementation of the AR visualization and interactive module for building a maintenance guidance model is a crucial step in transforming all preliminary detection and diagnostic results into final, executable maintenance operations. Its core is an intelligent decision support system integrating expert knowledge, computer graphics, and natural interaction. The foundation of this system is a deeply structured defect type knowledge base. This is not a simple database table, but a professional knowledge graph built on ontology. This knowledge base is constructed from long-term accumulated experimental data, field cases, and industry standards. Each node represents a specific type of cable defect, and nodes are connected through attribute relationships. Each node encapsulates multi-dimensional attribute data for that type of defect: including its typical characteristic patterns under multi-physics scanning, its fundamental physicochemical causes, the risk level determined according to relevant standards, and most importantly, the validated standard maintenance procedures. This knowledge base is the intellectual source of the system's expert-level diagnostic and guidance capabilities.

[0137] Once the diagnostic report generated by the upstream deep defect diagnosis and decision-making module is transmitted to this module, the system immediately initiates the intelligent solution matching process. The diagnostic report contains the defect type identifier, confidence level, size, and coordinates. The system's core inference engine first parses the report, extracts the most critical type identifier, and uses it as the primary search key for precise querying and matching within the vast knowledge graph. The matching process is not a simple key-value search, but a complex process that may involve uncertain reasoning. For example, when the system's confidence level in identifying a certain defect is not 100%, the engine may search several of the most likely defect type nodes in parallel and, considering their risk levels and differences in repair processes, recommend the most conservative or comprehensive repair solution. After a successful match, the engine will extract a complete set of repair solutions tailored to the defect from the corresponding knowledge nodes. This information is structured into a data object to be rendered.

[0138] Subsequently, the AR rendering engine is activated, receiving processed data objects and executing multi-layered, multi-modal information visualization rendering tasks. This is a core process that transforms abstract data into an intuitive visual experience. Rendering is layered: at the lowest level, the level that directly interacts with the real world, the engine renders a semi-transparent and highlighted 3D warning model in real time based on the precise 3D coordinates and physical dimensions of the defect, at its corresponding real location in the AR view. This could be a pulsating red sphere enveloping the defect point, or a wireframe model precisely outlining the defect's edges. The purpose is to guide maintenance personnel to locate the target unambiguously in complex on-site environments. To the side or above the view, the engine generates a floating information panel HUD. This panel uses dynamic graphics and text to clearly list and progressively highlight the required specialized tools, consumable materials, and broken-down operation steps. For extremely complex or critical steps, the engine calls pre-made 3D animation assets to provide an immersive demonstration next to the panel or directly above the real object. The animation can show the movement paths of tools, the assembly sequence of parts, and other microscopic processes that are difficult to understand intuitively with the naked eye.

[0139] To free up maintenance personnel's hands and improve operational efficiency, the system has built a complete natural interactive interface to support the control of the entire guidance model. The entire interactive system constructs a closed loop from intelligent information push and multimodal visualization to natural interactive control, ultimately forming a highly intuitive, information-rich, and easy-to-operate immersive maintenance guidance model. This greatly reduces the technical threshold and error probability of maintenance operations, extending the knowledge and experience of experts directly to the fingertips and eyes of front-line personnel.

[0140] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0141] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cable defect detection system characterized by, The system comprises a multi-physical field composite sensing array module, an intelligent signal excitation module, a multi-source data fusion processing engine module, a deep defect diagnosis decision module, a lightweight mobile detection module and an AR visualization interaction module, wherein: The multi-physical field composite sensing array module is configured to perform multi-physical field coupling scanning on the surface of the cable relying on a reconfigurable probe shell to obtain a multi-dimensional original signal set of the cable; The intelligent signal excitation module is configured to call the multi-dimensional original signal set to perform genetic algorithm optimization on excitation waveform parameters to generate type adaptive excitation instructions of the cable; The multi-source data fusion processing engine module is configured to perform cross-domain feature fusion on multi-source sensing data by using the type adaptive excitation instructions to output a three-dimensional probability distribution map of the cable defects; The deep defect diagnosis decision module is configured to analyze the three-dimensional probability distribution map to perform double-flow Transformer analysis on defect features to form a diagnosis decision report of the cable defect type; The lightweight mobile detection module is configured to control path planning of a folding mechanical arm according to the diagnosis decision report to complete an autonomous mobile detection track of the cable; The AR visualization interaction module is configured to match the autonomous mobile detection track to perform LiDAR-AR calibration matching on a maintenance space to construct a visual maintenance guide model of the cable defects. When performing multi-physical field coupling scanning on the surface of the cable relying on the reconfigurable probe shell, the multi-physical field composite sensing array module comprises: The reconfigurable probe shell is configured with an annular slide rail mechanism to realize continuous circumferential scanning; A microwave antenna array is driven to rotate and switch polarization directions by a stepping motor to enhance defect response at different angles; A liquid metal coupled ultrasonic transducer is internally provided with a piezoelectric ceramic array to form a directional acoustic beam focus; A wide-spectrum infrared thermal imager is integrated with a narrow-band filter group to suppress environmental light interference; The optical axes of the microwave antenna array, the ultrasonic transducer and the infrared thermal imager are coincident to ensure spatial coordinate consistency; An initial feature set for evaluating the effectiveness of defect response under current excitation parameters is constituted by calculating the width and amplitude envelope of echo pulses from microwave reflection signals, extracting energy integration within a specific time window from ultrasonic echo sequences, and calculating the temperature difference standard deviation of a specific area from infrared image sequences. The ratio of defect echo signal energy to background noise energy is established as the only and most important fitness function of the genetic algorithm. When performing coupling scanning, the liquid metal coupled ultrasonic transducer comprises:

2. The cable defect detection system of claim 1, wherein, A gallium-indium alloy fills a cavity connected with a micro air pump to realize dynamic volume adjustment; An air pressure adjustment mechanism controls the pressure of the liquid metal in real time according to the curvature radius of the cable; An elastic silicone sealing film is arranged at the bottom of the cavity to prevent leakage of the liquid metal; Nanoparticles dispersed in the liquid metal enhance the acoustic impedance matching characteristics; The working frequency range of the ultrasonic transducer covers multiple harmonic frequency bands. When performing genetic algorithm optimization on excitation waveform parameters by calling the multi-dimensional original signal set, the intelligent signal excitation module comprises:

3. The cable defect detection system of claim 1, wherein, The fitness function calculates the energy density ratio of defect echo amplitude to background noise. ​ The iterative optimization process constrains the microwave pulse width in the sub-microsecond to millisecond range; The ultrasonic sweep range is dynamically adjusted according to the dielectric properties of the cable insulation material; The genetic algorithm adopts a tournament selection strategy to preserve superior genes; Output the optimized type adaptive excitation instruction to the programmable signal generating device.

4. The cable defect detection system of claim 1, wherein, When the multi-source data fusion processing engine module executes cross-domain feature fusion on multi-source sensing data using type adaptive excitation instructions, it includes: The microwave time domain reflection coefficient is extracted by quadrature demodulation to obtain complex envelope features; The ultrasonic echo attenuation slope is calculated by piecewise linear regression; The infrared temperature rise gradient is strengthened by the gradient operator to highlight the hot spot boundary; The unit size of the unified three-dimensional grid is associated and matched with the cable structure scale; The three-dimensional probability distribution map labels the defect spatial distribution and morphological parameters.

5. The cable defect detection system of claim 1, wherein, When the deep defect diagnosis decision module executes double-flow Transformer analysis on the three-dimensional probability distribution map to analyze defect features, it includes: The time-frequency domain signal sequence of the first flow is input into the multi-head self-attention encoding layer; The spatial topology features of the second flow are extracted by the graph neural network to obtain neighborhood correlation; The cross-attention layer establishes a time-space feature correlation mapping matrix; The fully connected layer outputs the probability distribution vector of the cable defect type; The diagnosis decision report includes defect material degradation mechanism analysis.

6. The cable defect detection system of claim 1, wherein, When the lightweight mobile detection module executes path planning control on the folding mechanical arm according to the diagnosis decision report, it includes: The cable CAD model imports the parameterized curved surface geometry structure; Real-time point clouds are collected by depth vision sensors and registered to the world coordinate system; A heuristic search algorithm generates an optimized detection path that avoids joints and supports; The mechanical arm joint motion trajectory uses a spline curve for smooth transition; The anti-collision monitoring unit interrupts abnormal motion trajectories in real time.

7. The cable defect detection system of claim 6, wherein, The lightweight mobile detection module further includes: The annular coil of the electromagnetic induction power unit coaxially surrounds the cable conductor; The alternating magnetic field frequency is adaptively matched with the fundamental component of the power supply network; The rectifier and voltage stabilizing circuit converts the induced current into a stable DC voltage; The distributed energy storage module caches the power supply current; The dynamic power distribution unit adjusts the power output according to system load.

8. The cable defect detection system of claim 1, wherein, When the AR visualization interaction module performs LiDAR-AR calibration matching on the maintenance space by matching the autonomous mobile detection trajectory, it includes: The defect three-dimensional coordinates are projected to the display coordinate system by the rigid transformation matrix; The AR coordinate system takes the head-mounted device inertial navigation system as the spatial reference; The spatial anchor points are bound to physical locations through feature point cloud registration; The visual rendering parameters of the virtual identifier are associated with the physical properties of the defect; The LiDAR point cloud data stream is synchronously coupled with the AR display frame rate.

9. The cable defect detection system of claim 8, wherein, When the AR visualization interaction module constructs a visual maintenance guide model, it includes: The defect type database is associated with the failure mode influence analysis knowledge base; The tool list is intelligently filtered and recommended according to electrical insulation levels; The operation steps are decomposed into spatially registered augmented reality animations; The gesture recognition engine verifies the compliance of maintenance actions in real time; The augmented reality overlay guide integrates a natural language interaction interface.

Images