Electric wire and cable nondestructive on-line detection method

Abstract

The invention provides an electric wire and cable nondestructive on-line detection method which comprises the following steps: acquiring an electromagnetic signal, temperature distribution data, chemical component change data and mechanical vibration data of an electric wire and cable to be detected through a multi-mode sensor network; the electromagnetic signals, the temperature distribution data, the chemical component change data and the mechanical vibration data are subjected to data preprocessing, and a multi-dimensional data set is constructed; performing data fusion on the multi-dimensional data set through a multi-physics field coupling model to generate multi-dimensional feature data; performing defect identification based on the multi-dimensional feature data to generate a defect data set; and performing dynamic early warning according to the defect data set to generate an early warning data set. Electromagnetic, temperature, chemical and vibration data are comprehensively covered through the multi-mode sensing network, accurate recognition and early warning of wire and cable defects are achieved through preprocessing and multi-physics field coupling model fusion, the method has the advantages of being lossless online, comprehensive, intelligent and dynamic, and detection reliability and operation and maintenance efficiency are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wire and cable detection, and in particular to a non-destructive online detection method for wires and cables. Background Art

[0002] Over the long term, wires and cables are inevitably affected by a variety of factors, such as aging, wear, external damage, and environmental factors. This can lead to performance degradation or even failure, seriously threatening the normal operation of the system. Therefore, timely and effective testing of wires and cables to identify potential problems and repair them as early as possible is of great significance to ensuring the safe operation of wires and cables.

[0003] Existing nondestructive online testing methods primarily include electromagnetic induction-based testing, ultrasonic testing, and infrared testing. These methods can detect internal and external defects in wires and cables without damaging their structure. For example, electromagnetic induction testing can verify the continuity of conductors and the presence of short circuits; ultrasonic testing can detect cracks and voids within wires and cables; and infrared testing can identify potential faults caused by localized overheating.

[0004] However, existing non-destructive online testing methods generally suffer from low detection accuracy. On the one hand, due to the complex structure of wires and cables, wires and cables of different materials and specifications respond differently to detection signals, resulting in detection results that are easily interfered with and difficult to accurately reflect the true condition of the wires and cables. For example, when testing wires and cables with multi-layer insulation structures, the detection signal may be reflected and scattered between the layers, causing deviations in the detection results. On the other hand, external environmental factors such as temperature, humidity, and electromagnetic interference can also have a significant impact on the detection results. In high temperature and high humidity environments, the electrical properties of wires and cables will change, thereby interfering with the accuracy of the detection signal; strong electromagnetic interference may mask the true defect signal, resulting in missed detection or misjudgment. Summary of the Invention

[0005] In view of this, the present invention proposes a non-destructive online detection method for wires and cables, which can improve the accuracy of non-destructive online detection results of wires and cables and ensure the reliable operation of wires and cables.

[0006] The technical solution of the present invention is achieved as follows:

[0007] A non-destructive online detection method for electric wires and cables comprises the following steps:

[0008] Step S1: Collect electromagnetic signals, temperature distribution data, chemical composition change data, and mechanical vibration data of the wires and cables to be tested through a multimodal sensor network;

[0009] Step S2, performing data preprocessing on the electromagnetic signal, the temperature distribution data, the chemical composition change data, and the mechanical vibration data to construct a multidimensional data set;

[0010] Step S3, fusing the multidimensional data set through a multi-physics field coupling model to generate multidimensional feature data;

[0011] Step S4: performing defect identification based on the multidimensional feature data to generate a defect data set;

[0012] Step S5: Perform dynamic early warning according to the defect data set to generate an early warning data set.

[0013] Optionally, the multimodal sensor network includes electromagnetic sensors, thermal imaging sensors, chemical sensors, and mechanical vibration sensors;

[0014] The electromagnetic sensors are distributed at intervals along the axial direction of the wires and cables, and are used to detect the current distribution and magnetic field changes of the wires and cables;

[0015] The thermal imaging sensor covers the key nodes of the wires and cables to capture local overheating areas;

[0016] The chemical sensor is embedded in the insulation layer or sheath material of the wire and cable to monitor aging products of the insulation material;

[0017] The mechanical vibration sensor is attached to the outer surface of the wire and cable and is used to sense slight deformation of the wire and cable caused by external stress.

[0018] Optionally, the specific steps of step S2 are:

[0019] Step S21: performing abnormal value detection and repair on the electromagnetic signal, the temperature distribution data, the chemical composition change data, and the mechanical vibration data to generate initial electromagnetic signal, initial temperature distribution data, initial chemical composition change data, and initial mechanical vibration data;

[0020] Step S22: performing feature extraction on the initial electromagnetic signal, the initial temperature distribution data, the initial chemical composition change data, and the initial mechanical vibration data to generate a target electromagnetic signal, target temperature distribution data, target chemical composition change data, and target mechanical vibration data;

[0021] Step S23 : normalizing the target electromagnetic signal, the target temperature distribution data, the target chemical composition change data, and the target mechanical vibration data to construct a multidimensional data set.

[0022] Optionally, the specific steps of step S22 are:

[0023] S221, performing fast Fourier transform on the initial electromagnetic signal to generate a frequency domain feature vector;

[0024] S222. Perform feature learning on the frequency domain feature vector through a convolutional neural network, and construct a feature matrix to generate a target electromagnetic signal;

[0025] S223, extracting features from the initial temperature distribution data using a spatiotemporal convolutional neural network to generate target temperature distribution data;

[0026] S224, mapping the high-dimensional chemical composition data in the initial chemical composition change data to a low-dimensional space through linear transformation to generate intermediate chemical composition change data;

[0027] S225, using a local linear embedding method to capture the local manifold structure in the intermediate chemical composition change data to generate target chemical composition change data;

[0028] S226, performing wavelet packet decomposition on the initial mechanical vibration data to generate intermediate mechanical vibration data;

[0029] S227 : Calculate the energy value of each frequency band signal in the intermediate mechanical vibration data to generate target mechanical vibration data.

[0030] Optionally, the multi-physics field coupling model includes:

[0031] An electromagnetic field module is used to simulate the current distribution and magnetic field strength in the wires and cables and construct an electromagnetic feature space;

[0032] A thermal field module is used to calculate the temperature field distribution and heat conduction path of the wires and cables during operation and to construct a thermal feature space;

[0033] Chemical field module, used to quantify the changes in chemical composition and aging rate of insulating materials and construct chemical feature space;

[0034] The mechanical field module is used to analyze the stress-strain relationship of the wire and cable under the action of external force and construct a mechanical feature space.

[0035] Optionally, the specific steps of step S3 are:

[0036] Step S31: Mapping various types of data in the multidimensional dataset to corresponding feature spaces to generate multiple target feature spaces;

[0037] Step S32: coupling all the target feature spaces through a tensor decomposition algorithm to generate multi-dimensional feature data.

[0038] Optionally, the specific steps of step S4 are:

[0039] Step S41: performing principal component analysis on the multidimensional feature data to reduce the data dimension and extract the main feature components;

[0040] Step S42: performing cluster analysis on the main characteristic components using a clustering algorithm, and dividing the data into normal data clusters and abnormal data clusters;

[0041] Step S43: Mark the data in the abnormal data cluster as defect data, and generate a defect data set including defect locations and feature descriptions.

[0042] Optionally, the specific steps of step S42 are:

[0043] Step S421: Calculate the density distribution of the main characteristic components using a kernel density estimation method to determine high-density areas and low-density areas in the data space;

[0044] Step S422: using an improved K-means clustering algorithm, taking the high-density area as the initial cluster center, iteratively optimizing the main characteristic components to generate cluster data;

[0045] Step S423: Evaluate the compactness and separation of the cluster data by using the silhouette coefficient to generate a silhouette coefficient;

[0046] Step S424: if the silhouette coefficient is greater than a preset silhouette threshold, the clustering result is divided into a normal data cluster and an abnormal data cluster;

[0047] Step S425: If the silhouette coefficient is less than or equal to the preset silhouette threshold, adjust the parameters of the clustering algorithm and jump to step S422.

[0048] Optionally, the specific steps of step S5 are:

[0049] Step S51: Using a Bayesian network to jointly model the defect features in the defect data set and the wire and cable fault history data to generate a defect-fault association probability matrix;

[0050] Step S52: Calculate the probability value of the defect evolving into an actual fault based on the defect-fault association probability matrix. If the probability value is greater than a preset fault threshold, trigger an early warning signal.

[0051] Step S53: Associating the warning signal with the defect location information to generate a warning data set including the defect location, failure probability, warning level and recommended disposal measures.

[0052] Optionally, the specific steps of step S52 are:

[0053] Step S521: quantify the uncertainty of the defect-fault association probability matrix using a Monte Carlo simulation method to generate a probability distribution interval;

[0054] Step S522: Calculate the mathematical expectation and variance of the probability distribution interval to construct a confidence interval;

[0055] Step S523: If the lower limit of the confidence interval is greater than the preset fault threshold, a deterministic warning signal is triggered;

[0056] Step S524: If only the upper limit of the confidence interval is greater than the preset fault threshold, an uncertainty warning signal is triggered.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] The present invention provides a nondestructive online inspection method for wires and cables by simultaneously collecting data from four different modalities: electromagnetic signals, temperature distribution, chemical composition changes, and mechanical vibration. This multimodal data provides a more comprehensive view of the wire and cable status. When one modal data is interfered with or insensitive to a particular defect, data from other modalities can provide supplementary or verification information, significantly reducing the risk of failure of a single method and improving the coverage of defect detection. Data fusion is performed using a multi-physics coupling model, which can identify signal changes caused by environmental changes and distinguish them from localized defects. This model can integrate signals of varying penetration depths and sensitivities to better analyze signal propagation and interaction within multilayer structures, reducing misjudgments caused by reflection / scattering. By utilizing data from multiple related sensors that are subject to different interference patterns, a fusion algorithm can partially offset the effects of specific environmental interference. Defect identification based on the fused multidimensional feature data can effectively improve detection sensitivity, reduce false alarm rates, and enhance defect location and classification accuracy. Dynamic early warning based on this defect dataset can capture early, weak signals that indicate potential faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only preferred embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0060] Figure 1 This is a flow chart of a non-destructive online detection method for electric wires and cables of the present invention;

[0061] Figure 2 Flowchart of step S2 of a method for nondestructive online detection of electric wires and cables of the present invention;

[0062] Figure 3 Flowchart of step S3 of a method for nondestructive online detection of electric wires and cables of the present invention;

[0063] Figure 4 Flowchart of step S4 of a method for nondestructive online detection of electric wires and cables of the present invention;

[0064] Figure 5 This is a flow chart of step S5 of a method for non-destructive online detection of wires and cables of the present invention. DETAILED DESCRIPTION

[0065] In order to better understand the technical content of the present invention, a specific embodiment is provided below, and the present invention is further described in conjunction with the accompanying drawings.

[0066] like Figure 1 The figure shows a flow chart of a method for non-destructive online detection of wires and cables.

[0067] In an embodiment of the present invention, a multimodal sensor network includes electromagnetic sensors, thermal imaging sensors, chemical sensors, and mechanical vibration sensors. Electromagnetic sensors are distributed at intervals along the axial direction of wires and cables to monitor current distribution and magnetic field changes (such as local short circuits and leakage currents) in real time. Thermal imaging sensors cover key nodes of wires and cables (such as joints and bends) to capture local overheating areas (such as temperature rises caused by poor contact). Chemical sensors are embedded in insulating layers or sheath materials to monitor aging products (such as CO2 and aldehydes produced by the decomposition of insulating materials). Mechanical vibration sensors are attached to the outer surface of wires and cables to sense tiny deformations caused by external stress (such as mechanical damage and vibration fatigue). By integrating multiple sensors, the electrical, thermal, chemical, and mechanical status information of wires and cables can be fully acquired to achieve comprehensive monitoring of wire and cable aging, mechanical damage, and electrical faults. All sensors are online non-destructive testing, without the need for power outages or damage to the wire and cable structure, providing a multi-dimensional data foundation for subsequent defect detection.

[0068] The electromagnetic signals, temperature distribution data, chemical composition change data and mechanical vibration data are subjected to outlier detection and repair to generate initial electromagnetic signals, initial temperature distribution data, initial chemical composition change data and initial mechanical vibration data; the initial electromagnetic signals, initial temperature distribution data, initial chemical composition change data and initial mechanical vibration data are subjected to feature extraction to generate target electromagnetic signals, target temperature distribution data, target chemical composition change data and target mechanical vibration data; the target electromagnetic signals, target temperature distribution data, target chemical composition change data and target mechanical vibration data are subjected to data normalization to construct a multidimensional data set.

[0069] Each type of data in the multidimensional dataset is mapped to its corresponding feature space to generate multiple target feature spaces. All target feature spaces are coupled using a tensor decomposition algorithm to generate multidimensional feature data. Principal component analysis is performed on the multidimensional feature data to reduce the data dimension and extract the main feature components. Clustering algorithms are used to cluster the main feature components and divide the data into normal data clusters and abnormal data clusters. Data in the abnormal data clusters are labeled as defect data to generate a defect dataset containing defect locations and feature descriptions. A Bayesian network is used to jointly model the defect features in the defect dataset with historical wire and cable fault data to generate a defect-fault association probability matrix. The probability of a defect evolving into an actual fault is calculated based on the defect-fault association probability matrix. If the probability value exceeds a preset fault threshold, a warning signal is triggered. The warning signal is then associated with the defect location information to generate a warning dataset containing the defect location, failure probability, warning level, and recommended remedial measures.

[0070] Preferably, the multimodal sensor network includes electromagnetic sensors, thermal imaging sensors, chemical sensors, and mechanical vibration sensors;

[0071] Electromagnetic sensors are distributed at intervals along the axial direction of the wires and cables to detect the current distribution and magnetic field changes of the wires and cables;

[0072] Thermal imaging sensors cover key nodes of wires and cables to capture localized overheating areas;

[0073] Chemical sensors are embedded in the insulation or sheath material of wires and cables to monitor aging products of the insulation material;

[0074] Mechanical vibration sensors are attached to the outer surface of wires and cables to sense tiny deformations of wires and cables caused by external stress.

[0075] In an embodiment of the present invention, an electromagnetic sensor is used to monitor the current distribution and magnetic field changes of wires and cables in real time, and to capture electrical anomalies (such as local short circuits, leakage currents, insulation breakdown, etc.). The electromagnetic sensors are distributed along the axial direction of the wires and cables (such as arranging a sensor node every 10-50 meters), covering the entire body and key parts (such as joints and terminals) of the wires and cables to be detected. The signal is collected contactlessly through the principle of electromagnetic coupling (such as high-frequency current sensor HFCT or ultra-high frequency sensor UHF) to avoid interference with the structure of the wires and cables. The electromagnetic sensor can identify load fluctuations, local short circuits or ground faults by measuring the current waveform in the conductor of the wire and cable; and capture abnormal fluctuations in the magnetic field around the wire and cable (such as eddy currents caused by partial discharge or insulation degradation). The data collected by the electromagnetic sensor can be processed using fast Fourier transform and convolutional neural network. Specifically, the time domain signal is converted into a frequency domain feature vector through fast Fourier transform (FFT), and specific frequency components (such as high-frequency pulses generated by partial discharge) are extracted; the frequency domain features are deep-learned and modeled through convolutional neural network (CNN) to identify complex patterns (such as different types of electrical faults). Through dynamic monitoring and frequency domain analysis, transient anomalies (such as partial discharge) can be identified.

[0076] Thermal imaging sensors are used to capture localized overheating at key points in wires and cables in real time, preventing insulation aging or fires caused by overheating. Thermal imaging sensors are deployed at stress concentration areas such as wire and cable joints, terminals, and bends. They use non-contact infrared thermal imaging technology to convert thermal radiation into electrical signals through the piezoelectric effect. The data collected by thermal imaging sensors can be used to monitor temperature distribution, generate thermal images of the wire and cable surfaces, and identify localized hotspots (such as temperature rise caused by poor contact). This data can be combined with historical temperature data to analyze temperature trends using a spatiotemporal convolutional network.

[0077] Chemical sensors are used to monitor the aging products of wire and cable insulation materials and assess the degree of insulation degradation. Chemical sensors are directly embedded in the insulation layer or sheath material of wires and cables, and data is collected in real time through miniaturized chemical sensing modules (such as gas sensors and electrochemical sensors). Chemical sensors can detect gases (such as CO2, aldehydes) or changes in chemical composition (such as dielectric constant and polarization characteristics) produced by the decomposition of insulating materials. The data collected by the chemical sensor can be linearly transformed and locally linearly embedded. Specifically, high-dimensional chemical data is mapped to a low-dimensional space (such as PCA dimensionality reduction); the local manifold structure of the chemical composition is captured to distinguish normal aging from abnormal degradation.

[0078] Mechanical vibration sensors are used to sense tiny deformations of wires and cables caused by external stresses (such as mechanical damage and vibration fatigue) and identify potential mechanical defects. Mechanical vibration sensors are fixed to the outer surface of wires and cables by flexible adhesive or magnetic attraction. They are deployed in a distributed manner (such as one node every 5-10 meters) to cover areas where wires and cables are easily damaged (such as pipeline entrances and support points). Mechanical vibration sensors capture mechanical vibrations (such as the 0.1Hz-10kHz frequency band) through piezoelectric ceramics or fiber grating sensors. The data collected by the mechanical vibration sensor can be subjected to wavelet packet decomposition and energy threshold judgment. Specifically, the vibration signal is decomposed into multiple frequency bands and the energy characteristics are extracted; by calculating the energy value of each frequency band, abnormal vibrations (such as mechanical collisions and looseness) are identified.

[0079] The specific mechanism of sensor collaboration is as follows: electromagnetic sensors, thermal imaging sensors, chemical sensors and mechanical vibration sensors are used to collect electrical, thermal, chemical and mechanical status data of wires and cables respectively. The collected data are subjected to outlier detection and repair (such as the Z-score method), feature extraction (FFT, wavelet packet decomposition) and normalization. The four types of pre-processed data are mapped to electromagnetic, thermal, chemical and mechanical feature spaces, and data fusion is performed through tensor decomposition algorithms. Based on the fused feature data, defects are identified through principal component analysis (PCA) and clustering algorithms (improved K-means). Dynamic warnings are generated using Bayesian networks and Monte Carlo simulations. This multimodal sensor network achieves comprehensive monitoring of the status of wires and cables through the collaborative deployment of four types of sensors: electromagnetic, thermal, chemical and mechanical.

[0080] Preferably, Figure 2 The flowchart of step S2 of a method for non-destructive online detection of wires and cables is shown.

[0081] In an embodiment of the present invention, interference factors (such as noise, outliers, and dimensional differences) in the original sensor data are eliminated through systematic preprocessing, key features are extracted, and a multidimensional data set of unified scale is constructed to provide high-quality input for subsequent multi-physics field coupling modeling and defect identification. Electromagnetic signals, temperature distribution data, chemical composition change data, and mechanical vibration data are subjected to outlier detection and repair to generate initial electromagnetic signals, initial temperature distribution data, initial chemical composition change data, and initial mechanical vibration data. Statistical methods and machine learning methods are used in outlier detection. The statistical methods used include Z-score and interquartile range (IQR). Z-score is used to calculate the deviation of a data point from the mean. If the deviation exceeds a threshold (such as Z-score>3), it is marked as an outlier. IQR is used to identify outliers (such as Q1-1.5IQR or Q3+1.5IQR) based on the upper and lower quartiles of the data distribution. The machine learning methods used include Isolation Forest and cluster analysis (such as DBSCAN). Isolation Forest is used to determine outliers by randomly splitting the depth of the data tree and is suitable for high-dimensional data. Cluster analysis is used to identify outliers in low-density areas. When repairing outliers, interpolation, filtering technology, and data replacement are used. Interpolation is used to fill missing or abnormal data points using linear interpolation or spline interpolation. Filtering technology is used for electromagnetic signals and mechanical vibration data, using low-pass filters or sliding average filters to remove high-frequency noise. Data replacement is used for outliers in chemical sensors, using historical means or neighboring node data.

[0082] Features sensitive to defect detection are extracted from the preprocessed data, reducing data dimensionality and enhancing key information. Fast Fourier transforms (FFTs) are used to extract frequency-domain feature vectors (such as harmonic components and high-frequency pulses of partial discharge) from the electromagnetic signal. A convolutional neural network (CNN) is then used to extract features from the time-series data to generate the target electromagnetic signal. A spatiotemporal convolutional network combines the spatial distribution and time-series characteristics of the thermal imaging data to extract the spatiotemporal correlation features of local hotspots and generate target temperature distribution data. Principal component analysis (PCA) is used to map the high-dimensional chemical data to a low-dimensional space, preserving the principal components. Local linear embedding (LLE) is then used to capture the local manifold structure of the chemical composition, distinguishing normal aging from abnormal degradation, and generate target chemical composition change data. Multi-band energy analysis is performed on the vibration signal to extract features of subtle deformations and generate target mechanical vibration data. Through these feature extraction steps, the target electromagnetic signal, target temperature distribution data, target chemical composition change data, and target mechanical vibration data are obtained.

[0083] By eliminating the dimensional differences between different modal data, a unified-scale multidimensional dataset is constructed to facilitate subsequent multi-physics field modeling. Suitable normalization methods are selected for different sensor data (such as electromagnetic signals and chemical composition changes). The normalized electromagnetic, thermal, chemical, and mechanical feature vectors are concatenated according to the time step to form a unified multidimensional data matrix. A high-order tensor structure is used to store multimodal data, preserving the spatiotemporal correlation of the original data. Through outlier detection and repair, feature extraction, data normalization, and multidimensional dataset construction, the quality and availability of multimodal sensor data are systematically improved.

[0084] Preferably, the specific steps of step S22 are:

[0085] S221, performing fast Fourier transform on the initial electromagnetic signal to generate a frequency domain feature vector;

[0086] S222. Perform feature learning on the frequency domain feature vector through a convolutional neural network, and construct a feature matrix to generate a target electromagnetic signal;

[0087] S223, extracting features from the initial temperature distribution data using a spatiotemporal convolutional neural network to generate target temperature distribution data;

[0088] S224, mapping the high-dimensional chemical composition data in the initial chemical composition change data to a low-dimensional space through linear transformation to generate intermediate chemical composition change data;

[0089] S225, using a local linear embedding method to capture the local manifold structure in the intermediate chemical composition change data to generate target chemical composition change data;

[0090] S226, performing wavelet packet decomposition on the initial mechanical vibration data to generate intermediate mechanical vibration data;

[0091] S227 : Calculate the energy value of each frequency band signal in the intermediate mechanical vibration data to generate target mechanical vibration data.

[0092] In an embodiment of the present invention, the initial electromagnetic signal (time series data) is converted into a frequency domain eigenvector to capture the energy distribution of different frequency components. The system focuses on extracting high-frequency components (such as pulse signals generated by partial discharge) and low-frequency components (such as fundamental frequency changes caused by load fluctuations). The system outputs a frequency domain eigenvector containing information such as frequency, amplitude, and phase for subsequent defect identification. By dynamically adjusting the FFT frequency resolution based on the operating status of the wire and cable (such as load fluctuations), the system improves sensitivity to transient faults.

[0093] The frequency domain feature vector is input and local features (such as the high-frequency pulse pattern of partial discharge) are extracted through multiple layers of convolution kernels (such as 1D convolution). Pooling layers (such as max pooling) are used to reduce the feature dimension and retain key information. The feature vectors output by the CNN are concatenated according to the time step to form the feature matrix of the target electromagnetic signal.

[0094] Initial temperature distribution data is fed into a spatiotemporal convolutional neural network, where a 3D convolution kernel is used to simultaneously extract spatial (inter-pixel relationships) and temporal (inter-frame variations) features. An attention mechanism (such as the SE attention module) is used to enhance the feature weights of local hotspots. The output is a feature vector containing spatiotemporal features, representing the target temperature distribution data, which is used for subsequent defect identification.

[0095] Principal component analysis (PCA) is performed on the initial chemical composition change data (such as gas concentration and dielectric constant) to extract the top A principal components (A is typically 5-10) with the highest variance contributions. This generates intermediate chemical composition change data, preserving the core characteristics of insulation aging. Manifold learning is performed on the intermediate chemical composition change data to preserve the local neighborhood relationships of the data and generate target chemical composition change data for subsequent defect classification.

[0096] Perform multi-level wavelet packet decomposition on the initial mechanical vibration data (such as piezoelectric sensor signals) to generate multiple frequency band sub-signals (such as 0.1-1kHz, 1-10kHz). Retain key frequency bands (such as the frequency bands corresponding to mechanical collision and looseness) to generate intermediate mechanical vibration data. Calculate the energy value of each frequency band signal of the intermediate mechanical vibration data to generate target mechanical vibration data (energy vector) for subsequent defect classification. The energy value calculation formula is: E is the total energy of the vibration signal in a specific frequency band, reflecting the vibration intensity of the mechanical system in this frequency band; N is the total number of frequency band signal samples selected when calculating energy; x i is the frequency band signal value of the i-th sampling point.

[0097] Preferably, the multi-physics coupling model includes:

[0098] The electromagnetic field module is used to simulate the current distribution and magnetic field strength in wires and cables and construct the electromagnetic feature space;

[0099] Thermal field module, used to calculate the temperature field distribution and heat conduction path of wires and cables during operation, and to construct thermal feature space;

[0100] Chemical field module, used to quantify the changes in chemical composition and aging rate of insulating materials and construct chemical feature space;

[0101] The mechanical field module is used to analyze the stress-strain relationship of wires and cables under external forces and construct the mechanical feature space.

[0102] In an embodiment of the present invention, the electromagnetic field module is used to simulate the current distribution and magnetic field strength in wires and cables, construct an electromagnetic feature space, and provide key inputs for defects such as partial discharge and insulation degradation. The current density distribution (J) and magnetic field strength (H) in the conductor of the wire and cable are calculated by Maxwell's equations. Combined with the electromagnetic frequency domain feature vector extracted in step S22 (such as the FFT result), an electromagnetic feature space is constructed. The correlation between the high-frequency pulse signal generated by partial discharge and the electromagnetic field distribution (such as the relationship between the harmonic component and the degree of insulation degradation) is quantified. The time domain (current transient response) and frequency domain (partial discharge harmonics) features are combined to improve the accuracy of defect identification. Adjust the electromagnetic field calculation parameters (such as the influence of conductor temperature on resistance) according to the load state of the wire and cable to enhance the adaptability of the model.

[0103] The thermal field module is used to calculate the temperature field distribution and heat conduction path of wires and cables during operation, construct a thermal feature space, and assist in identifying hot spots and aging areas. The temperature distribution (T(x, y, z, t)) of the conductor, insulation layer and environment of the wire and cable is calculated based on the heat transfer equation (such as Fourier's law). Combined with the temperature distribution characteristics extracted by the spatiotemporal convolutional network in step S22 (such as the spatiotemporal correlation of local hot spots), a thermal feature space is constructed. The correlation between the temperature gradient and the aging rate of the insulation material is quantified (such as Arrhenius equation modeling). Combined with the structure of the wire and cable (such as conductor diameter, insulation thickness) and load fluctuations, the heat conduction parameters are dynamically adjusted. The SE attention module (Squeeze-and-Excitation) is used to enhance the feature weight of the hot spot and improve the accuracy of defect location.

[0104] The chemical field module is used to quantify the changes in chemical composition and aging rate in insulating materials, construct a chemical feature space, and provide a basis for insulation life prediction. The concentration data of decomposition products of insulating materials (such as CO2 and aldehyde gases) are collected through gas sensors or electrochemical sensors. Combined with the local manifold structure extracted by manifold learning in step S22 (such as LLE dimensionality reduction results), a chemical feature space is constructed. The correlation between gas concentration and the aging rate of insulating materials is quantified (such as linear regression model or neural network). Combined with the gas concentration data and the change in dielectric constant, the chemical field parameters are dynamically adjusted. Adjust the PCA dimensionality reduction dimension according to the sensor type (such as infrared spectrometer or mass spectrometer) to improve adaptability in different scenarios.

[0105] The mechanical field module is used to analyze the stress-strain relationship of wires and cables under external forces, construct a mechanical feature space, and assist in identifying mechanical damage (such as bending and extrusion). The mechanical vibration signal of the wires and cables is collected by a piezoelectric sensor, and the frequency band energy value extracted by the wavelet packet decomposition in step S22 is combined to construct a mechanical feature space. Quantify the correlation between vibration energy and mechanical damage (such as a sudden increase in energy corresponding to looseness or breakage). Calculate the stress distribution of wires and cables under bending or stretching based on Hooke's law, and adjust the model parameters based on the installation environment (such as pipeline inlet, support point). Dynamically adjust the wavelet packet decomposition parameters (such as basis function selection) according to the installation environment of the wires and cables to improve sensitivity to small deformations. Combine the energy values ​​of multiple frequency bands (such as 0.1-1kHz, 1-10kHz) to construct a feature vector to improve the ability to identify complex mechanical defects.

[0106] Preferably, Figure 3 The flowchart of step S3 of a method for non-destructive online detection of wires and cables is shown.

[0107] In this embodiment of the present invention, data from different modalities (e.g., electromagnetic, thermal, chemical, and mechanical) in a multidimensional dataset are mapped to corresponding feature spaces, forming independent but related target feature spaces. Specifically, the target electromagnetic signal generated in step S22 (e.g., the frequency domain feature matrix extracted by a CNN) is input and mapped to a low-dimensional electromagnetic feature space through a linear transformation (e.g., principal component analysis (PCA)). This yields a target feature space corresponding to the target electromagnetic signal. The dimensionality of the feature space is determined by the number of frequency bands in the electromagnetic signal and the number of CNN output channels.

[0108] The target temperature distribution data generated in step S22 (e.g., the spatiotemporal feature vectors extracted by ST-CNN) is input and mapped to the thermal feature space through time series modeling (e.g., LSTM network) to obtain the target feature space corresponding to the target temperature distribution data. The feature space dimension is determined by the time step and spatial resolution.

[0109] The target chemical composition change data generated in step S22 (e.g., the manifold features after LLE dimensionality reduction) are input and embedded into the chemical feature space through nonlinear mapping (e.g., kernel PCA) to obtain the target feature space corresponding to the target chemical change data. The dimension of the feature space is determined by the number of sensitive components of the chemical sensor.

[0110] The target mechanical vibration data generated in step S22 (e.g., the energy vector after wavelet packet decomposition) is input and mapped to the mechanical feature space through energy normalization (e.g., Min-Max normalization) to obtain the target feature space corresponding to the target mechanical vibration data. The feature space dimension is determined by the number of frequency bands (e.g., 0.1-1 kHz, 1-10 kHz).

[0111] The multiple target feature spaces (electromagnetic, thermal, chemical, and mechanical) generated by S31 are spliced ​​into high-order tensors according to the modal dimension, preserving the spatiotemporal correlation of the original data and obtaining multidimensional feature data. The specific execution process of the tensor decomposition algorithm is as follows: decomposing the high-order tensor into the sum of multiple rank-1 tensors, extracting the potential factors shared across modalities; approximating the original tensor by multiplying the core tensor with the factor matrix, preserving the complex interaction relationship between modalities; the result of the decomposition is multidimensional feature data (such as R K×L , where K is the decomposition rank and L is the number of modes. The decomposed low-dimensional feature vector is used as input to defect recognition models (such as support vector machines and random forests) to improve classification accuracy. For the first time, CP / Tucker decomposition is combined with multi-physics feature space to achieve efficient cross-modal data fusion. The decomposition rank and number of iterations are adjusted based on the operating status of the wire and cable (such as load fluctuations) to improve model adaptability.

[0112] Preferably, Figure 4 The flowchart of step S4 of a method for non-destructive online detection of wires and cables is shown.

[0113] In an embodiment of the present invention, the multidimensional feature data generated in step S3 (such as the low-dimensional feature vector after tensor decomposition) is input, and the top k principal components with the highest variance contribution rate are extracted by PCA (k is usually 5-10). The main characteristic components (such as the harmonic components of the electromagnetic signal, the temperature gradient of the thermal field, and the energy surge of the mechanical vibration) are retained, the noise and redundant information are removed, and the low-dimensional feature matrix is ​​output for subsequent cluster analysis. The number of principal components (k value) is adjusted according to the operating status of the wire and cable (such as load fluctuation) to improve the adaptability in different scenarios. Combined with domain knowledge (such as the aging mechanism of insulating materials), key features (such as high-frequency partial discharge signals) are screened to obtain the main characteristic components to avoid information loss.

[0114] Use algorithms such as K-means or DBSCAN to cluster the main feature components output in step S41. Dynamically adjust the clustering parameters (such as the number of clusters for K-means or the neighborhood radius for DBSCAN) according to the data distribution density. Divide the feature data into normal data clusters (such as the characteristic distribution of healthy wires and cables) and abnormal data clusters (such as the characteristic distribution of partial discharge, overheating, and mechanical damage). Use visualization (such as t-SNE dimensionality reduction graph) to assist in verifying the clustering results. Combine multimodal features such as electromagnetic, thermal, chemical, and mechanical to enhance the sensitivity of clustering to complex defects. Adjust the clustering parameters according to the wire and cable installation environment (such as pipeline inlet, support point) to avoid interference from environmental noise.

[0115] For the abnormal data cluster identified in step S42, the defect location is marked (such as "partial discharge of the insulation layer of the third section of wire and cable") in combination with the physical location information of the wire and cable (such as the installation point coordinates, sensor number). The defect feature description (such as the high-frequency surge of the electromagnetic signal, the hot spot temperature of the thermal field, and the energy anomaly of the mechanical vibration) is extracted. A structured data set is constructed (such as a header containing timestamp, location, feature type, and defect level) to obtain a defect data set for subsequent defect warning and repair decisions. Describe defects in combination with multi-physical field characteristics (such as electromagnetic-thermal-mechanical correlation) to improve the comprehensiveness of diagnosis. Adjust the defect marking threshold according to the operating status of the wire and cable (such as load fluctuation) to avoid false alarms.

[0116] Preferably, the specific steps of step S42 are:

[0117] Step S421: Calculate the density distribution of the main characteristic components using a kernel density estimation method to determine high-density areas and low-density areas in the data space;

[0118] Step S422: using an improved K-means clustering algorithm, taking the high-density area as the initial cluster center, iteratively optimizing the main characteristic components to generate cluster data;

[0119] Step S423: Evaluate the compactness and separation of the cluster data using the silhouette coefficient to generate a silhouette coefficient;

[0120] Step S424: If the silhouette coefficient is greater than the preset silhouette threshold, the clustering result is divided into a normal data cluster and an abnormal data cluster;

[0121] Step S425: If the silhouette coefficient is less than or equal to the preset silhouette threshold, the parameters of the clustering algorithm are adjusted and the process jumps to step S422.

[0122] In an embodiment of the present invention, the low-dimensional feature data after dimensionality reduction in step S41 (such as the principal component vector extracted by PCA) is input to calculate the density value of each data point. The high-density area corresponds to the normal data distribution (such as the characteristic clustering area of ​​healthy wires and cables), and the low-density area corresponds to the abnormal data distribution (such as the outlier of the defective sample). The density threshold (such as the 95% confidence interval) is determined by sliding window or histogram statistics to divide the high / low density areas. The density estimation is optimized by combining multi-physical field characteristics (such as electromagnetic-thermal-mechanical correlation) to enhance the sensitivity to complex defects. The kernel function bandwidth (such as the Gaussian kernel) is adaptively adjusted according to the data distribution density to avoid local noise interference.

[0123] Select the initial cluster center based on the high-density area of ​​S421 (such as K-means++ strategy) to avoid the local optimal problem caused by traditional random initialization. Iteratively update the cluster center by minimizing the Euclidean distance until convergence (such as the number of iterations ≤ 100 or the rate of change < 0.1%). Output cluster labels (such as "normal cluster" and "abnormal cluster") to obtain cluster data and provide input for subsequent evaluation. Combined with multi-physical field density distribution (such as joint modeling of electromagnetic signal density and thermal field density) to improve the rationality of initial center selection. Adjust the number of iterations or convergence threshold according to the complexity of data distribution to improve algorithm efficiency.

[0124] For the cluster data generated in step S422, calculate the silhouette coefficient of each sample Where a is the average distance from the sample to other samples in the same cluster; b is the average distance from the sample to the nearest cluster sample). The overall silhouette coefficient takes the mean of the silhouette coefficients of all samples, ranging from [-1, 1]. The larger the value, the higher the clustering quality. Generate a silhouette coefficient (such as 0.82) for subsequent threshold judgment. Combine the multi-physics field silhouette coefficient (such as the weighted average of the electromagnetic-thermal-mechanical silhouette coefficient) to improve the comprehensiveness of the evaluation. Adjust the silhouette coefficient threshold θ according to the operating status of the wire and cable (such as load fluctuation) to avoid misjudgment. If the silhouette coefficient S>θ (such as θ=0.7), the clustering result is valid and is directly divided into normal data clusters (such as healthy wires and cables) and abnormal data clusters (such as defective samples). If S≤θ, it is considered that the clustering quality is insufficient, and jump to S422 to adjust the parameters and combine the multi-physics field silhouette coefficient (such as the joint threshold of the electromagnetic silhouette coefficient and the thermal field silhouette coefficient) to improve the rationality of the division. Adjust the silhouette coefficient threshold θ according to the installation environment of the wire and cable (such as the pipeline entrance, support point) to avoid environmental noise interference.

[0125] Preferably, Figure 5 The flowchart of step S5 of a method for non-destructive online detection of wires and cables is shown.

[0126] In an embodiment of the present invention, the defect data set (such as defect location, feature description, energy vector) output by step S43 is input, and combined with the historical fault database (such as insulation aging, partial discharge, and mechanical damage records) to construct a Bayesian Network. The causal relationship between defect characteristics (such as electromagnetic signal harmonic components, thermal field temperature gradient) and fault types (such as insulation breakdown and mechanical fracture) is represented by a probabilistic graphical model (PGM). The joint probability distribution of defect characteristics and fault types is calculated to generate a defect-fault association probability matrix. The multi-physical field defect characteristics (such as electromagnetic-thermal-mechanical correlation) and historical fault data are combined to enhance the sensitivity of the model to complex faults. The Bayesian network node weights are dynamically adjusted according to the operating status of the wires and cables (such as load fluctuations) to avoid model obsolescence.

[0127] A Bayesian network inference algorithm (such as the variable elimination method) is used to calculate the conditional probability of a defect characteristic evolving into a fault type. For example, if the defect characteristic is a "sudden increase in partial discharge harmonics," the corresponding probability value of insulation breakdown is calculated. If the probability value is greater than the preset fault threshold, a warning signal (such as an audible and visual alarm, system notification) is triggered. The probability calculation weight is dynamically adjusted by combining the defect characteristic intensity (such as the energy vector amplitude) with the historical fault frequency. The preset fault threshold is adjusted according to the wire and cable installation environment (such as the pipeline entrance and support points) to avoid interference from environmental noise.

[0128] The warning signal is associated with the defect location information to generate a structured warning data set to guide subsequent disposal. The warning data set contains the following fields:

[0129] Defect location: such as "partial discharge of the insulation layer of the third section of wire and cable" (based on the defect location marked in step S43).

[0130] Failure probability: such as "insulation breakdown probability 92%" (from S52 calculation results).

[0131] Warning level: such as "high risk" (based on probability value mapping level rules).

[0132] Recommended disposal measures: such as "disconnect power immediately and replace the insulation layer" (generated in combination with the domain knowledge base).

[0133] Generation of treatment measures: Matching fault types with treatment plans through rule engines or expert systems.

[0134] By combining defect location, probability, and treatment measures, the operability of early warnings is improved. Treatment priorities are adjusted based on the load status of the wires and cables (e.g., high-risk defects are prioritized under high load). Step S5 systematically implements fault prediction and intelligent treatment of wire and cable defects through Bayesian network modeling, multimodal probability calculation, and dynamic early warning generation.

[0135] Preferably, the specific steps of step S52 are:

[0136] Step S521: quantify the uncertainty of the defect-fault association probability matrix using a Monte Carlo simulation method to generate a probability distribution interval;

[0137] Step S522: Calculate the mathematical expectation and variance of the probability distribution interval and construct a confidence interval;

[0138] Step S523: If the lower limit of the confidence interval is greater than the preset fault threshold, a deterministic warning signal is triggered;

[0139] Step S524: If only the upper limit of the confidence interval is greater than the preset fault threshold, an uncertainty warning signal is triggered.

[0140] In an embodiment of the present invention, the uncertainty in the defect-fault association probability matrix is ​​evaluated by the Monte Carlo simulation method, and a probability distribution interval is generated to provide input for subsequent confidence analysis. Specifically, the defect-fault association probability matrix generated in step S51 is input, and each probability value in the matrix is ​​randomly sampled (such as based on Beta distribution or normal distribution) to simulate the uncertainty under different operating conditions (such as wire and cable load fluctuations, environmental noise), and generate a large number of probability samples (such as 10,000 iterations). Construct a probability distribution interval (such as a 95% confidence interval) for the evolution of defects into faults, and output the probability distribution interval to reflect the potential risk range of the evolution of defects into faults. Combined with defect characteristics (such as electromagnetic signal harmonic components) and historical fault data, the parameters of the Monte Carlo simulation (such as distribution type, sampling times) are dynamically adjusted to improve the uncertainty quantification accuracy. Adjust the probability distribution type (such as switching from Beta distribution to Gamma distribution) according to the operating status of the wire and cable (such as load fluctuations) to avoid model obsolescence.

[0141] For the probability distribution interval generated in step S521, calculate the mathematical expectation ( Among them, μ is the mathematical expectation (mean); P g is the probability value of the defect evolving into a fault in the g-th simulation; G is the total number of simulations) and the variance Among them, γ is the variance, which is used to measure the discreteness of the simulation results and reflect the probability fluctuation of the defect evolving into a fault; μ is the mathematical expectation (mean); P g is the probability value of the defect evolving into a fault in the g-th simulation; G is the total number of simulations). Based on the normal distribution assumption, a confidence interval is constructed (e.g., a 95% confidence interval: [μ-1.96γ, μ+1.96γ]), reflecting the core probability range of the defect evolving into a fault. The confidence interval width is dynamically adjusted (e.g., a higher confidence threshold is used for high-risk defects) by combining the defect feature intensity (e.g., energy vector amplitude) with the historical fault frequency. The confidence calculation parameters are adjusted according to the wire and cable installation environment (e.g., pipeline entrance, support point) to avoid interference from environmental noise.

[0142] If the lower limit of the confidence interval is greater than the preset fault threshold (preset fault threshold = 0.85), it is determined that the defect has a high probability of evolving into a fault (such as insulation breakdown, mechanical fracture), triggering a deterministic early warning signal (such as a red alarm). Output a high-priority early warning (such as "immediately cut off the power and replace the insulation layer"), and prompt the operation and maintenance personnel through sound and light alarms, system notifications, etc. Combined with the defect location (such as the third section of wire and cable) and the confidence interval, dynamically adjust the early warning priority (such as giving priority to high-risk defects under high load). Adjust the preset fault threshold according to the operating status of the wire and cable (such as load fluctuations) to avoid misjudgment.

[0143] When the upper limit of the confidence interval exceeds the preset fault threshold (such as 0.85), but the lower limit does not, it indicates that the defect presents a high potential risk (for example, partial discharge may evolve into insulation breakdown), but the current data is insufficient to fully confirm the fault. An uncertainty warning signal (such as a yellow alarm) is generated to alert operations and maintenance personnel to the defect, increase monitoring frequency, or optimize operating parameters to prevent further risk.

[0144] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A non-destructive online detection method for wires and cables, characterized in that: The following steps are involved: Step S1: Collect electromagnetic signals, temperature distribution data, chemical composition change data, and mechanical vibration data of the wires and cables to be tested through a multimodal sensor network; Step S2, performing data preprocessing on the electromagnetic signal, the temperature distribution data, the chemical composition change data, and the mechanical vibration data to construct a multidimensional data set; Step S3, fusing the multidimensional data set through a multi-physics field coupling model to generate multidimensional feature data; Step S4: performing defect identification based on the multidimensional feature data to generate a defect data set; Step S5: Perform dynamic early warning according to the defect data set to generate an early warning data set.

2. A non-destructive online detection method for electric wires and cables according to claim 1, characterized in that: The multimodal sensor network includes electromagnetic sensors, thermal imaging sensors, chemical sensors and mechanical vibration sensors; The electromagnetic sensors are distributed at intervals along the axial direction of the wires and cables, and are used to detect the current distribution and magnetic field changes of the wires and cables; The thermal imaging sensor covers the key nodes of the wires and cables to capture local overheating areas; The chemical sensor is embedded in the insulation layer or sheath material of the wire and cable to monitor aging products of the insulation material; The mechanical vibration sensor is attached to the outer surface of the wire and cable and is used to sense slight deformation of the wire and cable caused by external stress.

3. A non-destructive online detection method for electric wires and cables according to claim 2, characterized in that: The specific steps of step S2 are: Step S21: performing abnormal value detection and repair on the electromagnetic signal, the temperature distribution data, the chemical composition change data, and the mechanical vibration data to generate initial electromagnetic signal, initial temperature distribution data, initial chemical composition change data, and initial mechanical vibration data; Step S22: performing feature extraction on the initial electromagnetic signal, the initial temperature distribution data, the initial chemical composition change data, and the initial mechanical vibration data to generate a target electromagnetic signal, target temperature distribution data, target chemical composition change data, and target mechanical vibration data; Step S23 : normalizing the target electromagnetic signal, the target temperature distribution data, the target chemical composition change data, and the target mechanical vibration data to construct a multidimensional data set.

4. A non-destructive online detection method for electric wires and cables according to claim 3, characterized in that: The specific steps of step S22 are: S221, performing fast Fourier transform on the initial electromagnetic signal to generate a frequency domain feature vector; S222. Perform feature learning on the frequency domain feature vector through a convolutional neural network, and construct a feature matrix to generate a target electromagnetic signal; S223, extracting features from the initial temperature distribution data using a spatiotemporal convolutional neural network to generate target temperature distribution data; S224, mapping the high-dimensional chemical composition data in the initial chemical composition change data to a low-dimensional space through linear transformation to generate intermediate chemical composition change data; S225, using a local linear embedding method to capture the local manifold structure in the intermediate chemical composition change data to generate target chemical composition change data; S226, performing wavelet packet decomposition on the initial mechanical vibration data to generate intermediate mechanical vibration data; S227 : Calculate the energy value of each frequency band signal in the intermediate mechanical vibration data to generate target mechanical vibration data.

5. The method for nondestructive online detection of electric wires and cables according to claim 1, characterized in that: The multi-physics coupling model includes: An electromagnetic field module is used to simulate the current distribution and magnetic field strength in the wires and cables and construct an electromagnetic feature space; A thermal field module is used to calculate the temperature field distribution and heat conduction path of the wires and cables during operation and to construct a thermal feature space; Chemical field module, used to quantify the changes in chemical composition and aging rate of insulating materials and construct chemical feature space; The mechanical field module is used to analyze the stress-strain relationship of the wire and cable under the action of external force and construct a mechanical feature space.

6. A non-destructive online detection method for electric wires and cables according to claim 1 or 5, characterized in that: The specific steps of step S3 are: Step S31: Mapping various types of data in the multidimensional dataset to corresponding feature spaces to generate multiple target feature spaces; Step S32: coupling all the target feature spaces through a tensor decomposition algorithm to generate multi-dimensional feature data.

7. The method for nondestructive online detection of electric wires and cables according to claim 1, characterized in that: The specific steps of step S4 are: Step S41: performing principal component analysis on the multidimensional feature data to reduce the data dimension and extract the main feature components; Step S42: performing cluster analysis on the main characteristic components using a clustering algorithm, and dividing the data into normal data clusters and abnormal data clusters; Step S43: Mark the data in the abnormal data cluster as defect data, and generate a defect data set including defect locations and feature descriptions.

8. A non-destructive online detection method for electric wires and cables according to claim 7, characterized in that: The specific steps of step S42 are: Step S421: Calculate the density distribution of the main characteristic components using a kernel density estimation method to determine high-density areas and low-density areas in the data space; Step S422: using an improved K-means clustering algorithm, taking the high-density area as the initial cluster center, iteratively optimizing the main characteristic components to generate cluster data; Step S423: Evaluate the compactness and separation of the cluster data by using the silhouette coefficient to generate a silhouette coefficient; Step S424: if the silhouette coefficient is greater than a preset silhouette threshold, the clustering result is divided into a normal data cluster and an abnormal data cluster; Step S425: If the silhouette coefficient is less than or equal to the preset silhouette threshold, adjust the parameters of the clustering algorithm and jump to step S422.

9. The method for nondestructive online detection of electric wires and cables according to claim 1, characterized in that: The specific steps of step S5 are: Step S51: Using a Bayesian network to jointly model the defect features in the defect data set and the wire and cable fault history data to generate a defect-fault association probability matrix; Step S52: Calculate the probability value of the defect evolving into an actual fault based on the defect-fault association probability matrix. If the probability value is greater than a preset fault threshold, trigger an early warning signal. Step S53: Associating the warning signal with the defect location information to generate a warning data set including the defect location, failure probability, warning level and recommended disposal measures.

10. A non-destructive online detection method for electric wires and cables according to claim 9, characterized in that: The specific steps of step S52 are: Step S521: quantify the uncertainty of the defect-fault association probability matrix using a Monte Carlo simulation method to generate a probability distribution interval; Step S522: Calculate the mathematical expectation and variance of the probability distribution interval to construct a confidence interval; Step S523: If the lower limit of the confidence interval is greater than the preset fault threshold, a deterministic warning signal is triggered; Step S524: If only the upper limit of the confidence interval is greater than the preset fault threshold, an uncertainty warning signal is triggered.

Images