Phi-OTDR optical cable fault diagnosis method and system based on multi-channel topological data analysis

Abstract

The invention discloses a phi-OTDR optical cable fault diagnosis method and system based on multi-channel topological data analysis, and is applied to the technical field of optical fiber sensing and intelligent operation and maintenance. The method comprises the following steps: acquiring a backward Rayleigh scattering signal of a sensing optical cable to obtain a spatio-temporal data matrix; preprocessing and multi-scale slicing are carried out to obtain spatio-temporal data segments, and initial position information based on the optical time domain reflection principle is added to each spatio-temporal data segment; different types of spatio-temporal data fragments are routed to corresponding topology analysis channels for parallel processing; extracting topological features and positioning information, and fusing the topological features and the positioning information to form optical cable full-state vectors; and outputting a diagnosis result based on the pre-trained intelligent diagnosis model, wherein the diagnosis result comprises a fault type, accurate position information and a health degree evaluation result. According to the method, the positioning robustness in a strong noise environment is effectively improved, and accurate fault classification and early capture and early warning of progressive performance reduction of the optical cable are realized.

Description

Technical Field

[0001] This invention relates to the field of fiber optic sensing and intelligent operation and maintenance technology, and more specifically to a Φ-OTDR optical cable fault diagnosis method and system based on multi-channel topology data analysis. Background Technology

[0002] In power communication and dispatching systems, optical cables serve as the core carrier of information transmission, and their reliability and security directly affect the stable operation of the power grid. In the field of fiber optic sensing, distributed fiber optic sensing technology, especially phase-sensitive optical time-domain reflectometers (Φ-OTDR), has become a key technology for the safety monitoring of power optical cables due to its ability to sense physical quantities such as vibrations and sound waves along the fiber optic path over long distances, continuously and in real time. However, with the expansion of the power grid and the increasing complexity of the environment, the signal processing capabilities of traditional Φ-OTDR systems have gradually revealed several bottlenecks, particularly in fault location, where they can no longer fully meet the high reliability and high precision operation and maintenance requirements of the power system: Firstly, in terms of fault location, most existing fault location methods are based on the simple principle of optical time-domain reflectometry, achieving location by judging the time delay of the backscattered Rayleigh signal. However, power optical cables are often laid in complex environments, easily affected by strong electromagnetic interference, mechanical vibration, temperature changes, and line load fluctuations, leading to signal saturation, distortion, or severe deterioration of the signal-to-noise ratio. Under signal saturation or strong noise backgrounds, the accuracy of traditional time-domain location algorithms will decrease significantly, and they may even fail to effectively identify the fault point, resulting in missed or false alarms, delaying fault repair opportunities, and posing a serious challenge to real-time monitoring of power systems. Secondly, in terms of event identification and classification, most existing solutions rely on manual extraction of time-domain or frequency-domain features for analysis. Power optical cables face diverse fault types, including fiber breakage, bending, compression, and impact from suspended objects, and their signal characteristics often exhibit high similarity in complex noise backgrounds. Traditional methods struggle to deeply capture the essential patterns and subtle differences between various faults, resulting in limited identification accuracy. In particular, they are unable to distinguish between similar events such as construction excavation and ordinary mechanical vibration, or human-caused damage and natural interference, thus hindering the accuracy and efficiency of maintenance responses. Finally, regarding health status assessment, most existing solutions remain at the passive response stage of alarming after a fault occurs. However, the degradation of power optical cables is often a gradual process, such as joint aging, increased microbending loss, and fiber core stress accumulation. Traditional monitoring methods lack the ability to quantitatively assess and analyze the long-term health status of optical cables, making early warning and predictive maintenance impossible. This results in maintenance work always being in a reactive state, making it difficult to prevent potential faults and unable to provide data support for optical cable life assessment and replacement decisions. Therefore, how to provide a Φ-OTDR optical cable fault diagnosis method and system based on multi-channel topology data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0003] In view of this, the present invention provides a Φ-OTDR optical cable fault diagnosis method and system based on multi-channel topology data analysis. By extracting features from multi-channel topology, it performs full-state fusion and intelligent diagnosis to generate a standardized diagnostic report that includes fault type, precise location, severity level, health status, and confidence level.

[0004] To achieve the above objectives, the present invention provides the following technical solution: A method for fault diagnosis of Φ-OTDR optical cables based on multi-channel topology data analysis includes the following steps: S1. Obtain the back Rayleigh scattering signal of the sensing optical cable through the Φ-OTDR system to obtain the spatiotemporal data matrix; S2. Preprocess and multi-scale slice the spatiotemporal data matrix to obtain spatiotemporal data segments corresponding to transient events, periodic vibrations, and long-term degradation, respectively. Add preliminary location information based on the principle of optical temporal reflection to each spatiotemporal data segment. S3. Routing different types of spatiotemporal data fragments to the corresponding topology analysis channels for parallel processing; S4. Extract the topological features and positioning information output from each topology analysis channel and fuse them to form a full-state vector of the optical cable; S5. Input the full state vector of the optical cable into the pre-trained intelligent diagnostic model and output the state diagnostic results of the optical cable, including fault type, precise location information and health assessment results.

[0005] Optionally, the acquisition of the backscattered Rayleigh signal from the sensing optical cable by the Φ-OTDR system in S1 includes three stages: physical process, system operation, and data construction. Specifically: Physical process: Φ-OTDR senses vibrations and sound waves by detecting changes in the intensity and phase of Rayleigh scattered light; System operation: The transmitter generates a series of light pulses, which are injected into the sensing fiber and propagate forward. The data is received by the data acquisition card and the high-speed photodetector. The same position is detected multiple times, and the corresponding return signals are coherently accumulated. Data construction: A one-dimensional distance axis is constructed. The data acquisition card collects data points in chronological order after each optical pulse is emitted, corresponding to a specific physical location on the optical fiber. The intensity and phase distribution of the backscattered Rayleigh signal on the entire optical fiber at the corresponding moment are recorded through the mapping from the time domain to the spatial domain. The detection and acquisition are continuously performed at a fixed pulse repetition frequency. The distance sequence obtained from the first acquisition is used as the first row of the spatiotemporal data matrix, the second acquisition is used as the second row, and so on, to form a two-dimensional spatiotemporal data matrix.

[0006] Optionally, the multi-scale slicing of the spatiotemporal data matrix in S2 is specifically as follows: Multiple criteria are established based on the time-frequency characteristics of the signal to realize the intelligent distribution of spatiotemporal data segments. The three segments correspond to three different judgment criteria. When the maximum absolute value of a time series signal exceeds the amplitude threshold and the duration is shorter than the short-term time threshold, it is determined to be a transient event. Significant spectral peaks are detected by power spectral density. When there are spectral peaks exceeding the threshold and the duration is within the set range, they are determined to be periodic vibrations. When the duration of a signal exceeds a long-term time threshold and the attenuation slope is less than a slope threshold, it is determined to be a long-term degradation.

[0007] Optionally, the topology analysis channels in S3 include: The transient analysis channel constructs spatiotemporal dynamic graphs and extracts graph topology features for transient event data segments, enabling event identification and precise location; the vibration analysis channel reconstructs phase space and performs continuous coherence analysis on periodic vibration data segments, quantifying vibration modes; and the degradation analysis channel performs scaling space topology analysis on the shape of decay curves for long-term trend data segments, assessing health status and locating degradation points.

[0008] Optionally, the transient analysis channel performs spatiotemporal dynamic graph construction and graph topology feature extraction on transient event data segments as follows: Each data acquisition location on the sensing fiber is marked as a point on a map, and each point is connected to its neighboring points with lines to form a spatiotemporal dynamic map. ,in, It is a set of nodes, where each node corresponds to a spatial sampling point. Given a set of edges, connections are established based on spatial proximity. This is a time-varying weight matrix, representing the dynamic correlation strength between nodes; Accurately locate the fault epicenter by calculating nodal betweenness centrality: ; In the formula, Represents a node betweenness centrality, Indicates from node To the node Of all the shortest paths, the one that passes through the node The number of paths, Indicates from node To the node The total number of all shortest paths; The clustering coefficient is used to analyze the local clustering patterns that characterize the impact of faults. For a single node, the formula for measuring the clustering degree of its neighboring nodes is: ; In the formula, Clustering coefficient, This represents the actual number of connections, or nodes. The actual number of connections between neighboring nodes. For nodes Number of neighboring nodes Eigenvector centrality is used to evaluate the global influence of a node in a network. The centrality of eigenvectors is measured by the importance of the node. ; In the formula, Represents a node eigenvector centrality, It is the adjacency matrix of the graph. The first in the adjacency matrix i Line 1 j One element, These are the eigenvalues ​​corresponding to the eigenvectors. It represents the total number of nodes.

[0009] Optionally, the vibration analysis channel performs phase space reconstruction and continuous coherence analysis on periodic vibration data segments as follows: The mutual information method is used to calculate the statistical dependency between the current signal value and its corresponding delayed version to determine the time delay; the pseudo-nearest neighbor method is used to determine the embedding dimension of the reconstruction space; the one-dimensional observation time series is mapped to a high-dimensional phase space through a time delay embedding method, and a time delay embedding method is constructed for each starting time. m A dimensional state vector m To embed the dimension, after processing all valid points, a point cloud composed of points is generated. The geometry of the point cloud is topologically equivalent to the unknown attractor of the original dynamic system, thus restoring the implicit dynamic information from the one-dimensional time domain to the high-dimensional geometric domain. Topological analysis is performed on the reconstructed point cloud to construct a filter shape. Starting from a scale parameter of 0, the scale parameter is gradually increased and the evolution of the point cloud topology is observed. The birth scale b and death scale d of the high-dimensional topological features are recorded to generate a continuous map or barcode. Each feature corresponds to a point or a line segment. By extracting topological invariants, the complex dynamic characteristics of the vibration mode are transformed into noise-robust geometric features. By transforming topological invariants into vibration mode classifications with explicit physical meaning, a multidimensional feature vector is constructed from the output of persistent homology analysis to describe the topological characteristics of the vibration system, including the number of modes, total persistence, uniformity of energy distribution, topological scale, and frequency characteristics. The signal undergoes initial rule-based classification, followed by refined classification based on machine learning. Physical parameter inversion is then performed to map topological features back to traditional physical parameters. The formulas include: Formula for estimating dominant frequency: ; In the formula, The function represents the rounding function, which passes through the characteristic timescale of the most persistent ring. Phase space reconstruction delay The quantitative relationship is used to calculate the dominant vibration frequency; Modal energy ratio formula: ; In the formula, Indicates the first The energy percentage of each vibration mode is based on the persistence length of each characteristic loop. With total durability The ratio quantifies the relative energy contribution of each vibration mode; Stability index formula: ; In the formula, This represents the stability score of the system mode. This represents the variance of different components of the feature vector within the sliding window.

[0010] Optionally, the degradation analysis channel performs scale space topology analysis on the decay curve shape of long-term trend data segments as follows: Constructing the scale space: ; In the formula, The standard deviation is The Gaussian kernel function transforms a one-dimensional curve Convert to two-dimensional scaled space image For each smoothing scale The curve below Extreme point detection and pairing are performed. By identifying local maxima and minima and recording their occurrence and disappearance scales, a set representation describing the intervals where topological features exist is generated. , The scale by which topological features appear. The scale representing the vanishing of topological features is used to calculate the persistence length of each feature based on the feature interval: ; In the formula, Indicates the first n The scale at which a feature disappears. Indicates the first nThe scale at which each feature appears; calculate total persistence: ; In the formula, The sum of the persistence lengths of all individual features is represented by the structural properties of the curve shape quantified by topological persistence entropy: ; ; In the formula, For topological persistence, For the first n The percentage of the persistence length of each feature; the higher the entropy value, the more complex and irregular the topological features of the curve, indicating a defective area; the lower the entropy value, the smoother the curve and the simpler the structure, indicating a healthy state. Degraded segment localization identifies short-lived feature interval clusters near specific locations by comparing the current feature interval set with the healthy baseline interval set. These short-lived feature interval clusters are defined as follows: and, These clusters indicate localized anomalies such as microbending or aging; among them, For the first n The scale at which a feature disappears. For the first n The scale at which each feature appears L Threshold for short-lived feature regions For position The characteristic density at that location, For position The critical threshold at; Health score is calculated by weighted aggregation of multi-scale entropy values: ; In the formula, Indicates the first Weights for each scale Indicates the first Topological entropy at a scale, The historical maximum entropy value is used for trend warnings. By fitting a sliding window to the health score sequence, a warning signal is triggered if a significant downward trend or abrupt change is detected.

[0011] A Φ-OTDR optical cable fault diagnosis system based on multi-channel topology data analysis, applying the aforementioned Φ-OTDR optical cable fault diagnosis method based on multi-channel topology data analysis, includes: The signal acquisition and processing module is used to acquire and preprocess the back Rayleigh scattering signal of the sensing optical cable. The multi-scale slicing routing module is connected to the signal acquisition and processing module to slice the spatiotemporal data matrix at multiple scales and route it to the corresponding topology analysis channel. The topology feature calculation module, connected to the multi-scale slice routing module, is used to perform topology analysis calculations for each channel; The full-state fusion diagnostic module, connected to the topology feature calculation module, is used to construct a full-state vector and perform intelligent diagnosis. The early warning output module, connected to the full-state fusion diagnostic module, is used to generate and output visualized diagnostic reports and alarm information.

[0012] Optionally, the topology feature calculation module includes a transient analysis unit, a vibration analysis unit, and a degradation analysis unit.

[0013] As can be seen from the above technical solution, compared with the prior art, the present invention provides a Φ-OTDR optical cable fault diagnosis method and system based on multi-channel topology data analysis, which has the following beneficial effects: 1. Accurate and reliable positioning: This invention not only performs basic positioning of fault events based on the basic formula of optical time domain reflectance (OTDR), but also achieves stable and accurate ranging of fault points in complex real environments through topology refinement, achieving unprecedented accuracy and robustness; 2. Comprehensive Functionality: This invention integrates transient events, periodic vibration events, and long-term degradation events into a unified framework, outputting a comprehensive diagnostic report that includes location, type, and health score, thus upgrading from simple alarms to comprehensive diagnosis; 3. Early warning and prediction: This invention solves the problem that traditional fault diagnosis methods cannot predict slow degradation faults in optical cables by analyzing the topological characteristics of long-term operating data. It can identify slowly developing defects such as aging of optical cable joints and micro-bending of optical fibers, and realize predictive maintenance. 4. Accurate identification: This invention uses topological data analysis, which pays more attention to the relationship between data. By analyzing the "shape" of the data, it can effectively distinguish complex fault modes that are difficult to identify by traditional methods. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0015] Figure 1 This is a flowchart of the Φ-OTDR optical cable fault diagnosis method based on multi-channel topology data analysis of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] This invention discloses a method for fault diagnosis of Φ-OTDR optical cables based on multi-channel topology data analysis, such as... Figure 1 As shown, it includes the following steps: S1. Obtain the back Rayleigh scattering signal of the sensing optical cable through the Φ-OTDR system to obtain the spatiotemporal data matrix; S2. Preprocess and multi-scale slice the spatiotemporal data matrix to obtain spatiotemporal data segments corresponding to transient events, periodic vibrations, and long-term degradation, respectively. Add preliminary location information based on the principle of optical temporal reflection to each spatiotemporal data segment. S3. Routing different types of spatiotemporal data fragments to the corresponding topology analysis channels for parallel processing; S4. Extract the topological features and positioning information output from each topology analysis channel and fuse them to form a full-state vector of the optical cable; S5. Input the full state vector of the optical cable into the pre-trained intelligent diagnostic model and output the state diagnostic results of the optical cable, including fault type, precise location information and health assessment results.

[0018] In this embodiment, the intelligent diagnostic model is a multi-task integrated machine learning system designed specifically for optical cable health monitoring. It includes a random forest classifier, an XGBoost regressor, and a neural network regression model, which are used for fault type identification, precise location positioning, and quantitative health assessment.

[0019] Furthermore, the acquisition of the backscattered Rayleigh signal from the sensing optical cable by the Φ-OTDR system in S1 includes three stages: physical process, system operation, and data construction. Specifically: Physical process: The physical basis of Φ-OTDR is coherent Rayleigh scattering. External perturbations ultimately manifest as random fluctuations in the intensity of the returned Rayleigh scattered light and abrupt changes in its phase. Therefore, Φ-OTDR senses vibrations and sound waves by detecting changes in the intensity and phase of the Rayleigh scattered light. System operation: The transmitter generates a series of high repetition frequency, extremely narrow width light pulses, which are injected into the sensing fiber and propagate forward. They are received by the data acquisition card and high-speed photodetector. In order to improve the signal-to-noise ratio, the same position is detected multiple times and the corresponding return signals are coherently accumulated. Data construction: A one-dimensional distance axis is constructed. The data acquisition card collects data points in chronological order after each optical pulse is emitted, corresponding to a specific physical location on the optical fiber. The intensity and phase distribution of the backscattered Rayleigh signal on the entire optical fiber at the corresponding moment are recorded through the mapping from the time domain to the spatial domain. The detection and acquisition are continuously performed at a fixed pulse repetition frequency. The distance sequence obtained from the first acquisition is used as the first row of the spatiotemporal data matrix, the second acquisition is used as the second row, and so on, to form a two-dimensional spatiotemporal data matrix.

[0020] In this embodiment, the precise distance to a specific physical location can be calculated using a formula, which is: L =( c ×Δ t ) / (2× n ); In the formula, c For the speed of light, Δ t For the time difference, n The refractive index; A two-dimensional spatiotemporal data matrix can be represented as In the formula, row index t Representing the time dimension, column index x The matrix represents the spatial dimension; the element values ​​quantify the spatial dimension. At a specific moment Specific location The intensity or phase value of the backscattered Rayleigh light at that location.

[0021] After obtaining spatiotemporal data through the system in S1, to eliminate signal distortion caused by light source fluctuations, link attenuation, etc., noise reduction, detrending, and standardization operations are performed to establish a unified and clean data benchmark for subsequent topology data analysis. The specific steps are as follows: Noise reduction: First, the original signal sequence Multi-resolution analysis is performed, selecting appropriate wavelet basis functions and decomposition level N to decompose the signal into coefficients of different frequency sub-bands. All obtained detail coefficients are then analyzed. Quantization is performed using a threshold function. In this embodiment, an improved threshold selection strategy is employed, with the threshold value... The calculation formula is as follows: ; ; In the formula, M For signal length, This is an estimate of the standard deviation of the noise intensity, which can be derived from the data details in Section 1. That is, when the detail coefficients are ini Details of the first section of data when the value is 1; The coefficients are then processed using a soft thresholding function, which is: ; In the formula, This represents the coefficients obtained after the signal is transformed. This function effectively avoids the Gibbs phenomenon and makes the denoised signal smoother. Finally, the processed detail factor will be... The inverse process of wavelet decomposition, along with the approximation coefficients, is used to reconstruct the signal, resulting in the denoised signal. This process preserves signal edge and abrupt change features that are crucial for fault diagnosis.

[0022] Detrend: This step aims to eliminate long-term, slowly varying background components in the signal. These low-frequency trend terms not only interfere with the analysis of the true dynamic characteristics of the vibration signal but also distort the signal geometry upon which topological data analysis relies. This step should first establish a trend phase model, where arbitrary spatial points are fixed within the given trend phase model. The time series signal at point can be represented as: ; In the formula, Represents a fixed spatial point The time series signal at a given time point is a vector containing signal values ​​at multiple time points. t Represents the time vector. They are time points The corresponding signal value.

[0023] Trend Item It can be represented as: ; In the formula, k Let be the order of the polynomial. Let be the coefficients of the polynomial to be determined, and To fully determine the trend term model Trend term parameter estimation is required, and the optimal polynomial coefficients are solved by minimizing the residual sum of squares (RSS) based on the least squares criterion. ; In the formula, express The corresponding actual time series values, The optimal estimate can be obtained by solving this formula for the trend term model prediction value at that time.

[0024] After the above operations, the trend term can be obtained. Subtracting the trend term from the original signal yields the pure signal. This operation effectively removes the long-term, slowly changing components from the signal, making the signal baseline zero.

[0025] Standardized operating procedures: This step is crucial for establishing a consistent discrimination benchmark in multi-channel topology analysis. This process ensures a uniform dimension and scale across different times and spatial locations, making the extracted topological features comparable and improving the classifier's training efficiency and generalization ability. This step applies Z-score normalization to the data matrix X after denoising and detrending. This normalization method linearly transforms the processed data to conform to a normal distribution with a mean of zero and a standard deviation of 1.

[0026] Next, for each column of the data matrix, its mean and standard deviation are calculated. The resulting values ​​are then standardized to unify the signal to a uniform numerical range, eliminating inherent amplitude differences caused by location variations. However, when the data distribution does not satisfy the normal distribution assumption, the min-max standardization method can be used, and its formula can be expressed as: ; In the formula, Indicates the first j The spatial point, the first i The signal value after time standardization Indicates the first j The spatial point, the first i The original signal value at time , where and The first j The method finds the minimum and maximum values ​​of a spatial point signal. It linearly maps the signal to the interval [0,1].

[0027] Furthermore, the multi-scale slicing of the spatiotemporal data matrix in S2 is specifically as follows: Multiple criteria are established based on the time-frequency characteristics of the signal to realize the intelligent distribution of spatiotemporal data segments. The three segments correspond to three different judgment criteria. When the maximum absolute value of a time series signal exceeds the amplitude threshold and the duration is shorter than the short-term time threshold, it is determined to be a transient event. Significant spectral peaks are detected by power spectral density. When there are spectral peaks exceeding the threshold and the duration is within the set range, they are determined to be periodic vibrations. When the duration of a signal exceeds a long-term time threshold and the attenuation slope is less than a slope threshold, it is determined to be a long-term degradation.

[0028] Specifically, the instantaneous event indicator function is: ; In the formula, It is a time series signal. For duration, For amplitude threshold, This is a time threshold; The periodic vibration indication function is: ; In the formula, For long-term time thresholds, The peak threshold; The long-term degradation indicator function is: ; In the formula, This is the slope threshold.

[0029] Furthermore, the topology analysis channels in S3 include: The transient analysis channel constructs spatiotemporal dynamic graphs and extracts graph topology features for transient event data segments, enabling event identification and precise location; the vibration analysis channel reconstructs phase space and performs continuous coherence analysis on periodic vibration data segments, quantifying vibration modes; and the degradation analysis channel performs scaling space topology analysis on the shape of decay curves for long-term trend data segments, assessing health status and locating degradation points.

[0030] Furthermore, for transient events, the channel primarily handles sudden events such as lightning strikes and external damage. Specifically, the transient analysis channel constructs a spatiotemporal dynamic graph and extracts graph topology features from transient event data fragments as follows: Each data acquisition location on the sensing fiber is marked as a point on a map. Each point is then connected to its neighboring points with lines, forming a structure similar to a fishing net or a social network, thus obtaining a spatiotemporal dynamic map. ,in, It is a set of nodes, where each node corresponds to a spatial sampling point. Given a set of edges, connections are established based on spatial proximity. This is a time-varying weight matrix, representing the dynamic correlation strength between nodes; In this embodiment, the signal similarity between any two adjacent points at the moment of the fault is calculated. If the two points are very similar, a high weight is assigned to the line connecting them; otherwise, a low weight is assigned. The time-varying weight matrix is ​​an adaptive Gaussian kernel function, as shown in the following formula: ; In the formula, , They are nodes i and j At any moment t signal strength, This is a scaling parameter that is adaptively adjusted based on the signal's dynamic range. After constructing the spatiotemporal state diagram, graph topological features are extracted to identify the most critical points and regional features. In this step, the fault epicenter is accurately located by calculating the betweenness centrality of nodes, the clustering coefficient is analyzed to characterize the local clustering patterns of the fault's impact, and the eigenvector centrality is used to assess the global influence of nodes in the network. These three factors combined constitute a "topological fingerprint" describing the spatiotemporal propagation structure of the fault. This provides a key criterion for subsequent intelligent diagnosis to distinguish the nature of the fault, achieving a cognitive leap from traditional signal analysis to network relationship insight. Accurately locate the fault epicenter by calculating nodal betweenness centrality: ; In the formula, Represents a node betweenness centrality, Indicates from node To the node Of all the shortest paths, the one that passes through the node The number of paths, Indicates from node To the node The total number of all shortest paths; The clustering coefficient is used to analyze the local clustering patterns that characterize the impact of faults. For a single node, the formula for measuring the clustering degree of its neighboring nodes is: ; In the formula, Clustering coefficient, This represents the actual number of connections, or nodes. The actual number of connections between neighboring nodes. For nodes Number of neighboring nodes Eigenvector centrality is used to evaluate the global influence of a node in a network. The centrality of eigenvectors is measured by the importance of the node. ; In the formula, Represents a node eigenvector centrality, It is the adjacency matrix of the graph. The first in the adjacency matrix i Line 1 j One element, These are the eigenvalues ​​corresponding to the eigenvectors. It represents the total number of nodes.

[0031] The above operations yielded a series of quantified feature values, which collectively describe the distribution pattern and core strength of faults on the network. To further diagnose and pinpoint the location of faults, event identification and localization are necessary. In this embodiment, the node with the highest betweenness centrality is directly identified; this node is the radiation center of the entire fault energy, i.e., the exact location of the fault. The fault is then identified, and the extracted features are combined into a fault fingerprint, which is input into a pre-trained intelligent classifier to determine the type of fault. When the energy is too high, traditional positioning methods may fail. This invention also provides an improved method for calculating betweenness centrality, the formula of which is as follows: ; In the formula, based on the original formula, the denominator is multiplied by... ,Right now s arrive t The topological distance is used, and this improved algorithm can maintain positioning accuracy even when the signal is distorted.

[0032] Furthermore, for periodic vibration events, this channel primarily handles periodic events such as galloping and light wind vibrations. Before this step, signal selection and preparation are necessary. A time series signal at a representative spatial location can be selected from the preprocessed spatiotemporal data matrix. This selected location is typically in the region of most pronounced vibration, i.e., the location of the most significant galloping. By selecting this location, a one-dimensional time series can be obtained. Number of sampling points; The vibration analysis channel performs phase space reconstruction and continuous homology analysis on periodic vibration data segments as follows: The theoretical basis for phase space reconstruction is Takens' embedding theorem, which states that the topological equivalent structure of the original high-dimensional dynamical system can be reconstructed using the delay coordinates of a univariate time series, without needing to measure all dimensions; a single-dimensional time series is sufficient to restore the core dynamic characteristics of the system. The first step in reconstruction is to determine the parameters, the core of which is finding the time delay and the embedding dimension. The mutual information method is used to calculate the statistical dependency between the current signal value and its corresponding delayed version to determine the time delay; the pseudo-nearest neighbor method is used to determine the embedding dimension of the reconstruction space; the one-dimensional observation time series is mapped to a high-dimensional phase space through a time delay embedding method, and a time delay embedding method is constructed for each starting time. m A dimensional state vector m To embed the dimension, after processing all valid points, a point cloud composed of points is generated. The geometry of the point cloud is topologically equivalent to the unknown attractor of the original dynamic system, thus restoring the implicit dynamic information from the one-dimensional time domain to the high-dimensional geometric domain. In this embodiment, the specific formula for calculating statistical dependency is as follows: ; In the formula, For time delay The mutual information value below, For signal The marginal probability, The overall statistical dependence between the measures and The smaller the value, the lower the redundancy and the stronger the independence, which ensures that the reconstructed coordinates have sufficient correlation while avoiding information redundancy. As the dimension increases, the proportion of false neighbor points generated by projection decreases. When this proportion is lower than a preset threshold, it can be considered that the dimension is sufficient to fully expand the attractor of the dynamical system without producing self-intersections, thereby ensuring that the reconstructed phase space is topologically equivalent to the original system.

[0033] Topological analysis is performed on the reconstructed point cloud to construct a filter. In this embodiment, the Vietoris-Rips filter is used, which is a family of simple complexes nested with increasing scale. Starting from a scale parameter of 0, the scale parameter is gradually increased, and the evolution of the point cloud topology is observed. The birth scale b and death scale d of the high-dimensional topological features are recorded to generate a persistence map or barcode. Each feature corresponds to a point or a line segment. By extracting topological invariants, the complex dynamic characteristics of the vibration mode are transformed into noise-robust geometric features. In this embodiment, the topological invariants include the number of significant cycles, total persistence, and persistence entropy. By transforming topological invariants into vibration mode classifications with explicit physical meaning, a multidimensional feature vector is constructed from the output of persistent homology analysis to describe the topological characteristics of the vibration system, including the number of modes, total persistence, uniformity of energy distribution, topological scale, and frequency characteristics. The number of modes directly corresponds to the number of independent and stable oscillation modes in a vibration system; total persistence measures the overall energy and stability of all vibration modes; a larger value indicates stronger and more stable vibration; uniform energy distribution means that when energy is concentrated in one dominant mode, it is simple periodic vibration, and when energy is evenly distributed across all modes, it is complex vibration; topological scale measures the overall scale of topological features across the entire scale range; frequency characteristics reflect the relationship between vibration modes and the overall time characteristics of the signal; The signal undergoes initial rule-based classification, followed by refined classification based on machine learning. Physical parameter inversion is then performed to map topological features back to traditional physical parameters. The formulas include: Formula for estimating dominant frequency: ; In the formula, The function represents the rounding function, which passes through the characteristic timescale of the most persistent ring. Phase space reconstruction delay The quantitative relationship is used to calculate the dominant vibration frequency; Modal energy ratio formula: ; In the formula, Indicates the first The energy percentage of each vibration mode is based on the persistence length of each characteristic loop. With total durability The ratio quantifies the relative energy contribution of each vibration mode; Stability index formula: ; In the formula, This represents the stability score of the system mode. Let be the variance of different components of the eigenvector within the sliding window. This formula combines the variances of each component of the eigenvector within the sliding window to generate a stability score between 0 and 1, where a score close to 1 indicates a stable pattern and a score close to 0 indicates drastic change.

[0034] In the final diagnostic output stage, the system integrates all the above analysis results into a structured report, which includes several aspects such as mode type, number of modes, dominant frequency, energy distribution, stability score, confidence level, and recommended measures.

[0035] In this embodiment of the invention, the preliminary classification based on rules is as follows: when the signal features a single dominant frequency and the phase space trajectory is a smooth limiting loop, it can be used as a criterion for simple periodic vibration; when the signal is nonlinearly coupled with multiple frequencies and the phase space trajectory is a high-dimensional torus, it can be used as a criterion for multimodal coupled vibration; when the frequency ratio is an irrational number and the phase space trajectory is densely filled on the torus, it can be used as a criterion for quasi-periodic vibration; and others can be used for chaotic signal processing. The fine-grained classification based on machine learning specifically involves using a labeled training dataset, expressed by the following formula: ; In the formula, H For the labeled training dataset, For category labels, For the first i The feature vectors of each sample include simple periodicity, multimodality, quasi-periodicity, and chaos. A classifier is trained using this dataset; in this embodiment, the classifier uses a support vector machine, and its decision function can be expressed as: ; In the formula, For support vector coefficients, For the first n The class labels of the training samples, where the kernel function... Used to handle nonlinearly separable problems, and for coefficients By solving optimization problems And satisfy constraints The model determines the decision function by solving a constrained optimization problem, which uses kernel tricks to map features to a high-dimensional space to find the optimal classification hyperplane. Finally, the confidence level of the classification result is evaluated by calculating the relative distance from the sample to the decision boundary. A confidence level close to 1 indicates reliable classification, while a confidence level close to 0 indicates that the sample is near the classification boundary, and the result has high uncertainty. The formula for calculating the confidence level is: ; For confidence level, The absolute value of the decision function for the current sample. It is the maximum absolute value of the decision function for all training samples.

[0036] Furthermore, for long-term degradation events, this channel mainly focuses on the slow aging of optical cables, signal attenuation, and potential defects. The degradation analysis channel performs scaling space topology analysis on the attenuation curve shape of long-term trend data segments as follows: Constructing the scale space: ; In the formula, The standard deviation is Gaussian kernel function, Take a series of incrementing values, corresponding to different smoothing scales, smaller ones Preserve details, larger Highlighting the overall trend, the one-dimensional curve Convert to two-dimensional scaled space image For each smoothing scale The curve below Extreme point detection and pairing are performed. By identifying local maxima and minima and recording their occurrence and disappearance scales, a set representation describing the intervals where topological features exist is generated. , The scale by which topological features appear. The scale representing the vanishing of topological features is used to calculate the persistence length of each feature based on the feature interval: ; In the formula, Indicates the first n The scale at which a feature disappears. Indicates the first nThe scale at which each feature appears; calculate total persistence: ; In the formula, The sum of the persistence lengths of all individual features is represented by the structural properties of the curve shape quantified by topological persistence entropy: ; ; In the formula, For topological persistence, For the first n The percentage of the persistence length of each feature; the higher the entropy value, the more complex and irregular the topological features of the curve, indicating a defective area; the lower the entropy value, the smoother the curve and the simpler the structure, indicating a healthy state. Degraded segment localization identifies short-lived feature interval clusters near specific locations by comparing the current feature interval set with the healthy baseline interval set. These short-lived feature interval clusters are defined as follows: and, These clusters indicate localized anomalies such as microbending or aging; among them, For the first n The scale at which a feature disappears. For the first n The scale at which each feature appears L Threshold for short-lived feature regions For position The characteristic density at that location, For position The critical threshold at that point.

[0037] Health score is calculated by weighted aggregation of multi-scale entropy values: ; In the formula, Indicates the first Weights for each scale Indicates the first Topological entropy at a scale, The historical maximum entropy value is used for trend warnings. By fitting a sliding window to the health score sequence, a warning signal is triggered if a significant downward trend or abrupt change is detected.

[0038] Furthermore, S4 unifies and integrates the features and location information obtained from the three channels into a complete optical cable status feature package, providing standardized input data for the final intelligent diagnosis; After S3 processes the transient event channel, periodic vibration event channel, and long-term degradation channel respectively, this step standardizes and integrates the output topological features and location information to construct a unified and complete mathematical description—the optical cable full-state vector—providing structured input for the final comprehensive diagnosis. First, feature standardization and alignment are performed. For spatial alignment, this means unifying the preliminary location information of each channel to the same fiber optic mileage coordinate system. The formula for this location information is as follows: ; In the formula, Represents fiber optic mileage coordinates. At the speed of light, For the time difference, The refractive index of the optical fiber is given; then, feature normalization is performed to normalize the topological features (graph node centrality, persistence length, topological entropy, etc.) with different numerical ranges and dimensions.

[0039] After standardizing the three types of features, they are concatenated into a high-order vector. This high-dimensional vector is then encapsulated into a complete topological map of the optical cable in three dimensions: transient time, vibration event, and degradation event. This realizes a transformation from multi-source heterogeneous data to unified feature expression, laying the foundation for the final output. Furthermore, S5 analyzes the full-state characteristics through an intelligent model, outputting a comprehensive diagnostic report that includes fault type, precise location, health score, and maintenance recommendations, achieving a closed loop from data to decision-making. Its core task is to use the full-state feature vector built based on S4 to output a comprehensive diagnostic conclusion on the optical cable's health status through an intelligent analysis model. First, fault identification and classification are performed. A trained classification model performs pattern recognition on the input feature vector, outputting the fault type and probability. Then, a quantitative health assessment is conducted, outputting interpretable quantitative health indicators including a comprehensive health score and risk level. The comprehensive health score is a weighted score based on the degree of feature deviation from the baseline, and the risk level is classified into low, medium, and high levels, dynamically adjusted based on the rate of trend change. Next, predictive maintenance recommendations are given, which are forward-looking outputs generated based on trend extrapolation of time-series features. Finally, the system achieves a closed-loop intelligent diagnosis from raw data to the entire chain of "identification-location-assessment-prediction."

[0040] and Figure 1 Corresponding to the method described above, this invention also discloses a Φ-OTDR optical cable fault diagnosis system based on multi-channel topology data analysis. The application of the aforementioned Φ-OTDR optical cable fault diagnosis method based on multi-channel topology data analysis includes: The signal acquisition and processing module is used to acquire and preprocess the back Rayleigh scattering signal of the sensing optical cable. The multi-scale slicing routing module is connected to the signal acquisition and processing module to slice the spatiotemporal data matrix at multiple scales and route it to the corresponding topology analysis channel. The topology feature calculation module, connected to the multi-scale slice routing module, is used to perform topology analysis calculations for each channel; The full-state fusion diagnostic module, connected to the topology feature calculation module, is used to construct a full-state vector and perform intelligent diagnosis. The early warning output module, connected to the full-state fusion diagnostic module, is used to generate and output visualized diagnostic reports and alarm information.

[0041] Furthermore, the topology feature calculation module includes a transient analysis unit, a vibration analysis unit, and a degradation analysis unit.

[0042] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0043] Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for fault diagnosis of Φ-OTDR optical cables based on multi-channel topology data analysis, characterized in that, Includes the following steps: S1. Obtain the back Rayleigh scattering signal of the sensing optical cable through the Φ-OTDR system to obtain the spatiotemporal data matrix; S2. Preprocess and multi-scale slice the spatiotemporal data matrix to obtain spatiotemporal data segments corresponding to transient events, periodic vibrations, and long-term degradation, respectively. Add preliminary location information based on the principle of optical temporal reflection to each spatiotemporal data segment. S3. Routing different types of spatiotemporal data fragments to the corresponding topology analysis channels for parallel processing; S4. Extract the topological features and positioning information output from each topology analysis channel and fuse them to form a full-state vector of the optical cable; S5. Input the full state vector of the optical cable into the pre-trained intelligent diagnostic model and output the state diagnostic results of the optical cable, including fault type, precise location information and health assessment results.

2. The Φ-OTDR optical cable fault diagnosis method based on multi-channel topology data analysis according to claim 1, characterized in that, The acquisition of the backscattered Rayleigh signal from the sensing optical cable by the Φ-OTDR system in S1 involves three stages: physical process, system operation, and data construction. Specifically: Physical process: Φ-OTDR senses vibrations and sound waves by detecting changes in the intensity and phase of Rayleigh scattered light; System operation: The transmitter generates a series of light pulses, which are injected into the sensing fiber and propagate forward. The data is received by the data acquisition card and the high-speed photodetector. The same position is detected multiple times, and the corresponding return signals are coherently accumulated. Data construction: A one-dimensional distance axis is constructed. The data acquisition card collects data points in chronological order after each optical pulse is emitted, corresponding to a specific physical location on the optical fiber. The intensity and phase distribution of the backscattered Rayleigh signal on the entire optical fiber at the corresponding moment are recorded through the mapping from the time domain to the spatial domain. The detection and acquisition are continuously performed at a fixed pulse repetition frequency. The distance sequence obtained from the first acquisition is used as the first row of the spatiotemporal data matrix, the second acquisition is used as the second row, and so on, to form a two-dimensional spatiotemporal data matrix.

3. The Φ-OTDR optical cable fault diagnosis method based on multi-channel topology data analysis according to claim 1, characterized in that, The multi-scale slicing of the spatiotemporal data matrix in S2 is specifically as follows: Multiple criteria are established based on the time-frequency characteristics of the signal to realize the intelligent distribution of spatiotemporal data segments. The three segments correspond to three different judgment criteria. When the maximum absolute value of a time series signal exceeds the amplitude threshold and the duration is shorter than the short-term time threshold, it is determined to be a transient event. Significant spectral peaks are detected by power spectral density. When there are spectral peaks exceeding the threshold and the duration is within the set range, they are determined to be periodic vibrations. When the duration of a signal exceeds a long-term time threshold and the attenuation slope is less than a slope threshold, it is determined to be a long-term degradation.

4. The Φ-OTDR optical cable fault diagnosis method based on multi-channel topology data analysis according to claim 1, characterized in that, The topology analysis channels in S3 include: The transient analysis channel constructs spatiotemporal dynamic graphs and extracts graph topology features for transient event data segments, enabling event identification and precise location; the vibration analysis channel reconstructs phase space and performs continuous coherence analysis on periodic vibration data segments, quantifying vibration modes; and the degradation analysis channel performs scaling space topology analysis on the shape of decay curves for long-term trend data segments, assessing health status and locating degradation points.

5. The Φ-OTDR optical cable fault diagnosis method based on multi-channel topology data analysis according to claim 4, characterized in that, The transient analysis channel performs spatiotemporal dynamic graph construction and graph topology feature extraction on transient event data fragments as follows: Each data acquisition location on the sensing fiber is marked as a point on a map, and each point is connected to its neighboring points with lines to form a spatiotemporal dynamic map. ,in, It is a set of nodes, where each node corresponds to a spatial sampling point. Given a set of edges, connections are established based on spatial proximity. This is a time-varying weight matrix, representing the dynamic correlation strength between nodes; Accurately locate the fault epicenter by calculating nodal betweenness centrality: ； In the formula, Represents a node betweenness centrality, Indicates from node To the node Of all the shortest paths, the one that passes through the node The number of paths, Indicates from node To the node The total number of all shortest paths; The clustering coefficient is used to analyze the local clustering patterns that characterize the impact of faults. For a single node, the formula for measuring the clustering degree of its neighboring nodes is: ； In the formula, Clustering coefficient, This represents the actual number of connections, or nodes. The actual number of connections between neighboring nodes. For nodes Number of neighboring nodes Eigenvector centrality is used to evaluate the global influence of a node in a network. The centrality of eigenvectors is measured by the importance of the node. ； In the formula, Represents a node eigenvector centrality, It is the adjacency matrix of the graph. The first in the adjacency matrix i Line number j One element, These are the eigenvalues ​​corresponding to the eigenvectors. It represents the total number of nodes.

6. The Φ-OTDR optical cable fault diagnosis method based on multi-channel topology data analysis according to claim 4, characterized in that, The vibration analysis channel performs phase space reconstruction and continuous homology analysis on periodic vibration data segments as follows: The mutual information method is used to calculate the statistical dependency between the current signal value and the corresponding delayed version to determine the time delay; the pseudo-nearest neighbor method is used to determine the embedding dimension of the reconstruction space. One-dimensional observation time series are mapped to a high-dimensional phase space using a time-delay embedding method. For each initial time, a... m A dimensional state vector m To embed the dimension, after processing all valid points, a point cloud composed of points is generated. The geometry of the point cloud is topologically equivalent to the unknown attractor of the original dynamic system, thus restoring the implicit dynamic information from the one-dimensional time domain to the high-dimensional geometric domain. Topological analysis is performed on the reconstructed point cloud to construct a filter shape. Starting from a scale parameter of 0, the scale parameter is gradually increased and the evolution of the point cloud topology is observed. The birth scale b and death scale d of the high-dimensional topological features are recorded to generate a persistent map or barcode. Each feature corresponds to a point or a line segment. By extracting topological invariants, the complex dynamic characteristics of the vibration mode are transformed into noise-robust geometric features. By transforming topological invariants into vibration mode classifications with explicit physical meaning, a multidimensional feature vector is constructed from the output of persistent homology analysis to describe the topological characteristics of the vibration system, including the number of modes, total persistence, uniformity of energy distribution, topological scale, and frequency characteristics. The signal undergoes initial rule-based classification, followed by refined classification based on machine learning. Physical parameter inversion is then performed to map topological features back to traditional physical parameters. The formulas include: Formula for estimating dominant frequency: ； In the formula, The function represents the rounding function, which passes through the characteristic timescale of the most persistent ring. Phase space reconstruction delay The quantitative relationship is used to calculate the dominant vibration frequency; Modal energy ratio formula: ； In the formula, Indicates the first The energy percentage of each vibration mode is based on the persistence length of each characteristic loop. With total durability The ratio quantifies the relative energy contribution of each vibration mode; Stability index formula: ； In the formula, This represents the stability score of the system mode. This represents the variance of different components of the feature vector within the sliding window.

7. The Φ-OTDR optical cable fault diagnosis method based on multi-channel topology data analysis according to claim 4, characterized in that, The degradation analysis channel performs scale-space topology analysis on the shape of the decay curve for long-term trend data segments as follows: Constructing the scale space: ； In the formula, The standard deviation is The Gaussian kernel function transforms a one-dimensional curve Convert to two-dimensional scaled space image For each smoothing scale The curve below Extreme point detection and pairing are performed. By identifying local maxima and minima and recording their occurrence and disappearance scales, a set representation describing the intervals where topological features exist is generated. , The scale by which topological features appear. The scale representing the vanishing of topological features is used to calculate the persistence length of each feature based on the feature interval: ； In the formula, Indicates the first n The scale at which a feature disappears. Indicates the first n The scale at which each feature appears; Calculate total durability: ； In the formula, The sum of the persistence lengths of all individual features is represented by the structural properties of the curve shape quantified by topological persistence entropy: ； ； In the formula, For topological persistence, For the first n The percentage of the persistence length of each feature; the higher the entropy value, the more complex and irregular the topological features of the curve, indicating a defective area; the lower the entropy value, the smoother the curve and the simpler the structure, indicating a healthy state. Degraded segment localization identifies clusters of short-lived feature intervals near specific locations by comparing the current feature interval set with the healthy baseline interval set. Short-lived feature interval clusters are defined as and, These clusters indicate localized anomalies such as microbending or aging; among them, For the first n The scale at which a feature disappears. For the first n The scale at which each feature appears L Threshold for short-lived feature regions For position The characteristic density at that location, For position The critical threshold at; Health score is calculated by weighted aggregation of multi-scale entropy values: ； In the formula, Indicates the first Weights for each scale Indicates the first Topological entropy at a scale, The maximum entropy value in history is used; the trend warning is triggered by fitting a sliding window to the health score sequence and if a significant downward trend or abrupt change is detected, an early warning signal is triggered.

8. A Φ-OTDR optical cable fault diagnosis system based on multi-channel topology data analysis, characterized in that, The Φ-OTDR optical cable fault diagnosis method based on multi-channel topology data analysis according to any one of claims 1-7 includes: The signal acquisition and processing module is used to acquire and preprocess the back Rayleigh scattering signal of the sensing optical cable. The multi-scale slicing routing module is connected to the signal acquisition and processing module to slice the spatiotemporal data matrix at multiple scales and route it to the corresponding topology analysis channel. The topology feature calculation module, connected to the multi-scale slice routing module, is used to perform topology analysis calculations for each channel; The full-state fusion diagnostic module, connected to the topology feature calculation module, is used to construct a full-state vector and perform intelligent diagnosis. The early warning output module, connected to the full-state fusion diagnostic module, is used to generate and output visualized diagnostic reports and alarm information.

9. The Φ-OTDR optical cable fault diagnosis system based on multi-channel topology data analysis according to claim 8, characterized in that, The topology feature calculation module includes a transient analysis unit, a vibration analysis unit, and a degradation analysis unit.

Images