Network cable fault diagnosis system and method based on signal analysis

Abstract

The invention relates to the technical field of network signals, in particular to a network cable fault diagnosis system and method based on signal analysis, in the network cable fault diagnosis system and method based on the signal analysis, multi-scale convolution extraction is carried out on amplitude and frequency information of network cable time sequence signals, disturbance of different periodic characteristics can be distinguished, and therefore accurate recognition of signal abnormal forms is achieved, classification is carried out at the same time, and the network cable fault diagnosis accuracy is improved. The method comprises the following steps of: constructing a characteristic expression with a rich hierarchical structure through fusion, so that an abnormal trend of a signal in a transmission process can be revealed in a clearer form, modeling a relation between network cable nodes by virtue of a graph structure, and calculating a characteristic propagation and connection relation between the nodes, the propagation path of the fault influence in the network structure can be quantitatively evaluated, the path topology entropy is used for judging the path complexity, the redundant nodes are removed, the influence sorting is combined, the prominent expression of the key fault path is strengthened, the fault type can be quickly locked during fault classification and position information extraction, and the fault classification efficiency is improved. And a specific influence area can be determined.

Description

Technical Field

[0001] The present invention relates to the technical field of network signals, and in particular to a network cable fault diagnosis system and method based on signal analysis. Background Art

[0002] The field of network signal technology refers to the technology of transmitting and communicating information, data, or signals through wireless transmission media such as radio waves or light waves. This field covers a variety of technologies and concepts, including wireless local area networks (WLAN), Bluetooth, long-term evolution (LTE), fifth-generation mobile communication networks (5G), near-field communication (NFC), radio frequency identification (RFID), and global positioning systems (GPS). The signal analysis-based network cable fault diagnosis system aims to improve the maintenance efficiency of local area network cabling and device connections. By implementing an automated and visual fault identification mechanism, this solution is committed to reducing network downtime, improving device availability, and reducing operation and maintenance costs. It is particularly suitable for high-density network scenarios such as data centers, smart buildings, and industrial sites. Existing network cable fault diagnosis systems are unable to accurately identify and classify abnormal signal patterns during signal transmission, resulting in misjudgment of fault causes and low operation and maintenance efficiency. Summary of the Invention

[0003] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a network cable fault diagnosis system and method based on signal analysis.

[0004] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: a network cable fault diagnosis system based on signal analysis includes: Signal disturbance identification module: Receives network cable timing signals through sensors and uses a multi-scale convolutional neural network to extract the amplitude and frequency information of each sampling point. It determines whether there are short-term sharp disturbances or long-term trend changes. By comparing with the reference signal, it confirms the signal type and generates a signal disturbance classification identifier. Multi-scale feature fusion module: Based on the signal disturbance classification identifier, the module performs multi-scale convolution on the signal using different convolution kernels to extract short-period mutations, medium impedance offsets, and long-period trends. The feature maps of each scale are concatenated and fused through 1×1 convolution to generate a fused multi-scale feature map. Fault propagation graph analysis module: Based on the fused multi-scale feature graph, a graph convolutional network is used to construct a network node graph, calculate the connection relationship between nodes and propagate features, and optimize the relationship between nodes through graph convolution operations to generate a node propagation feature graph; Path optimization module: Based on the node propagation feature graph, it calculates the path topology entropy and determines whether the path meets the threshold. If the threshold is exceeded, it removes redundant nodes by pruning, optimizes the path based on the node influence ranking, and generates an optimized fault path. Diagnosis confirmation and decision module: Based on the optimized fault path, classify the fault mode and confirm the fault location, determine the subsequent processing steps, and generate a fault diagnosis confirmation result.

[0005] As a further solution of the present invention, the signal disturbance identification module includes: Signal amplitude extraction submodule: Receives network cable timing signals through sensors, uses a multi-scale convolutional neural network to calculate the amplitude of each sampling point, extracts the maximum and average values of the time domain signal, further calculates the frequency components, and generates amplitude and frequency characteristics; Signal period determination submodule: Based on the amplitude and frequency characteristics, the collected signal is analyzed for periodicity, and a periodic detection method is used to evaluate the periodic changes of the signal to determine whether it is a short-period sharp disturbance or a long-period trend change, and generate a periodic feature identifier; Signal type classification submodule: Based on the periodic feature identifier, the collected signal is compared with the reference signal point by point to determine whether it belongs to a specific type such as a short-circuit reflection spike, a poor contact waveform, etc., and a signal disturbance classification identifier is generated.

[0006] As a further solution of the present invention, the multi-scale convolutional neural network constructs three parallel convolution paths, and uses 3×1, 7×1, and 15×1 convolution kernels to perform convolution operations on the waveform sequence of each sampling point, respectively, to extract disturbance features of different scales such as short-period spikes, impedance changes, and slow-changing trends, and enhances the response strength of the key channels in each path through the channel attention module. The feature maps of the three convolution paths are then spliced, and a 1×1 convolution operation is performed on the spliced results to compress the channel dimension, fuse the disturbance expressions at different scales, and finally establish a cross-scale disturbance feature expression result.

[0007] As a further solution of the present invention, the multi-scale feature fusion module includes: Short-period feature extraction submodule: Based on the signal disturbance classification identifier, a 3×1 convolution kernel is used to perform a convolution operation on the signal to extract the features of the high-frequency mutation part, calculate the fluctuation amount in each time window, and generate a short-period feature map; Medium- and long-period feature extraction submodule: Based on the signal disturbance classification identifier, the signal is convolved using 7×1 and 15×1 convolution kernels, respectively, to extract medium-scale impedance changes and long-period trends, and the amplitude change rate and average change in the window are calculated to generate medium- and long-period feature maps; Multi-scale feature fusion submodule: Based on the short-period feature map and the medium- and long-period feature map, the two are spliced and feature compressed and fused through a 1×1 convolution operation, and the feature map output by the convolution is combined to generate a fused multi-scale feature map.

[0008] As a further solution of the present invention, the fault propagation graph analysis module includes: Node relationship construction submodule: Based on the fused multi-scale feature map, a graph convolutional network is used to extract the connection points of the network cable signal. By analyzing the connection path between the signal acquisition point of each node and the node, the physical distance between the nodes, signal transmission delay and other parameters are calculated. The connection relationship between the nodes is constructed and annotated to generate a node connection relationship graph; Feature propagation submodule: Based on the node connection relationship graph, it transmits the signal characteristics of each node to adjacent nodes, calculates the strength and influence range of signal propagation, and updates the propagation relationship between nodes according to certain rules to ensure the correct flow of signals in the graph and generate node propagation characteristics; Relationship optimization submodule: Based on the node propagation characteristics, by calculating the weights of the propagation characteristics between nodes, the signal interaction relationship between nodes is analyzed, the propagation path and strength between nodes are adjusted, the effect of signal propagation is optimized, and a node propagation feature graph is generated.

[0009] As a further solution of the present invention, the graph convolutional network selects each signal collection point in the network cable as a node in the graph, constructs an initial adjacency structure, and establishes an edge weight matrix by calculating the physical distance between nodes, signal transmission delay and connection mode. In each round of propagation, the feature information of each node is weighted and aggregated with the features of the adjacent nodes according to the connection weight, and a weighted sum operation is performed between the node's own features and the aggregated features to update the node status, ultimately forming a node connection relationship graph that reflects the global signal propagation structure.

[0010] As a further solution of the present invention, the path optimization module includes: Path topology calculation submodule: Based on the node propagation feature graph, it performs topological structure analysis on each node in the path, calculates the connectivity of the path, evaluates the topological entropy of each path, and analyzes the propagation efficiency through the connection relationship between nodes to generate a path topology entropy graph; Path pruning submodule: Based on the path topology entropy graph, evaluate the redundant nodes and inefficient paths in the path, calculate the entropy value of each path, and perform pruning operations if the topology entropy of the path exceeds the set threshold, remove redundant nodes, optimize the path structure, and generate an optimized path graph; Path sorting submodule: Based on the optimized path graph, analyze the influence of each node, sort the paths by importance, prioritize the critical paths, determine the most important paths as the final diagnostic paths, and generate the optimized fault paths.

[0011] As a further solution of the present invention, the pruning operation calculates the sum of the edge weights, propagation delay value and received signal strength for each node, and compares the differences to identify nodes with low contribution, weak connection strength or abnormal delay during the propagation process. If the node simultaneously meets the conditions that the edge weight is lower than the average edge weight, the signal strength is lower than the set lower limit and it does not constitute a critical hop, it is marked as a redundant node. Subsequently, the node and the connecting edge are removed from the path graph, and the previous and next hops are reconnected to retain the main path structure, and finally a path graph with structural optimization is output.

[0012] As a further solution of the present invention, the diagnosis confirmation and decision module includes: Fault mode classification submodule: Based on the optimized fault path, by comparing the propagation characteristics of each path node one by one, the matching degree with the known fault mode is determined. According to the similarity of the matching characteristics, the type of the current fault is classified, such as short circuit, poor contact, etc., and the fault mode classification result is generated.

[0013] Fault location confirmation submodule: Based on the fault mode classification results, analyze each node in the path, locate the most likely fault source and extension area based on the signal strength and delay propagated between nodes, determine the influence relationship between the fault point and surrounding nodes based on characteristic propagation, and generate the fault location.

[0014] Subsequent processing decision submodule: Based on the fault location, analyze the impact range and priority of the fault point, determine the required processing measures according to the fault type, such as repair, replacement or further investigation, evaluate the urgency of the processing steps, and generate fault diagnosis confirmation results.

[0015] A network cable fault diagnosis method based on signal analysis is implemented based on the above-mentioned network cable fault diagnosis system based on signal analysis, and includes the following steps: S1: Based on the network cable timing signal, the amplitude and frequency information of each sampling point is extracted, the maximum amplitude, frequency distribution, and time interval are calculated, and the presence of short-period spike disturbances or long-period slow-changing trends is determined. The characteristics are compared with the reference waveform signal to generate a periodic disturbance type label. S2: Based on the periodic disturbance type label, a multi-scale convolutional neural network is used to perform three-scale window division on the signal, calculate the fluctuation mean, amplitude offset and direction change rate at each scale, establish a multi-scale feature sequence, perform splicing and compression operations on multiple feature path outputs, and generate a fused scale feature map; S3: Based on the fused scale feature map, a graph convolutional network is used to locate the sampling point position in the network node structure diagram, extract the time feature sequence corresponding to each node, analyze the connection path length, signal delay, and interference coefficient value between nodes, construct a node connection relationship diagram and propagate node information values, update the node propagation strength and frequency response, and generate a topological propagation feature structure; S4: Based on the topological propagation feature structure, perform structural entropy calculation on all subgraphs in the connection path, identify paths with structural entropy greater than a threshold, prune redundant connections by comparing node influence values, select the path with the minimum entropy weight and perform path priority sorting to generate a path screening optimization graph sequence; S5: Based on the path screening optimization graph sequence, the propagation characteristics between the first node and the last node of each path are extracted, the node response strength judgment and type mapping are performed, the current fault mode type and the key propagation starting point position are determined, and the processing operation code is generated according to the fault mode number and propagation direction to establish the fault diagnosis confirmation result.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are: 1. This invention uses multi-scale convolution to extract the amplitude and frequency information of network cable timing signals, distinguishing disturbances with different periodic characteristics. This allows for accurate identification of signal anomalies, simultaneous classification, and fusion to construct a feature representation with a rich hierarchical structure, enabling a clearer understanding of abnormal trends in signal transmission. 2. This method uses a graph structure to model the relationships between network nodes. By combining feature propagation and connection calculations between nodes, it enables a quantitative assessment of the propagation paths of fault impacts within the network structure. Path topology entropy is used to determine path complexity. By removing redundant nodes and combining influence ranking, the redundant structure is further compressed, enhancing the prominence of critical fault paths. 3. This invention not only rapidly identifies the fault type and extracts location information during fault classification, but also pinpoints the specific affected area, providing precise positioning support for operations and maintenance. The hierarchical expression of signal anomaly characteristics, optimized inter-node structure, and clear fault path identification collectively improve the efficiency and accuracy of fault identification, while significantly reducing the false positive rate and time consumption. This has a positive impact on the precise diagnosis of complex faults in high-density network environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a system flow chart of the present invention; Figure 2 Schematic diagram of the system framework of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0019] See also Figure 1 The present invention provides a technical solution: a network cable fault diagnosis system based on signal analysis includes: Signal disturbance identification module: Receives network cable timing signals through sensors and uses a multi-scale convolutional neural network to extract the amplitude and frequency information of each sampling point. It determines whether there are short-term sharp disturbances or long-term trend changes. By comparing with the reference signal, it confirms the signal type and generates a signal disturbance classification identifier. Multi-scale feature fusion module: Based on the signal disturbance classification and identification, it uses different convolution kernels to perform multi-scale convolution on the signal to extract short-period mutations, medium impedance offsets, and long-period trends. It then concatenates the feature maps at each scale and fuses them through 1×1 convolution to generate a fused multi-scale feature map. Fault propagation graph analysis module: Based on the fused multi-scale feature graph, a graph convolutional network is used to construct a network node graph, calculate the connection relationships between nodes and propagate features. Graph convolution operations are used to optimize the relationships between nodes and generate a node propagation feature graph. Path optimization module: Based on the node propagation feature graph, it calculates the path topology entropy and determines whether the path meets the threshold. If the threshold is exceeded, it removes redundant nodes through pruning, optimizes the path based on the node influence, and generates the optimized fault path. Diagnosis confirmation and decision module: Based on the optimized fault path, classify the fault mode and confirm the fault location, determine the subsequent processing steps, and generate the fault diagnosis confirmation result.

[0020] See also Figure 2 , the signal disturbance identification module includes: Signal amplitude extraction submodule: Receives network cable timing signals through sensors, uses a multi-scale convolutional neural network to calculate the amplitude of each sampling point, extracts the maximum and average values of the time domain signal, further calculates the frequency components, and generates amplitude and frequency characteristics; Signal period determination submodule: Based on the amplitude and frequency characteristics, the collected signal is analyzed for periodicity. The periodicity detection method is used to evaluate the periodic changes of the signal, determine whether it is a short-period sharp disturbance or a long-period trend change, and generate a periodic feature identifier. Signal type classification submodule: Based on periodic feature identification, the collected signal is compared with the reference signal point by point to determine whether it belongs to a specific type such as a short-circuit reflection spike or a poor contact waveform, and generates a signal disturbance classification identification; Signal amplitude extraction submodule: Based on the network cable timing signal, a multi-scale convolutional neural network is used to construct three sets of convolution kernel operation paths for the input signal. Convolution kernels of sizes 3×1, 7×1, and 15×1 are used to slide each sampling window in sequence, and a one-dimensional convolution operation is performed on the 5-point amplitude sequence in each window. A step size of 1 and zero-filling are used to maintain the original sequence length. The output results of each convolution path are taken as the channel mean and normalized to a decimal between 0 and 1. After merging the channel outputs, a maximum pooling operation is performed on them to extract the maximum amplitude. Mean pooling is used to obtain the window average amplitude. Fast Fourier transform is then called to perform frequency component analysis on each 8-point data segment in the window. The first three main frequencies and corresponding amplitudes are extracted as frequency domain features. Finally, the frequency domain features and the pooled amplitude features are spliced together to form a node signal expression vector to generate amplitude and frequency features. Signal period determination submodule: Based on the amplitude and frequency characteristics, the peak spacing calculation method is used to perform peak detection on the maximum amplitude sequence of each node signal. The minimum peak interval is set to 5 sampling points, and the minimum amplitude change threshold is 0.1 times the input signal amplitude mean. By detecting whether the sampling point interval between two consecutive peaks is stable in the range of 3 to 6 points and occurs more than 5 times in a row, it is determined to be a short-period spike disturbance. If the peak interval exceeds 15 points and the amplitude changes slowly, it is determined to be a long-period slow-changing trend. Finally, according to the detection results, the period type is marked as spike type or slow-changing type in the node structure to generate a periodic feature identifier; Signal type classification submodule: Based on periodic feature identification, the point-by-point distance comparison method is used to calculate the Euclidean distance between the amplitude of the input signal and the reference signal at each moment. The window width is set to 10 sampling points, and the distance values in each window are averaged. If the average distance is less than the set threshold of 0.15 and the period type is a spike type, it is classified as a short-circuit reflection spike. If the distance value is stably distributed in the range of 0.1 to 0.2 and the period type is a slow-changing type, it is classified as a poor contact slow-changing waveform. After the classification is completed, the result is stored in the node fault label field to generate a signal disturbance classification identifier.

[0021] A multi-scale convolutional neural network constructs three parallel convolution paths, and uses convolution kernels of 3×1, 7×1, and 15×1 sizes to perform convolution operations on the waveform sequence of each sampling point, respectively. It extracts disturbance features of different scales, such as short-period spikes, impedance changes, and slow-changing trends. The channel attention module is used to enhance the response strength of the key channels in each path. The feature maps of the three convolution paths are then spliced together, and a 1×1 convolution operation is performed on the spliced result to compress the channel dimension. The disturbance expressions at different scales are integrated to finally establish a cross-scale disturbance feature expression result.

[0022] See also Figure 2 , the multi-scale feature fusion module includes: Short-period feature extraction submodule: Based on the signal disturbance classification identification, a 3×1 convolution kernel is used to perform a convolution operation on the signal to extract the features of the high-frequency mutation part, calculate the fluctuation amount in each time window, and generate a short-period feature map; Medium- and long-period feature extraction submodule: Based on the signal disturbance classification identification, the signal is convolved using 7×1 and 15×1 convolution kernels, respectively, to extract medium-scale impedance changes and long-period trends. The amplitude change rate and average change within the window are calculated to generate medium- and long-period feature maps. Multi-scale feature fusion submodule: Based on the short-term feature map and the medium- and long-term feature map, the two are spliced and the features are compressed and fused through a 1×1 convolution operation. The fused multi-scale feature map is generated by combining the feature map output by the convolution. Short-term feature extraction submodule: Based on the signal disturbance classification identification, a multi-scale convolutional neural network is used to construct the first channel convolution path for the input signal. The convolution kernel size is 3×1, the step size is set to 1, the padding method is set to symmetric zero padding, and the sliding window width is 3 sampling points. In each convolution operation, the amplitude change sequence within the window is extracted, and the result is activated by ReLU and the feature map is output. Then, the sliding window fluctuation difference calculation is performed on every 10 convolution results. The calculation formula is the difference between the maximum and minimum amplitudes in each window. All window fluctuations are marked as feature response maps to generate short-term feature maps. Medium- and long-term feature extraction submodule: Based on the signal perturbation classification and identification, a multi-scale convolutional neural network is used to establish the second and third channel convolution paths respectively. A 7×1 convolution kernel is used for the second channel, and a 15×1 convolution kernel is used for the third channel. The step size is set to 1 in the convolution operation, and zero padding is used to maintain the output length. The convolution output is activated by ReLU to extract the medium impedance offset and long-term trend. In each sliding window of 20 points, the amplitude change rate of the convolution output is calculated and normalized by the cumulative difference between adjacent points. At the same time, the mean calculation is performed on each window to mark the change trend characteristic value. The output results of the two channels are merged to generate a medium- and long-term feature map. Multi-scale feature fusion submodule: Based on the short-cycle feature map and the medium- and long-cycle feature map, a multi-scale convolutional neural network is used. The two feature maps are first spliced in the channel dimension to obtain a fused feature stack with a channel number of 2. Then, a 1×1 convolution kernel is used to perform channel-by-channel compression convolution on the fused features. In the convolution operation, the step size is set to 1 and the number of channels is compressed to 1. The output result is then standardized again, and the normalization constants are set to 0 and the standard deviation to 1. Finally, the integrated convolution feature results are output to generate a fused multi-scale feature map.

[0023] See also Figure 2 ,The fault propagation graph analysis module includes: Node relationship construction submodule: Based on the fused multi-scale feature map, a graph convolutional network is used to extract the connection points of the network cable signal. By analyzing the connection path between each node's signal collection point and the node, the physical distance between nodes, signal transmission delay and other parameters are calculated. The connection relationship between nodes is constructed and annotated to generate a node connection relationship diagram. Feature propagation submodule: Based on the node connection relationship graph, it transmits the signal characteristics of each node to adjacent nodes, calculates the strength and influence range of signal propagation, and updates the propagation relationship between nodes according to certain rules to ensure the correct flow of signals in the graph and generate node propagation characteristics; Relationship optimization submodule: Based on the node propagation characteristics, by calculating the weights of the propagation characteristics between nodes, the signal interaction relationship between nodes is analyzed, the propagation path and strength between nodes are adjusted, the effect of signal propagation is optimized, and a node propagation feature graph is generated; Node relationship construction submodule: Based on the fused multi-scale feature map, a graph convolutional network is used to set each physical port and intermediate test point of the network cable as an independent node in the graph structure. The three-dimensional Euclidean distance calculation is performed on the location coordinates of each node, and the straight-line distance between it and all adjacent nodes is recorded. Nodes with a distance of less than 3 meters in each section of the wiring path are set as directly connected edge relationships and attached with a weight value of 1. If the distance is greater than 3 meters and less than 10 meters, the weight is set to 0.5. If there is a difference in the connection medium, a multiplication factor of 0.75 is added to the edge weight. The signal transmission delay between nodes is calculated at 5 meters per microsecond. The obtained weight matrix is merged with the adjacency matrix to construct the graph structure. An index table is established according to the node sequence number and the graph initialization operation is performed to generate a node connection relationship graph. Feature propagation submodule: Based on the node connection relationship graph, a graph convolutional network is used to represent the initial signal features of each node as a one-dimensional tensor with a length of 128. In each round of propagation, the current tensor value of the node and the tensor value of its adjacent nodes are called to perform a weighted sum operation, using the connection edge weight as the weight coefficient. A self-connection mechanism is introduced in the convolutional propagation to multiply the node's own features by 1.0 and add them to the propagation sum. The propagation result is activated using a linear transformation function and normalized to between 0 and 1. This propagation operation is performed on all nodes in sequence for one round. After two rounds of propagation, the current feature state tensor of the node is output to generate the node propagation feature; Relationship optimization submodule: Based on the node propagation characteristics, a graph convolutional network is used to reassign the edge weights between nodes, and weighted adjustment is performed according to the cosine similarity between the signal propagation values in the node pairs. If the similarity is greater than 0.9, the edge weight is set to 1. If the similarity is between 0.5 and 0.9, the edge weight is updated to 0.5. If it is less than 0.5, it is set to 0 and regarded as a weak connection. After removing the weak connection edges, the connected components of the entire graph structure are recalculated, the largest subgraph is retained and the adjacency list is updated. The feature states in all node propagation paths are reconstructed and compressed into a 64-dimensional tensor to generate a node propagation feature graph.

[0024] The graph convolutional network selects each signal collection point in the network cable as a node in the graph, constructs the initial adjacency structure, and establishes an edge weight matrix by calculating the physical distance between nodes, signal transmission delay and connection mode. In each round of propagation, the feature information of each node is weighted and aggregated with the features of the adjacent nodes according to the connection weight, and a weighted sum operation is performed between the node's own features and the aggregated features to update the node status, ultimately forming a node connection relationship graph that reflects the global signal propagation structure.

[0025] See also Figure 2 , the path optimization module includes: Path topology calculation submodule: Based on the node propagation feature graph, it performs topological structure analysis on each node in the path, calculates the connectivity of the path, evaluates the topological entropy of each path, analyzes the propagation efficiency through the connection relationship between nodes, and generates a path topology entropy graph; Path pruning submodule: Based on the path topology entropy graph, it evaluates redundant nodes and inefficient paths in the path and calculates the entropy value of each path. If the topological entropy of the path exceeds the set threshold, it performs a pruning operation to remove redundant nodes, optimize the path structure, and generate an optimized path graph. Path sorting submodule: Based on the optimized path graph, it analyzes the influence of each node, sorts the paths by importance, prioritizes the critical paths, determines the most important paths as the final diagnostic paths, and generates the optimized fault paths. Path topology calculation submodule: Based on the node propagation feature graph, the topology entropy calculation method is used to construct a directed path sequence for all nodes in each path according to the connection order. The connection edge weights between each adjacent node are normalized, and each path is traversed. The edge weight ratio is counted and the connection value range of each node in the path is recorded. The complexity of the path is determined by the distribution of the number of node connection differences and the change in the edge weight ratio. The connection ratio distribution and node connection structure encoding corresponding to each path are stored in a two-dimensional array table. A structure record table is established according to the path number. The set of all path topology states is output to generate a path topology entropy graph. Path pruning submodule: Based on the path topology entropy graph, the path structure entropy pruning method is adopted to check the node records in the path structure table one by one, obtain the propagation edge weight, connection frequency and total number of connections between each node and its adjacent nodes, and evaluate whether the edge weight ratio of each node in the path is lower than the overall average edge weight. At the same time, it is determined whether the node has a unique connection structure in the path. If it meets the conditions of weak edge connection and non-critical path jump point, the node is removed and the node index is recorded. The connection repair processing is performed on the remaining adjacent nodes, the adjacent edge relationship is rebuilt and the propagation direction is retained. After completing the node pruning in all paths, the updated path set is established to generate the optimized path graph; Path sorting submodule: Based on the optimized path graph, the Katz centrality sorting method is used to establish an adjacency matrix for all nodes in the path. The initial node score is set to a uniform value, and the score is iteratively transferred according to the adjacency matrix structure. In each round, the sum of the adjacency scores received by each node is counted and the scores are accumulated and updated. A cyclic scoring calculation is performed on all nodes until convergence. The average score of all nodes in each path is then taken as the basis for path sorting. A priority order table is established for all paths according to the score. The path with the highest score is identified as the most critical propagation path, generating the optimized fault path.

[0026] See also Figure 2 , pruning operation, calculates the sum of edge weights, propagation delay value and received signal strength for each node, and compares the differences to identify nodes with low contribution, weak connection strength or abnormal delay in the propagation process. If the node satisfies the conditions that the edge weight is lower than the average edge weight, the signal strength is lower than the set lower limit and it does not constitute a key jump point, it is marked as a redundant node. Then, the node and the connecting edge are removed from the path graph, and the previous and next jump points are reconnected to retain the main path structure. Finally, the path graph after structural optimization is output.

[0027] See also Figure 2 , the diagnosis confirmation and decision-making module includes: Fault mode classification submodule: Based on the optimized fault path, by comparing the propagation characteristics of each path node one by one, the matching degree with the known fault mode is judged. According to the similarity of the matching characteristics, the type of current fault is classified, such as short circuit, poor contact, etc., and the fault mode classification result is generated.

[0028] Fault location confirmation submodule: Based on the fault mode classification results, it analyzes each node in the path, locates the most likely fault source and extension area based on the signal strength and delay propagated between nodes, and determines the influence relationship between the fault point and surrounding nodes based on characteristic propagation to generate the fault location.

[0029] Subsequent processing decision submodule: Based on the fault location, analyze the impact scope and priority of the fault point, determine the required processing measures based on the fault type, such as repair, replacement or further investigation, evaluate the urgency of the processing steps, and generate fault diagnosis confirmation results; Fault mode classification submodule: Based on the optimized fault path, the K-nearest neighbor classification method is used to construct a standard fault mode sample set. The sample feature vector dimension is set to 64, including features such as node propagation frequency, adjacent node response difference, and path length. For each path node to be classified, feature vectors of the same dimension are sequentially extracted as input. The number of neighbors is set to K=5. The Euclidean distance function is used to calculate the distance between the input feature and each sample in the sample set. The fault type label is obtained for the five samples with the smallest distance. The current node type is determined by majority voting. The node type results are combined and classified and merged at the path level, marked with standard labels such as short circuit, poor contact, and breakpoint to generate the fault mode classification result. The fault location confirmation submodule uses a signal propagation reverse location algorithm based on the fault mode classification results to traverse the nodes of each fault path in reverse order. The signal response strength of each node and the number of transmission interval samples required to propagate to adjacent nodes are calculated, starting from the last node to the first node. The propagation delay value of each hop is recorded and the total propagation delay of the path is superimposed. All nodes are then scored based on their propagation delay difference and the decreasing trend of signal strength. Nodes with the fastest signal attenuation and the shortest propagation time are selected as source point candidates. Based on the path propagation trend, the node with the largest change is marked as the fault starting point, and the continuous structure before and after it is the expansion area to generate the fault location. Subsequent processing decision submodule: Based on the location of the fault, a rule-based priority decision method is adopted to extract the path level, location depth, propagation strength and type label of the fault point. The corresponding processing items in the rule table are compared. The path type of short circuit is marked as breakpoint repair priority, the path type of poor contact is marked as connection confirmation priority, and the node whose propagation strength decreases at a speed exceeding the threshold is set as replacement priority. At the same time, the response priority is set based on the path level number and node order, the processing instruction numbers are arranged in sequence to form an operation sequence, and the fault diagnosis confirmation result is generated.

[0030] The network cable fault diagnosis method based on signal analysis is performed based on the above-mentioned network cable fault diagnosis system based on signal analysis, and includes the following steps: S1: Based on the network cable timing signal, the amplitude and frequency information of each sampling point is extracted, the maximum amplitude, frequency distribution, and time interval are calculated, and the presence of short-period spike disturbances or long-period slow-changing trends is determined. The characteristics are compared with the reference waveform signal to generate a periodic disturbance type label. S2: Based on the periodic disturbance type label, a multi-scale convolutional neural network is used to perform three-scale window division on the signal. The fluctuation mean, amplitude offset and direction change rate at each scale are calculated to establish a multi-scale feature sequence. Multiple feature path outputs are spliced and compressed to generate a fused scale feature map. S3: Based on the fusion scale feature map, a graph convolutional network is used to locate the sampling point position in the network node structure diagram, extract the time feature sequence corresponding to each node, analyze the connection path length, signal delay, and interference coefficient between nodes, construct a node connection relationship diagram and propagate node information values, update the node propagation strength and frequency response, and generate a topological propagation feature structure; S4: Based on the topological propagation feature structure, the structural entropy of all subgraphs in the connection path is calculated, and the paths with structural entropy greater than the threshold are identified. The redundant connections are pruned by comparing the node influence values, and the path with the minimum entropy weight is selected and the path priority is sorted to generate a path screening optimization graph sequence. S5: Based on the path screening optimization graph sequence, the propagation characteristics between the first and last nodes of each path are extracted, the node response strength judgment and type mapping are performed, the current fault mode type and the key propagation starting point position are determined, the processing operation code is generated according to the fault mode number and propagation direction, and the fault diagnosis confirmation result is established.

[0031] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A network cable fault diagnosis system based on signal analysis, characterized in that: The system comprises: Signal disturbance identification module: Receives network cable timing signals through sensors and uses a multi-scale convolutional neural network to extract the amplitude and frequency information of each sampling point. It determines whether there are short-term sharp disturbances or long-term trend changes. By comparing with the reference signal, it confirms the signal type and generates a signal disturbance classification identifier. Multi-scale feature fusion module: Based on the signal disturbance classification identifier, the module performs multi-scale convolution on the signal using different convolution kernels to extract short-period mutations, medium impedance offsets, and long-period trends. The feature maps of each scale are concatenated and fused through 1×1 convolution to generate a fused multi-scale feature map. Fault propagation graph analysis module: Based on the fused multi-scale feature graph, a graph convolutional network is used to construct a network node graph, calculate the connection relationship between nodes and propagate features, and optimize the relationship between nodes through graph convolution operations to generate a node propagation feature graph; Path optimization module: Based on the node propagation feature graph, it calculates the path topology entropy and determines whether the path meets the threshold. If the threshold is exceeded, it removes redundant nodes by pruning, optimizes the path based on the node influence ranking, and generates an optimized fault path. Diagnosis confirmation and decision module: Based on the optimized fault path, classify the fault mode and confirm the fault location, determine the subsequent processing steps, and generate a fault diagnosis confirmation result.

2. The network cable fault diagnosis system based on signal analysis according to claim 1, characterized in that: The signal disturbance identification module includes: Signal amplitude extraction submodule: Receives network cable timing signals through sensors, uses a multi-scale convolutional neural network to calculate the amplitude of each sampling point, extracts the maximum and average values of the time domain signal, further calculates the frequency components, and generates amplitude and frequency characteristics; Signal period determination submodule: Based on the amplitude and frequency characteristics, the collected signal is analyzed for periodicity, and a periodic detection method is used to evaluate the periodic changes of the signal to determine whether it is a short-period sharp disturbance or a long-period trend change, and generate a periodic feature identifier; Signal type classification submodule: Based on the periodic feature identifier, the collected signal is compared with the reference signal point by point to determine whether it belongs to a specific type such as a short-circuit reflection spike, a poor contact waveform, etc., and a signal disturbance classification identifier is generated.

3. The network cable fault diagnosis system based on signal analysis according to claim 2, characterized in that: The multi-scale convolutional neural network constructs three parallel convolution paths, and uses convolution kernels of 3×1, 7×1, and 15×1 sizes to perform convolution operations on the waveform sequence of each sampling point, respectively, to extract disturbance features of different scales, such as short-period spikes, impedance changes, and slow-changing trends. The response strength of the key channels in each path is enhanced through the channel attention module. The feature maps of the three convolution paths are then spliced together, and a 1×1 convolution operation is performed on the spliced result to compress the channel dimension, fuse the disturbance expressions at different scales, and finally establish a cross-scale disturbance feature expression result.

4. The network cable fault diagnosis system based on signal analysis according to claim 1, characterized in that: The multi-scale feature fusion module includes: Short-period feature extraction submodule: Based on the signal disturbance classification identifier, a 3×1 convolution kernel is used to perform a convolution operation on the signal to extract the features of the high-frequency mutation part, calculate the fluctuation amount in each time window, and generate a short-period feature map; Medium- and long-period feature extraction submodule: Based on the signal disturbance classification identifier, the signal is convolved using 7×1 and 15×1 convolution kernels, respectively, to extract medium-scale impedance changes and long-period trends, and the amplitude change rate and average change in the window are calculated to generate medium- and long-period feature maps; Multi-scale feature fusion submodule: Based on the short-period feature map and the medium- and long-period feature map, the two are spliced and feature compressed and fused through a 1×1 convolution operation, and the feature map output by the convolution is combined to generate a fused multi-scale feature map.

5. The network cable fault diagnosis system based on signal analysis according to claim 4, characterized in that: The fault propagation graph analysis module includes: Node relationship construction submodule: Based on the fused multi-scale feature map, a graph convolutional network is used to extract the connection points of the network cable signal. By analyzing the connection path between the signal acquisition point of each node and the node, the physical distance between the nodes, signal transmission delay and other parameters are calculated. The connection relationship between the nodes is constructed and annotated to generate a node connection relationship graph; Feature propagation submodule: Based on the node connection relationship graph, it transmits the signal characteristics of each node to adjacent nodes, calculates the strength and influence range of signal propagation, and updates the propagation relationship between nodes according to certain rules to ensure the correct flow of signals in the graph and generate node propagation characteristics; Relationship optimization submodule: Based on the node propagation characteristics, by calculating the weights of the propagation characteristics between nodes, the signal interaction relationship between nodes is analyzed, the propagation path and strength between nodes are adjusted, the effect of signal propagation is optimized, and a node propagation feature graph is generated.

6. The network cable fault diagnosis system based on signal analysis according to claim 5, characterized in that: The graph convolutional network selects each signal collection point in the network cable as a node in the graph, constructs an initial adjacency structure, and establishes an edge weight matrix by calculating the physical distance between nodes, signal transmission delay and connection mode. In each round of propagation, the feature information of each node is weighted and aggregated with the features of the adjacent nodes according to the connection weight, and a weighted sum operation is performed between the node's own features and the aggregated features to update the node status, ultimately forming a node connection relationship graph that reflects the global signal propagation structure.

7. The network cable fault diagnosis system based on signal analysis according to claim 1, characterized in that: The path optimization module includes: Path topology calculation submodule: Based on the node propagation feature graph, it performs topological structure analysis on each node in the path, calculates the connectivity of the path, evaluates the topological entropy of each path, and analyzes the propagation efficiency through the connection relationship between nodes to generate a path topology entropy graph; Path pruning submodule: Based on the path topology entropy graph, evaluate the redundant nodes and inefficient paths in the path, calculate the entropy value of each path, and perform pruning operations if the topology entropy of the path exceeds the set threshold, remove redundant nodes, optimize the path structure, and generate an optimized path graph; Path sorting submodule: Based on the optimized path graph, analyze the influence of each node, sort the paths by importance, prioritize the critical paths, determine the most important paths as the final diagnostic paths, and generate the optimized fault paths.

8. The network cable fault diagnosis system based on signal analysis according to claim 7, characterized in that: The pruning operation calculates the sum of edge weights, propagation delay value, and received signal strength for each node, and compares the differences to identify nodes with low contribution, weak connection strength, or abnormal delay during the propagation process. If the node satisfies the conditions that the edge weight is lower than the average edge weight, the signal strength is lower than the set lower limit, and it does not constitute a critical hop, it is marked as a redundant node. Subsequently, the node and connecting edges are removed from the path graph, and the previous and next hops are reconnected to retain the main path structure. Finally, a path graph with structural optimization is output.

9. The network cable fault diagnosis system based on signal analysis according to claim 1, characterized in that: The diagnosis confirmation and decision-making module includes: Fault mode classification submodule: Based on the optimized fault path, by comparing the propagation characteristics of each path node one by one, the matching degree with the known fault mode is determined. According to the similarity of the matching characteristics, the type of the current fault is classified, such as short circuit, poor contact, etc., and the fault mode classification result is generated. Fault location confirmation submodule: Based on the fault mode classification results, analyze each node in the path, locate the most likely fault source and extension area based on the signal strength and delay propagated between nodes, determine the influence relationship between the fault point and surrounding nodes based on characteristic propagation, and generate the fault location. Subsequent processing decision submodule: Based on the fault location, analyze the impact range and priority of the fault point, determine the required processing measures according to the fault type, such as repair, replacement or further investigation, evaluate the urgency of the processing steps, and generate fault diagnosis confirmation results.

10. A network cable fault diagnosis method based on signal analysis, characterized in that: The network cable fault diagnosis system based on signal analysis according to any one of claims 1 to 9 comprises the following steps: S1: Based on the network cable timing signal, the amplitude and frequency information of each sampling point is extracted, the maximum amplitude, frequency distribution, and time interval are calculated, and the presence of short-period spike disturbances or long-period slow-changing trends is determined. The characteristics are compared with the reference waveform signal to generate a periodic disturbance type label. S2: Based on the periodic disturbance type label, a multi-scale convolutional neural network is used to perform three-scale window division on the signal, calculate the fluctuation mean, amplitude offset and direction change rate at each scale, establish a multi-scale feature sequence, perform splicing and compression operations on multiple feature path outputs, and generate a fused scale feature map; S3: Based on the fused scale feature map, a graph convolutional network is used to locate the sampling point position in the network node structure diagram, extract the time feature sequence corresponding to each node, analyze the connection path length, signal delay, and interference coefficient value between nodes, construct a node connection relationship diagram and propagate node information values, update the node propagation strength and frequency response, and generate a topological propagation feature structure; S4: Based on the topological propagation feature structure, perform structural entropy calculation on all subgraphs in the connection path, identify paths with structural entropy greater than a threshold, prune redundant connections by comparing node influence values, select the path with the minimum entropy weight and perform path priority sorting to generate a path screening optimization graph sequence; S5: Based on the path screening optimization graph sequence, the propagation characteristics between the first node and the last node of each path are extracted, the node response strength judgment and type mapping are performed, the current fault mode type and the key propagation starting point position are determined, and the processing operation code is generated according to the fault mode number and propagation direction to establish the fault diagnosis confirmation result.

Images