Sensor deployment method and system for railways

Abstract

The present application relates to the field of sensor deployment analysis, and particularly relates to a railway sensor deployment method and system, which comprises the following steps: performing multi-position point inspection analysis on railway inspection records to generate a plurality of position point inspection key degrees; acquiring a railway facility design drawing; performing structure key degree evaluation based on the railway facility design drawing to generate a structure key degree score; performing full-coverage deployment analysis based on the plurality of position point inspection key degrees and the structure key degree score to construct a full-coverage deployment scheme; performing sensor deployment processing on the railway based on the full-coverage deployment scheme and collecting sensor detection signals; and performing signal detection quality quantitative analysis on the sensor detection signals to generate the detection quality of each sensor. The present application can quickly adjust the deployment based on the line state change, improve the coverage range and detection accuracy of the sensor, and enhance the safety sensing and monitoring efficiency of the railway.

Description

Technical Field

[0001] This invention relates to the field of sensor deployment analysis, and more particularly to a sensor deployment method and system for railways. Background Technology

[0002] During long-term railway operation, deployed sensors may experience performance degradation due to various factors such as train vibration, climate change, track stress changes, and external interference, including decreased accuracy, deviation, and data drift. Furthermore, aging railway infrastructure, structural changes, line expansion, and operational strategy adjustments can also make it difficult for existing sensor deployment schemes to maintain optimal monitoring results. Simultaneously, the complex operating environment may present unpredictable risks such as abnormal vibrations, track cracks, equipment overload, and foreign object intrusion, posing new challenges to sensor density, monitoring sensitivity, and data validity. Traditional railway sensor deployment methods rely heavily on empirical rules and static planning, primarily based on fixed line structures and known risk locations for initial deployment. However, with the increasing dynamism of railway operation scenarios, such fixed deployment schemes often fail to meet the needs of real-time monitoring and long-term evolution. Especially in the context of rapidly growing multi-source, multi-modal sensor data, existing schemes often lack the ability to continuously analyze sensor effectiveness, coverage changes, and risk hotspot migration, making it difficult to promptly identify monitoring blind spots and enabling rapid deployment optimization in response to changes in line conditions. In addition, manual inspections and periodic deployment adjustments still account for a large proportion of existing methods, resulting in low adjustment efficiency and an inability to achieve dynamic and enhanced monitoring of risk areas. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention proposes a sensor deployment method and system for railways, thereby resolving at least one of the aforementioned technical issues.

[0004] To achieve the above objectives, the present invention provides a method for deploying sensors on railways, comprising the following steps: Step S1: Perform multi-location point maintenance analysis on railway inspection records to generate the maintenance criticality of multiple location points; Step S2: Obtain railway facility design drawings; conduct structural criticality assessment based on the railway facility design drawings and generate a structural criticality score; Step S3: Conduct a full-coverage deployment analysis based on the criticality scores of maintenance at multiple locations and the criticality scores of the structure, thereby constructing a full-coverage deployment plan; Step S4: Based on the full coverage deployment plan, perform sensor deployment processing on the railway and collect sensor detection signals; perform quantitative analysis of the signal detection quality of the sensor detection signals to generate the detection quality of each sensor; Step S5: Based on the detection quality, identify and update the location of the performance bottleneck.

[0005] This specification provides a railway sensor deployment system for performing the railway sensor deployment method described above, comprising: The inspection information analysis unit is used to perform multi-location point maintenance analysis on railway inspection records and generate the criticality of maintenance at multiple location points. The structural criticality assessment unit is used to obtain railway facility design drawings; based on the railway facility design drawings, it performs structural criticality assessment and generates a structural criticality score; The full-coverage deployment unit is used to conduct full-coverage deployment analysis based on the criticality of maintenance and structural criticality scores of multiple location points, thereby constructing a full-coverage deployment plan; The detection quality analysis unit is used to process sensor deployments on the railway based on a full-coverage deployment plan and to collect sensor detection signals; it performs quantitative analysis of the signal detection quality of the sensor detection signals and generates the detection quality of each sensor. The position adjustment unit is used to identify and update the position of the performance bottleneck based on the detection quality.

[0006] The specific benefits of this invention are as follows: By analyzing the maintenance frequency and severity of problems at each location point through historical inspection records, vulnerable or critical locations in railway facilities can be identified. Clearly defining the maintenance criticality of each location point provides a key reference for subsequent sensor deployment, avoiding resource waste. Prioritizing locations with high criticality helps improve the effectiveness and accuracy of the monitoring system. Structural criticality scoring reflects the structural importance of each component or location in the railway facility design, helping to identify potential risk points. Analysis combined with design drawings ensures that sensor deployment not only relies on historical fault records but also considers the safety and criticality of the structure itself. Ensuring sufficient monitoring of critical structural locations improves the overall safety of railway facilities. By combining historical maintenance criticality and structural criticality scoring, a comprehensive yet focused deployment strategy is achieved. Blind or redundant deployments are avoided, ensuring that limited sensor resources cover the most critical locations, maximizing monitoring accuracy. A quantifiable and adjustable deployment scheme is constructed, providing a foundation for subsequent iterative optimization. Quantitative analysis of signal detection quality determines the actual monitoring capability of each sensor. Detection quality analysis can reveal locations of insufficient signal coverage or sensor performance bottlenecks, providing a basis for iterative optimization. Ensuring the accuracy and reliability of collected monitoring data provides effective information for subsequent risk analysis and decision-making. Adjusting sensor positions based on detection quality eliminates performance bottlenecks and optimizes monitoring coverage. Through dynamic adjustment and optimization, the monitoring accuracy and responsiveness of the railway sensor network are improved. Key locations are fully monitored, reducing the probability of potential faults and safety hazards. Attached Figure Description

[0007] Figure 1 This is a schematic flowchart illustrating the steps of a sensor deployment method for railways according to the present invention. Figure 2 This is a detailed flowchart illustrating the implementation steps of step S1. Figure 3 This is a flowchart illustrating the detailed implementation steps of step S2. Detailed Implementation

[0008] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0009] This application provides a method and system for deploying sensors on a railway. The entities executing the method and system include, but are not limited to, mechanical equipment, data processing platforms, cloud server nodes, and network upload devices that can be considered general computing nodes in this application. The data processing platform includes, but is not limited to, at least one audio / image management system, information management system, and cloud data management system.

[0010] Please see Figures 1 to 3 This invention provides a method for deploying sensors on railways, comprising the following steps: Step S1: Perform multi-location point maintenance analysis on railway inspection records to generate the maintenance criticality of multiple location points; Step S2: Obtain railway facility design drawings; conduct structural criticality assessment based on the railway facility design drawings and generate a structural criticality score; Step S3: Conduct a full-coverage deployment analysis based on the criticality scores of maintenance at multiple locations and the criticality scores of the structure, thereby constructing a full-coverage deployment plan; Step S4: Based on the full coverage deployment plan, perform sensor deployment processing on the railway and collect sensor detection signals; perform quantitative analysis of the signal detection quality of the sensor detection signals to generate the detection quality of each sensor; Step S5: Based on the detection quality, identify the performance bottleneck location and update and adjust its location.

[0011] In the embodiments of the present invention, see Figure 1 This is a flowchart illustrating the steps of a railway sensor deployment method according to the present invention. In this example, the steps of the railway sensor deployment method include: Step S1: Perform multi-location point maintenance analysis on railway inspection records to generate the maintenance criticality of multiple location points; In this embodiment, during the multi-location-point maintenance analysis of railway inspection records, the event categories, maintenance work orders, fault descriptions, location markers, and time information involved in the inspection records are comprehensively structured to form a track maintenance dataset that can be used for quantitative analysis. The processing flow is typically based on semantic parsing. By vectorizing key phrases in the inspection text records, such as extracting features from words like "sleeper settlement," "fastener loosening," "rail cracks," "ballast disturbance," and "foreign object intrusion," the fault types at different locations can be accurately identified. Subsequently, the results after semantic parsing are categorized according to location numbers such as track kilometer markers, section numbers, and bridge and culvert locations. The frequency of recorded faults, the duration of each fault, whether repeated maintenance is required, and the corresponding impact level are combined and calculated to construct the basic maintenance feature vector for each location point. For example, parameters such as maintenance frequency (3–12 times per year), recurrence rate (0–0.45), and average event impact weight (1.0–3.5) can be calculated for each location point. A weighted model is then used to form a location criticality index K_loc, such as K_loc = 0.4F + 0.35R + 0.25I, where F represents the normalized number of failures, R represents the recurrence rate, and I represents the impact level. During the analysis, environmental factors in the track construction area, such as train passing frequency, gradient changes, and the geological conditions of the track structure, are also considered to more comprehensively reflect the potential risk levels of different locations.

[0012] Step S2: Obtain railway facility design drawings; conduct structural criticality assessment based on the railway facility design drawings and generate a structural criticality score; In this embodiment, after completing the criticality analysis of the inspection data, a structural-level evaluation of the railway facility design drawings is required to establish a structural criticality model. This process uses railway design drawings, BIM models, or structural layout diagrams as input. It involves topological decomposition of structural locations such as rails, sleepers, fasteners, connecting plates, supporting components, bridge nodes, and tunnel linings to form a computable railway structural model. The processing flow extracts structural nodes from the design drawings, including all location numbers, connection relationships, node types, load-bearing directions, and stress patterns, and generates a structural mesh or node link matrix based on the geometric distribution. For example, nodes can be set at 5–10 m intervals for track sections, and key stress points (such as turnout areas, bridge joints, and outer rails on curved sections) can be marked as critical nodes. Subsequently, the importance of each node is quantified through structural stability analysis. For example, a load-bearing influence factor S_w can be used for load-bearing nodes, a connectivity importance factor C_conn can be used for connecting nodes, and a geometric sensitivity factor G_sens can be used for curved segments. A structural criticality score S_str is then generated through weighted integration, such as S_str = 0.5S_w + 0.3C_conn + 0.2G_sens. For instance, load-bearing nodes, due to their significant impact on track stability, typically have a score higher than 0.75, while nodes on ordinary straight sections generally fall within the 0.2–0.45 range. The entire process can incorporate factors such as dimensional parameters, radius of curvature, track span, and rail specifications from the structural drawings to make the criticality score more objective and comprehensive.

[0013] Step S3: Conduct a full-coverage deployment analysis based on the criticality scores of maintenance at multiple locations and the criticality scores of the structure, thereby constructing a full-coverage deployment plan; In this embodiment, after obtaining the maintenance criticality and structural criticality scores, the two types of data need to be jointly analyzed to determine the priority areas and deployment density of sensors. The processing flow integrates the two scoring indicators. For example, using K_loc to represent the maintenance criticality of a location and S_str to represent the structural criticality score, a comprehensive deployment priority index U = 0.6K_loc + 0.4S_str can be constructed, allowing high-risk and structurally sensitive areas to be considered simultaneously. In actual deployment planning, sensor coverage radius parameters need to be set. For example, the typical coverage radius of a depth image sensor can be set in the range of 25–35 m, and the monitoring coverage area of ​​each candidate location is calculated in combination with the viewing angle range (e.g., horizontal 80–120°, vertical 40–60°). Subsequently, analysis is performed using a coverage optimization algorithm. A point-by-point coverage evaluation method is adopted, dividing the railway structure model into several monitoring units (e.g., a segment every 5 m), and calculating the probability and number of segments covered by candidate sensor points. Based on this, a greedy algorithm can be used to optimize coverage, that is, to prioritize the selection of deployment points that cover multiple high-critical sections, so as to minimize the overall number of sensors while ensuring that the coverage rate meets the preset requirements (such as no less than 95% track section coverage).

[0014] Step S4: Based on the full coverage deployment plan, perform sensor deployment processing on the railway and collect sensor detection signals; perform quantitative analysis of the signal detection quality of the sensor detection signals to generate the detection quality of each sensor; In this embodiment, after the deployment plan is completed, sensor nodes need to be deployed along the railway line according to the plan, and depth image signals need to be collected for subsequent analysis. After the sensor is started, it will continuously output indicators such as depth data, intensity echo, point cloud density, and contour edge sharpness. In order to obtain a reliable evaluation of detection quality, the collected signals need to be continuously analyzed, including factors such as signal noise level, target structure imaging sharpness, and environmental interference. Typically, parameters such as signal-to-noise ratio (SNR in the range of 18–35 dB), point cloud integrity ratio (effective point cloud percentage, such as 85–98%), and depth error mean square value (deviation range of 0.8–2.5 cm) are calculated, and these parameters are used to construct the detection quality index Q_sns. For example, Q_sns = 0.45·SNR_n + 0.35·C_pc + 0.2·D_acc can be used, where SNR_n is the normalized signal-to-noise ratio, C_pc is the point cloud integrity, and D_acc is the depth accuracy. If a sensor operates in a dusty area, its SNR may fall below 15 dB, and the sensor's detection quality will automatically degrade. Furthermore, analyzing the stability of continuous signals, such as depth fluctuations exceeding the normal range by 10–15% or a significant decrease in point cloud contour sharpness, will also affect the quality score.

[0015] Step S5: Based on the detection quality, identify and update the location of the performance bottleneck.

[0016] In this embodiment, after obtaining the detection quality of each sensor, it is necessary to identify nodes with performance bottlenecks in the overall coverage and adjust their positions accordingly. The performance bottleneck identification process typically sorts nodes based on their detection quality index Q_sns. For example, nodes with a Q_sns below 0.45 are considered primary bottlenecks, and nodes with a Q_sns between 0.45 and 0.60 are considered secondary bottlenecks. This is combined with a comprehensive screening based on the importance weight of the covered section. If a node covers a highly critical section but has low detection quality, it will be prioritized for adjustment. During the position adjustment process, the spatial distance, visibility, and monitoring angle between the bottleneck node and the target railway section need to be recalculated. For example, if a bottleneck node is more than 40 m away from the target abnormal area or its monitoring angle deviates from the main structural direction by 20–40°, the node can be moved 5–15 m along the track direction, or its installation height can be adjusted by 0.3–0.8 m to improve its imaging quality. Furthermore, the coverage of surrounding sensors must be considered to avoid coverage gaps caused by adjustments. The adjusted deployment points must be verified through coverage calculation to confirm that their coverage rate has increased to a preset threshold, such as no less than 95%. If the coverage rate is still insufficient after node adjustment, other nodes need to be adjusted or a third deployment point needs to be added in an adjacent area to ensure the integrity of overall monitoring.

[0017] In this embodiment, see Figure 2 The diagram below illustrates the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: Deep semantic analysis of railway inspection records is performed to extract railway maintenance information; Identify all fault repair events based on railway maintenance information; The maintenance frequency and number of regional inspections are calculated based on the aforementioned fault maintenance events at different railway locations. The location failure rate is calculated based on the maintenance frequency and the number of regional maintenance visits to obtain the failure probability for multiple locations; The fault repair events are analyzed for the fault time distribution at each fault point, and the fault time sequence distribution data of each fault point is extracted. Identify the severity of the fault repair event; quantify the maintenance dependency of the fault severity, the fault time-series distribution data, and the fault probability to generate maintenance criticality at multiple location points.

[0018] In this embodiment, deep semantic analysis of railway inspection records typically uses multi-source inspection text as input, including descriptions of equipment inspections along the railway line, path location text, maintenance log fragments, task description information, etc. By constructing a structured semantic analysis process, natural language content is mapped line by line to quantifiable maintenance elements. During the analysis process, a domain vocabulary graph needs to be established, consisting of fault entity words (such as "loose bolts," "damaged sleepers," "wearing of contact wires"), operation action words (such as "replace," "tighten," "correct"), and component location words (such as "left track," "insulation joint," "numbered section"), so that the text can be initially semantically focused at the input end based on feature words and semantic proximity. To improve the accuracy of the analysis, a deep semantic model based on 768-dimensional vector encoding can be introduced, allowing inspection statements to be clustered in the vector space according to semantic similarity, and then an attention mechanism is used to identify key components of the statements, thereby improving the extraction accuracy of component attributes, fault attributes, and maintenance behaviors. During processing, it is also necessary to align the text positions of information such as time tags, inspection personnel identification, and location record annotations to establish a solid association between semantic fragments and the actual inspection scenario. Valid records are filtered using specific threshold rules (such as semantic similarity > 0.82 and action word confidence > 0.75), and statements associated with the same object, component, or section are merged into structured maintenance information. After semantic extraction of the inspection records, the maintenance information scattered across various time nodes needs to be reorganized into independent fault repair events using a unified event expression structure. This process can be achieved by constructing a three-dimensional event aggregation model based on time windows, spatial locations, and component categories. The time window is typically set between 24 and 120 hours to determine whether multiple statements belong to the same fault process; spatial locations are calibrated using track mileage values ​​or section codes, allowing event merging within a maximum spatial drift range of 3–5 meters; and component categories are determined based on component classification labels generated during the semantic parsing stage, with consistency judged through component similarity and track lateral position alignment. When assigning events to each maintenance record, it is typically necessary to calculate event similarity, which is a weighted average of component matching score, location matching score, fault content matching score, and maintenance behavior continuity score. For example, a total score can be calculated using a weighting ratio of 0.35 / 0.30 / 0.25 / 0.10, and whether a record should be classified as a single event is determined based on whether the total score is greater than 0.72. In this way, maintenance items with temporal continuity, location proximity, and content consistency can be gradually grouped into individual fault maintenance events, ensuring that each event includes the complete process of fault discovery, status change, handling actions, and subsequent impacts.

[0019] For identified fault maintenance events, their distribution trends need to be reorganized spatially to calculate the maintenance frequency at different railway locations. This process typically uses railway section numbers, line mileage coordinates, or sleeper numbers as spatial division methods, and performs statistical aggregation according to a fixed spatial granularity (e.g., every 50 meters, every 100 meters, or every section number). Each event is categorized into a corresponding section based on its core area location, and then all maintenance events occurring within the same section are counted to form the maintenance frequency value for that location. To enhance statistical stability, a sliding segment approach with neighborhood weighting can be used, combining the central section with 1-2 adjacent sections at ratios of 0.6, 0.25, and 0.15 to form a weighted frequency, ensuring that a single abnormally high-frequency point does not disrupt the overall trend. Simultaneously, in addition to the location dimension, events also need to be statistically analyzed by component category to confirm whether certain components exhibit a higher fault tendency in specific areas. Furthermore, the number of maintenance operations will be calculated separately for different levels of railway areas (such as sections, stations, and key bridge and tunnel areas) to clearly express the maintenance pressure at the regional level. After obtaining the maintenance frequency and regional maintenance operations at the location level, it is necessary to estimate the failure rate at each location on the railway using statistical modeling methods to obtain the location failure probability. Failure rate calculation typically employs a normalization method based on time intervals or section mileage. If time is used as a reference, the form λ = N / T can be used, where N is the number of maintenance events in that section during the statistical period, and T is the total duration of the statistical period (e.g., 180 days or 365 days). If mileage is used as a reference, the number of failures per 10,000 kilometers can be calculated. To improve the stability of the results, exponential smoothing or Bayesian correction strategies can be introduced, such as setting a prior failure rate (e.g., 0.05 times per 100,000 kilometers) for sections with less data and correcting it according to the observation frequency. After the failure rate is estimated, the failure rate can be further converted into the failure probability using a segment risk distribution model, for example, by using the form 1 – exp(–λ·τ), where τ represents the reference time window, such as the prediction period of the next 30 to 90 days.

[0020] Beyond failure rate calculation, it's also necessary to analyze the distribution characteristics of failure events from a time perspective. To this end, historical time records for each failure point can be constructed into a time series, and the interval characteristics between failure events can be studied. Three types of indicators are typically used in the analysis: first, the failure interval distribution, i.e., the time difference between two consecutive failures, using its mean, variance, and skewness to determine whether failures are concentrated or uniformly distributed; second, the trend of failure intensity changes, which can be calculated based on segmented time series events with short and long cycles, such as 30-day density and 90-day density; and third, periodic fluctuation characteristics, which can be identified using windowed autocorrelation analysis to determine if there are periodic recurring failure behaviors, such as accelerated wear of overhead contact line components due to seasonal temperature differences. When establishing the time distribution model, different window lengths such as 5, 10, and 20 days can be used to calculate the moving average and moving volatility to form a stable and detailed time series description. Identifying the severity of failures requires a comprehensive assessment of the impact of each failure event on line safety, train operation stability, and subsequent maintenance cycles across multiple dimensions. During identification, scoring is typically based on factors such as fault type (e.g., track structure damage, signal equipment malfunction, power supply contact network wear), fault impact range (e.g., whether it affects double or single tracks), speed limit level caused by the fault, and potential risk level to train operation. Multi-factor grading rules can be constructed during the scoring process. For example, the risk of the fault type can be weighted at 0.2–0.4, the urgency of maintenance actions at 0.3, the recurrence rate of historical faults at that point at 0.2, and the potential chain reaction caused by component failure at 0.1. Each event, after being processed by the scoring model, will receive a severity value, which can be quantified in ranges of 0–1 or 0–100, and further divided into minor, moderate, important, and critical levels. During the identification process, key terms appearing in the fault text also need to be weighted. For example, words like "fracture," "loosening," and "burnt" represent higher severity, while "mild" and "slightly loose" represent lower levels.

[0021] After obtaining the failure probability, temporal distribution characteristics, and severity in the aforementioned steps, these three factors need to be integrated into a maintenance criticality index that can be used for priority ranking. This process typically employs a weighted fusion model, which can be designed with the following structure: Maintenance Criterion = α × Failure Probability + β × Temporal Concentration + γ × Severity. The weights of α, β, and γ can be selected within the range of 0.2 to 0.5, for example, α = 0.35, β = 0.30, and γ = 0.35. The failure probability reflects the likelihood of future occurrence, the temporal concentration reflects the recent risk accumulation trend, and the severity directly indicates the level of impact on railway operation safety. During the quantification process, to avoid a single feature dominating the results, the three features can be normalized to a range of 0 to 1, and a smoothing coefficient κ (e.g., 0.1 to 0.2) can be introduced after calculation to stabilize cross-section comparisons. Each location point will receive a maintenance criticality value, which can be divided into four levels: 0-0.25 indicates general concern, 0.25-0.5 indicates need for follow-up, 0.5-0.75 indicates key concern, and 0.75-1.0 indicates urgent maintenance.

[0022] In this embodiment, see Figure 3 The diagram below illustrates the detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include: Obtain railway facility design drawings; perform topology modeling based on the railway facility design drawings to construct a railway structure model; The key structural points of the railway structural model are identified in depth and marked; the key structural points include main connections, transition points and load-bearing structural points. Railway network connectivity analysis is performed on key structural points to generate structural point connectivity features; Structural criticality is assessed based on the connectivity features of structural points, and a structural criticality score is generated. The signal propagation quality of the key structural points is calculated based on the railway facility design drawings.

[0023] In this embodiment, after obtaining the railway facility design drawings, it is necessary to perform structured analysis on the line layout, track arrangement, ancillary structural nodes, and equipment interfaces presented in the design drawings to construct a railway structure model that can be used for topology analysis. To achieve this goal, the drawing content can be transformed into a structural network composed of multi-category linear components and multi-scale nodes based on component symbols, line segment labels, node numbers, length annotations, and location baselines on the design drawings. During the analysis process, it is usually necessary to perform geometric identification on the track centerline, branch lines, connecting lines, bridge sections, tunnel sections, electrical equipment layout, and related ancillary structures. The azimuth angle, stride length, and curvature information of each line segment in the design drawings are extracted using a vectorization method, and scale conversion is performed based on pixel coordinates or the drawing coordinate system to express all components under a unified coordinate framework. Then, a graph structure representation method can be used, with each line segment as an edge and each connecting or intersecting component as a node, to construct a connected graph model of the railway structure. To improve the adaptability of the topology model, nodes can be classified, including ordinary nodes, bifurcation nodes, transition nodes, and critical load-bearing nodes, and their attributes can be recorded in the model. After establishing the railway structure model, all nodes in the model need to be deeply identified to determine which nodes are critical structural points that have a significant impact on the overall structural stability, railway operation continuity, and load transmission path. During the identification process, the importance of nodes can be determined based on the node type, number of connecting edges, component cross-sectional characteristics, geometric location, and the force flow path associated with the node recorded in the model. For example, for intersection nodes that connect multiple tracks or bear force flow in multiple directions, their connectivity is often greater than 3, and their geometric location is in the structural transition area; such nodes can be directly identified as major connection points. For locations where the track direction changes significantly, where the track transitions from straight to curved, or where structural transitions occur in equipment interface sections, they can be identified as transition points. For nodes located in bridges, tunnels, support foundations, turnout hubs, etc., and participating in load transmission, they can be identified as load-bearing structural points based on the load-bearing path. The identification process can be judged by combining multiple indicators such as the connection strength of the node, the component length ratio, the change of the orbital angle, and the betweenness centrality of the node in the overall structural network. For example, nodes with betweenness centrality higher than 0.18, connection strength score higher than 0.72, or structural transformation rate higher than 0.25 are marked as key structural points.

[0024] After identifying key structural points, it is necessary to further analyze their position within the overall railway structure from the perspective of network connectivity to generate structural point connectivity characteristics. Connectivity analysis typically revolves around three dimensions: first, path connectivity, i.e., the number of routes accessible from a key point, the structural levels that can be traversed, and the number of shortest paths; second, redundancy characteristics, which assess the number of alternative paths around the key point, including whether there are more than one connecting path if the node fails or becomes isolated, and the differences in length between these alternative paths; and third, structural distribution characteristics, which calculate the hierarchical position of the key point in the network, such as whether it is a hub-level node, a regional connection point, or a terminal node. The analysis can employ a graph traversal-based depth-first search model to calculate the breadth distance from the key point to other nodes, the reachable area coverage, and the path distribution density. For example, indicators such as whether the number of nodes in the three-layer path structure extending from the key point exceeds 15 and whether the path density is higher than 0.12 can be used to determine the strength of connectivity. Higher connectivity typically means that a key point plays a broader connecting role in the railway network, and its failure will affect a wider area; while key points with lower connectivity are often located in peripheral areas and have a smaller impact on overall connectivity. After the connectivity characteristics of structural points are clarified, structural criticality assessment is needed based on their contribution to the overall stability and operational continuity of the railway structure, thereby generating structural criticality scores that reflect the importance levels of different nodes. The assessment process usually employs a multi-factor weighted model, where connectivity coverage, the number of alternative paths, path density, node betweenness centrality, and node type weights are all key parameters. For example, connectivity coverage directly reflects the impact of a key point on the line's extent and can be assigned a weight of 0.35; the number of alternative paths reflects the substitutability of a key point when it fails and can be assigned a weight of 0.25; node betweenness centrality indicates its importance in path transitions and can be assigned a weight of 0.20; node type (whether it belongs to a structural support point or is located in a structural transition section) can be assigned a weight of 0.20. By normalizing each indicator to the 0-1 range and then weighting and summing them, a total structural criticality score can be obtained, typically distributed within a quantification range of 0-1 or 0-100. Critical points with scores above 0.75 usually represent core nodes in the overall structure; failure of these nodes can lead to connection breaks or range-wide operational obstruction. Nodes with scores between 0.5 and 0.75 are important nodes that require frequent monitoring. Nodes with scores below 0.5 are considered general nodes with limited impact on the overall structure.

[0025] After the structural criticality assessment is completed, the communication reachability of key structural points also needs to be calculated from the perspective of signal propagation to ensure stable signal quality for key nodes in sensing monitoring, communication scheduling, and remote information transmission. Signal propagation quality is typically assessed based on factors such as material information, structural coverage, spatial obstacle distribution, and trackside equipment layout contained in the railway design drawings. For example, for key structural points located in tunnel sections, signal attenuation needs to be calculated based on parameters such as tunnel cross-sectional dimensions, curve radius, and wall absorption characteristics; for nodes located in bridge sections, the impact of signal reflection or scattering needs to be calculated by considering factors such as bridge beam reflectivity and steel structure interference. During the calculation process, geometric modeling of the propagation path can be performed. By assuming the signal propagation direction on 2 to 4 possible paths, the propagation loss of each path can be estimated. For example, attenuation calculations can be performed for different structural sections within a loss range of 0.8 to 1.5 dB / m. Simultaneously, the shielding coefficient of components needs to be considered; for example, the shielding coefficient for concrete structures can be set to 0.65 to 0.85, and for metal structures, it can be set to 0.4 to 0.6.

[0026] In this embodiment, the specific steps for calculating the signal propagation quality of the key structural points based on the railway facility design drawings are as follows: The physical structural parameters of the key structural points are extracted based on the railway facility design drawings; Based on the physical structure parameters, multi-band signal propagation simulation of structural points is performed to generate multi-band signal propagation simulation data. Calculate the signal transmission delay, distortion, and propagation attenuation characteristics of multi-band signal propagation simulation data; The signal propagation quality is calculated based on the signal transmission delay, distortion, and propagation attenuation characteristics to generate the signal propagation quality of key structural points.

[0027] In this embodiment, when processing railway facility design drawings, parameters are extracted item by item from key structural points based on the structural element annotations, component size markings, and engineering drawing standards in the design drawings. Key structural points typically include support nodes, sleeper connection ends, rail joints, load-bearing beam nodes, and transition points of turnaround sections; therefore, it is necessary to identify their corresponding dimensioned areas from the design drawings. Parameter extraction is based on two-dimensional design drawing data. By geometrically analyzing the line width annotations, component size ratios, material symbols, and node number information in the design drawings, the length, thickness, beam width, track gauge, elevation difference, material type, and relative positional relationships of components at the nodes are obtained. To ensure data integrity, the geometric intersection relationships of components around the nodes are also calculated during the extraction process, such as the included angles between components at the nodes, the extension length of welded positions, the spacing of connectors, and the number of fixing points, so that key structural points form a complete set of physical parameters including elastic modulus, structural dimensions, resistance parameters, material density, and geometric morphological characteristics. During the extraction process, a 1:200 or 1:500 scale conversion factor is typically used to normalize the numerical values ​​of the diagram annotations, ensuring that the parameters can be directly used as input variables in subsequent signal simulations. Based on the obtained physical parameters of key structural points, a propagation environment model for multi-band signal transmission characteristics needs to be constructed, and signal propagation simulations are conducted accordingly. Signal frequency bands can be set to multiple commonly used transmission bands such as 400 MHz, 1.2 GHz, 2.4 GHz, 3.5 GHz, and 5.8 GHz to cover different functional scenarios such as trackside monitoring, train control equipment communication, and track condition detection. In propagation simulations, the physical parameters of key structural points (such as the dielectric constant and conductivity of node materials, component thickness, geometric angles, and metal density variations in connection areas) directly affect the signal's reflection, refraction, diffraction, and absorption behavior; therefore, these parameters are considered core influencing factors in the propagation model. The propagation simulation typically employs the wave equation characteristic expression commonly used in electromagnetic wave propagation models, and a grid is divided according to the node's geometric characteristics, causing the signal to experience different amplitudes of attenuation or scattering in different grid regions. Meanwhile, the material type at the nodes (such as steel rails, concrete bases, metal connecting pieces, etc.) will cause frequency band-related propagation loss. Therefore, the simulation will calculate the signal reflection coefficient, absorption rate and transmittance under multiple frequency bands respectively.

[0028] After obtaining multi-band signal propagation simulation data, it is necessary to perform parameterized calculations on the propagation performance of each band to quantify the impact of key structural points on signal transmission. Based on the electromagnetic wave propagation time model, signal transmission delay is calculated. The delay is affected by factors such as the material density, dielectric constant, component thickness, and connection angle of the structural points. Therefore, it is necessary to accumulate and statistically analyze the end-to-end propagation time of each path to obtain the delay curve corresponding to each frequency band. Secondly, signal distortion is calculated. Distortion typically originates from multi-path superposition, local reflection angle deviation, non-uniform scattering, and waveform distortion caused by geometrical abrupt changes in node regions. Therefore, it is necessary to analyze the similarity of the signal waveform before and after propagation to obtain a quantifiable index of distortion. Multiple frequency bands can exhibit different degrees of distortion. Furthermore, propagation attenuation characteristics need to be calculated. The attenuation intensity is related to the energy loss of the signal when passing through node materials. The amplitude attenuation trend of the signal along the propagation path can be fitted to obtain the attenuation factor, energy loss rate, and additional path loss value caused by structural points. Through the quantitative calculation of delay, distortion, and attenuation, a set of signal propagation characteristics under multiple frequency bands can be formed. After obtaining the delay, distortion, and attenuation characteristics of key structural points, these parameters need to be comprehensively analyzed to form a signal propagation quality evaluation result. The calculation of signal propagation quality can be based on a multi-indicator weighted model, where delay reflects overall transmission efficiency, distortion reflects data integrity, and propagation attenuation characteristics reflect effective signal coverage. During the calculation, indicators for different frequency bands can be normalized to ensure they participate in the propagation quality calculation on a uniform scale. In the delay component, the difference between the average and maximum delay across different frequency bands can be used to measure the impact of the structural point on the overall communication link. In the distortion component, waveform distortion rate can be used for scoring; lower distortion indicates a smaller impact of the node on signal morphology. In the attenuation component, attenuation can be quantified using the attenuation factor in the energy loss model; lighter attenuation indicates higher propagation quality.

[0029] In this embodiment, step S3 includes the following steps: Multi-objective solution is performed based on the criticality of maintenance at multiple locations, the criticality score of the structure, and the signal propagation quality to extract the first sensor deployment point; Define the sensor's detection range; The physical detection boundary is calculated and extracted based on the sensor's detection range. Greedy calculations are performed on the railway structure model based on the physical detection boundaries to generate a greedy deployment scheme; Based on the greedy deployment scheme, a full-coverage deployment analysis is performed on the initial sensor deployment points to construct a full-coverage deployment scheme.

[0030] In this embodiment, when faced with three categories of indicators—maintenance criticality, structural criticality score, and signal propagation quality—at multiple locations, a multi-objective solution process needs to be constructed to select candidate sensor deployment points that maximize monitoring value from these indicators. The three types of indicators are quantified and normalized to ensure comparability within the same numerical range. For example, maintenance criticality is weighted according to fault association density, historical maintenance frequency, and temporal concentration to obtain a criticality index of 0–1; structural criticality is quantified based on node connectivity, load-bearing level, and topological dependency; and signal propagation quality is comprehensively scored based on delay, attenuation, and distortion. Subsequently, a multi-objective optimization approach is used to solve the three indicators, with the objective framework of "prioritizing high maintenance criticality, high structural criticality, and high signal propagation quality." Candidate locations are gradually selected through a weighted summation model, Pareto optimal selection, and hierarchical indicator ranking. During the solution process, the weight structure can be configured differently. For example, the maintenance criticality weight can be set to 0.45, the structural criticality weight to 0.35, and the signal propagation quality weight to 0.20, to ensure that the optimal location point can balance reliable monitoring, structural importance, and signal stability. After obtaining the first set of sensor deployment points, the effective detection range of the sensors needs to be strictly defined based on the sensor device attributes, the spatial geometry of the railway facilities, and the distribution characteristics of key structural points. The sensor detection range typically includes four types of parameters: angle detection range, distance coverage range, effective imaging area, and visible obstruction factors. The angle range can be set as a variable sector coverage area of ​​60°–120°, and the distance coverage range can be set to different levels such as 25 meters, 40 meters, or 70 meters based on the sensor's optical characteristics or band characteristics. When defining the detection range, it is also necessary to consider the actual spatial configuration in the railway structural model, such as track curvature, lateral obstacles, slope changes, and node connectivity characteristics, to avoid overlap between the detection sector and structural obstruction locations. In addition, the redundancy requirements between adjacent sensors must be considered when defining the detection range. To avoid blind spots, the detection range can be set to have 10%–20% overlap for subsequent coverage optimization.

[0031] After determining the sensor detection range, it is necessary to calculate the physical boundary of its actual coverage area in the railway structure model to clarify the spatial range that each sensor can cover. The determination of the physical detection boundary requires a geometric projection method, projecting the detection sector, detection distance, and angle constraints onto the spatial topology of the railway structure model. This ensures that the coverage area forms a clear spatial boundary along the track centerline, sleeper distribution, node coordinates, and structural contour. The calculation process must consider the intersection relationship between the detection range and the railway structure. For example, when a sensor is installed on the side of a contact wire pole, its detection sector may undergo geometric deformation when passing through curves. Therefore, a spatial polygonal projection of the detection range is required to form a detection shape that closely matches the actual railway structure. Furthermore, if there are elevation changes on the track, the detection boundary also needs to be vertically compensated for based on the slope, ensuring that the sensor detection area includes the true spatial contour affected by the elevation difference. After determining the physical detection boundary of each candidate sensor point, a greedy strategy is needed to perform preliminary calculations of the deployment scheme in the railway structure model. The greedy strategy aims to achieve maximum structural coverage with the fewest deployments. Therefore, during the calculation, the physical boundary of each sensor is considered a coverable area, and the coverage area or the number of covered key points is used as the basis for greedy selection. The greedy selection process involves iterative steps: selecting the deployment point that covers the most key structural points from all sensors and adding it to the deployment plan; then removing the covered area from the total coverage requirement, and selecting the point that covers the most remaining uncovered area from the remaining deployment points, repeating this process until the required area is basically covered. The process also considers node occlusion, coverage overlap, and structural hierarchy importance; for example, sensor boundaries that can cover nodes with high structural criticality can be prioritized. The advantage of the greedy strategy is that it can quickly obtain an approximate optimal solution, allowing key locations of the railway structure to be prioritized for monitoring, thus forming the initial deployment plan. After obtaining the greedy deployment plan, a comprehensive analysis of the overall coverage is needed to ensure that all key locations, key nodes, and related areas in the railway structure model can be effectively covered by sensors. Full coverage analysis requires merging the physical detection boundaries of each sensor in the greedy plan to identify uncovered spatial areas. If uncovered areas exist, it is necessary to find suitable deployment points from the initial sensor deployment point set to fill the gaps, prioritizing those with insufficient coverage. The analysis must consider factors such as coverage redundancy requirements, deployment spacing limitations, and sensor installation feasibility to ensure the solution achieves both complete coverage and a reasonable number of deployments. For areas with insufficient coverage, the coverage effect can be optimized by increasing the detection distance, adjusting the detection angle, or selecting deployment points closer to the area. After supplementing all uncovered areas, the entire set of sensor deployment points included in the solution is integrated into a comprehensive coverage deployment plan, enabling full spatial monitoring of the railway structure model.

[0032] In this embodiment, the specific steps of step S4 are as follows: Set a maintenance cycle window; deploy and process sensors on the railway based on a full-coverage deployment plan; collect signals according to the maintenance cycle window; and extract sensor detection signals. The sensor detects signals and performs environmental background noise identification to extract the environmental background noise. Calculate the signal-to-noise ratio and total harmonic distortion of the ambient background noise to generate the noise interference intensity; The accuracy of each sensor signal is calculated to generate the signal accuracy. The noise interference intensity and signal accuracy are quantitatively analyzed to determine the signal detection quality of each sensor.

[0033] In this embodiment, after constructing the full-coverage deployment scheme, it is necessary to set the maintenance cycle window based on railway maintenance patterns, structural fatigue characteristics, and track wear trends to ensure that signal acquisition has temporal continuity and complete characterization of the structural state. The maintenance cycle window can typically be set based on a 12-hour, 24-hour, or 72-hour timescale, while also considering periodic factors such as diurnal temperature differences, traffic density changes, and railway vibration loads, so that the cycle window can cover the structural response under typical working conditions. After setting the cycle window, each sensor in the full-coverage deployment scheme is installed at key structural locations according to preset deployment coordinates, enabling it to continuously acquire monitoring signals such as track deformation data, vibration response, structural micro-deformation, and optical offset within the cycle window. Signal acquisition is performed at a fixed sampling frequency within the cycle window, such as setting a sampling rate of 1kHz, 5kHz, or 10kHz, so that the acquired signals can cover the high-frequency impact characteristics and low-frequency deformation trends of the track. During the acquisition process, the sensor converts the physical changes within its detection range into vibration amplitude curves, light intensity change sequences, or displacement response sequences, and stores them sequentially using timestamps, thereby forming a continuous set of sensor detection signals. After obtaining the sensor detection signal, it is necessary to identify the environmental background noise contained in the signal in order to distinguish between the true structural response and invalid interference components. Environmental background noise mainly originates from trackside wind disturbance, ground vibration, slight metal friction, distant vehicle scattering noise, and electromagnetic radiation interference. Therefore, identification requires analysis of the signal's frequency domain characteristics, amplitude variation patterns, and steady-state noise distribution. Background noise identification can involve dual analysis of the signal in both the time and frequency domains. For example, in the time domain, it can identify persistent random noise, abrupt pulse noise, and periodic interference noise; in the frequency domain, it can identify high-frequency scattering noise, low-frequency foundation vibration, and irregular spectral components outside the structure's natural frequencies. During the identification process, the signal energy distribution characteristics should also be utilized to identify random components with amplitudes in the low-energy range of 5%–10% as candidate background noise, and further filter out frequency bands unrelated to the structural response.

[0034] After extracting background noise, to quantify its impact on the sensor's detection signal, signal-to-noise ratio (SNR) calculation and total harmonic distortion (THD) analysis are required. SNR is calculated based on the ratio of noise power to effective signal power; a lower SNR indicates stronger noise interference. Typically, the SNR calculation range can be set to multiple intervals such as 20 dB, 30 dB, and 40 dB to assess differences in noise stability at different locations. THD identifies the harmonic components in the noise. By calculating the proportion of energy occupied by harmonic components other than the fundamental frequency in the noise sequence, the degree of harmonic interference is determined. A high proportion of harmonic energy indicates that the noise will interfere with the periodic vibration characteristics of the structure, thus affecting the accuracy of the structural frequency analysis. After calculating SNR and THD, these two indicators are fused to generate the noise interference intensity; for example, a lower SNR and higher distortion correspond to a stronger noise interference intensity. Noise interference intensity is presented in numerical evaluation form. After obtaining noise characteristics, signal accuracy calculation is required for the signals collected by each sensor to determine its true measurement capability under the aforementioned noise interference. Signal accuracy can be comprehensively evaluated based on signal stability, structural response consistency, amplitude reconstruction error, and period fitting degree. For example, if the signal maintains a stable amplitude change trend across multiple period windows, the accuracy is high; if irregular jumps, waveform distortion, or non-structural frequency components appear in the signal, the accuracy decreases. When calculating signal accuracy, the amplitude error range can be set to 1%–5%. If the error exceeds the threshold, it indicates that the signal is heavily affected by noise interference. The period fitting degree can also be set to a matching degree range of 0.85, 0.90, or 0.95 to determine the signal's ability to reflect the true structural behavior. By combining indicators such as amplitude error, frequency deviation, waveform offset, and structural response consistency, a signal accuracy score is formed for each sensor. After obtaining the two key indicators of noise interference intensity and signal accuracy, a comprehensive quantitative analysis of signal detection quality is required to determine the effectiveness and reliability of each sensor in the actual monitoring process. Detection quality analysis can employ a weighted scoring method, where signal accuracy is typically the primary weight because it represents the sensor's ability to accurately characterize structural changes. Noise interference intensity serves as the inverse weight; a higher value indicates greater susceptibility to external interference. When constructing the detection quality model, the signal accuracy weight can be set to 0.6–0.75, and the noise interference intensity weight to 0.25–0.40, thus forming the calculated expression of detection quality. Quantitative analysis must also consider detection stability, temporal continuity, and waveform integrity to ensure that the detection quality accurately reflects the sensor's overall performance in the actual monitoring environment.

[0035] In this embodiment, the specific steps of step S5 are as follows: Based on sensor detection signals, abnormal location is located and sensor spatial distance is calculated to obtain the spatial distance between multiple abnormal locations and sensors. Based on the detection quality, the location of the performance bottleneck is identified, and the performance bottleneck sensor node is marked. Based on the spatial distance, the position adjustment analysis of the performance bottleneck sensor node is performed to generate the second sensor deployment point; The railway coverage rate is calculated based on the second sensor deployment point. If the railway coverage rate exceeds the preset railway coverage rate threshold, the other sensor locations are retained. When the railway coverage rate does not exceed the preset railway coverage rate threshold, the location update calculation of the remaining sensors is performed for the full coverage deployment scheme, and the deployment point of the third sensor is output.

[0036] In this embodiment, after parsing the raw monitoring signals acquired by the depth image sensing network, it is necessary to first perform unified coordinate calibration processing on the depth information, point cloud information, and low-frequency structural vibration information generated by the multi-source sensor nodes. Typically, the three-dimensional coordinates are referenced to the mining area's benchmark grid, for example, using X–Y–Z three-dimensional unified measurement coordinates (units can be set to cm or m). The fixed poses of multiple sensor nodes are matrix-described, and spatial registration is performed using a preset calibration point matrix to reduce positioning errors caused by coordinate drift. Subsequently, temporal difference calculations are performed on the depth images continuously acquired by each sensor node, revealing voxel-level structural changes, moving contours, unexpected deformations, and personnel and mechanical intrusion behaviors as significant depth offset points in three-dimensional space. To enhance the robustness of anomalies, the original depth difference results are typically subjected to shape dilation, noise removal, and gradient magnitude thresholding. For example, a depth variation threshold Δd ≥ 6–10 cm is set (based on the mine wall stability level and monitoring radius). This is combined with voxel clustering algorithms (such as Euclidean distance-based voxel clustering) to merge adjacent anomalies into actual anomaly region centers. After obtaining the anomaly center, a spatial distance matrix between the anomaly and each sensor node is constructed using the three-dimensional Euclidean distance formula, where d = √((x...). a x s )² + (y a y s )² + (z a z sThe distance from each anomaly location to each sensor node is calculated and recorded using a method similar to ()²), and the distance fluctuations are statistically analyzed over multiple time periods to obtain stable and accurate spatial distance data between anomaly locations and sensors. After obtaining the spatial distance between anomaly locations and sensors, a detection quality index is constructed to characterize sensor performance based on the imaging quality, depth recovery accuracy, data missing rate, and trigger latency of sensor nodes in different monitoring time periods. This index is typically composed of multiple weighted parameters, such as depth image noise rate (which can be determined by depth variance σ²), effective pixel ratio (used to determine environmental dust obstruction), timing synchronization error (which can be measured by multi-point triggering to check timing deviation), and target deformation recovery error (evaluated based on multi-view fusion reconstruction error E_m). To quantify node performance, several thresholds can be set, such as a noise rate exceeding 12%, an effective pixel ratio below 70%, a timing deviation exceeding 8 ms, or a reconstruction error greater than 4 cm, which are considered performance degradation. Combining the above indicators, a sliding window approach is used to statistically evaluate all nodes within the monitoring period, and a node performance score curve is constructed. Performance bottleneck nodes are identified based on their scores falling below a stable threshold (e.g., 60 points or an adaptive quantile threshold). Furthermore, performance is correlated with the average monitoring distance from the node to the anomaly location. For example, nodes with a distance exceeding 22–30 m may experience reliability degradation due to depth diffusion. A distance-quality two-dimensional distribution map can more clearly identify the distribution of bottleneck nodes.

[0037] After identifying performance bottleneck nodes, their positions need to be recalculated based on their spatial distance from the abnormal area, coverage blind spots, depth imaging angles, and structural occlusion. Typically, a combination of 3D spatial visibility analysis, viewpoint optimization models, and multi-point reconstruction redundancy calculation methods is used to determine the optimal candidate locations. A line-of-sight obstruction map needs to be constructed using a current mine depth obstacle model (obtained through point cloud fusion) to subdivide the visible area of ​​the performance bottleneck node and determine the specific angular range of the monitoring blind spot it causes. Subsequently, by setting target coverage constraints, such as ensuring that each potentially abnormal area is monitored by at least two nodes simultaneously, or guaranteeing a maximum monitoring distance of less than 20–25 m, and considering the terrain boundaries of the installable area in the mine, feasible locations for the nodes are determined. Typically, continuous location optimization based on gradient descent or heuristic optimization schemes based on discrete grid search are used to iteratively update the 3D coordinates of the nodes. Tilt and orientation constraints are also incorporated during the optimization process, such as maintaining the sensor's depression angle within the range of 15°–35° and the horizontal viewpoint deflection not exceeding 40°, to obtain higher depth reconstruction resolution. After obtaining the deployment point of the second sensor, it is necessary to assess the spatial coverage of the entire monitoring area, especially the track routes used for transportation and construction scheduling within the mining area. To this end, the mine track coordinate lines are first discretized into a set of track nodes at a fixed resolution (e.g., 1 m intervals), and the field-of-view coverage of each track node at the second deployment point is calculated. Coverage determination is typically based on the effective monitoring radius R of the depth sensor (e.g., 25–40 m, determined by equipment performance), the field of view angle (typically 60°–90°), and the depth model of obstructions. Spatial ray projection is used to determine whether a track node falls within the visible range. Then, the proportion of all track nodes covered by at least one deployment point is calculated to obtain the overall railway coverage. If the coverage exceeds a preset threshold (e.g., 85%, 90%, or a dynamic threshold set according to the mine's safety level), it indicates that the current second deployment point can meet the track monitoring requirements. In this case, no further adjustments to other existing nodes are necessary to maintain deployment stability and reduce construction workload.

[0038] If the railway coverage rate fails to reach the preset threshold, further global deployment optimization of all sensor nodes is required. This involves not only adjusting performance bottleneck nodes but also recalculating the positions of the remaining nodes to improve overall coverage. This process typically employs a global search strategy, visibility optimization, redundancy fusion enhancement algorithms, and a track priority-based coverage enhancement model. A coverage gap map is constructed based on the areas lacking track node coverage. Uncovered track sections are weighted according to length, geological features, and hazard level; for example, areas with steep slopes or frequent mechanical activity are given higher weights. Subsequently, a candidate point set is generated within the deployable area (usually defined based on stable ground areas, support areas, and equipment installation areas). Combining the effective monitoring radius R, field of view, and occlusion model, the improvement degree of each candidate point on the coverage gap is calculated. To avoid local optima, simulated annealing, genetic optimization, and other methods can be used for a global search to evaluate the coverage rate, distance distribution, redundancy (e.g., important areas observed by at least two nodes), and installation feasibility (e.g., avoiding narrow passages or dangerous areas) of each node location combination. After several iterations (e.g., 40–150 iterations), the optimal coverage layout is gradually approximated, and a new set of three-dimensional coordinates and orientation data is generated, namely the third sensor deployment point. This deployment point can significantly improve the overall coverage, enabling track monitoring to meet safety threshold requirements, while improving depth imaging redundancy, making mine construction safety early warning more stable and timely.

[0039] In this embodiment, the specific steps for locating abnormal locations and calculating sensor spatial distances based on sensor detection signals to obtain the spatial distances between multiple abnormal locations and sensors are as follows: Identify and mark abnormal railway parameters based on sensor-detected signals; Based on the anomaly detection parameters, the anomaly location is determined, and the coordinates of the anomaly location are extracted. The spatial distance between the sensor and the abnormal location is calculated based on the coordinates of the abnormal location, and the spatial distance between the sensor and the abnormal location is obtained.

[0040] In this embodiment, in a mine construction safety monitoring scenario, multiple depth image sensing nodes continuously capture the structural contours of the track area, sleeper deformation, rail torsion, soil settlement around the track, and foreign object intrusion. To identify anomalous parameters from these detection signals, the depth image sequence acquired by the sensors needs to be continuously analyzed in both spatial and temporal dimensions. The processing flow typically begins with depth image denoising and brightness equalization, for example, using bilateral filtering and median depth smoothing to control random noise generated in high-dust environments within 3–5% of the original depth fluctuation. Subsequently, temporal difference and voxel-level three-dimensional change detection are performed on the sequence images to reveal minute changes in the track's geometric contours. To enhance the robustness of anomaly identification, a track baseline model is usually introduced, such as using the depth profile curve of a normal track as a reference template. Threshold judgment is performed using a segmented curvature comparison curve K(x). If the track curvature deviation exceeds 8–12 mm, the sleeper settlement depth offset exceeds 15–20 mm, or the rail surface height gradient anomaly exceeds 4–6%, it is marked as a primary anomaly. Furthermore, an anomaly index function E_abn needs to be constructed by combining multiple parameters such as depth gradient direction, track boundary contour integrity, and changes in rail surface reflection intensity. A threshold (e.g., E_abn > 0.6) is then set to filter significant anomalies. When detecting railway operating areas, changes in foreign object volume are also monitored. For example, the presence of continuous three-dimensional entities (volume > 0.001–0.003 m³) in voxel clustering results can be identified as foreign object intrusion. After anomaly parameter identification, this anomaly information needs to be further converted into accurate three-dimensional location coordinates for subsequent distance analysis and regional risk localization. This process relies on the extrinsic calibration parameters of the depth image sensor, including node position (which can be set as a fixed point in the three-dimensional coordinate system), orientation matrix, tilt angle, and installation height. Based on these parameters, the pixels in each depth image can be reconstructed into a three-dimensional point cloud using an inverse projection model. Then, the pixel regions corresponding to the anomaly parameters are mapped to three-dimensional space, thus obtaining a preliminary voxel set of the anomaly region. To improve positioning accuracy, point cloud clustering and geometric center calculation are performed on the voxel set. For example, Euclidean distance clustering algorithm is used, with a cluster radius r_c = 15–25 cm, to remove scattered small noise clusters and form stable geometric center points in abnormal areas. Simultaneously, to reduce spatial drift caused by changes in viewing angle, the 3D reconstruction results from multiple sensors are fused. Weighted averaging or ICP (Iterative Closest Point) alignment methods are used to unify abnormal areas reconstructed from multiple views into the mine's reference coordinate framework. Typically, the fusion error can be controlled within the range of 1.5–2.5 cm. For track anomalies, such as sleeper subsidence or rail twisting, the anomaly points can be projected and corrected using the track centerline model to ensure they accurately fall within the track coordinate system.The 3D coordinates of each anomalous region are output in the form of (x, y, z), forming an anomalous location coordinate set for the next stage of spatial distance analysis. After obtaining the anomalous location coordinates, it is necessary to further calculate the spatial distance between these locations and each sensor node for subsequent coverage analysis, performance evaluation, and node deployment optimization. This calculation is based on 3D geometric distance, specifically using the Euclidean distance formula d = √((x...z)). a x s )² + (y a y s )² + (z a z s )²), where (x a , y a , z a (x) represents the coordinates of the outlier point. s , y s , z s The coordinates of the sensor node are shown below. To improve the physical accuracy of distance calculations, the actual monitoring field of view and monitoring radius of the sensor need to be considered. For example, if the effective monitoring radius of a sensor node is 25–35 m, and the calculated result d exceeds this range, the node and the anomaly location need to be marked as invalid. Furthermore, the distance needs to be corrected for visibility by combining the sensor's depression angle θ and horizontal deflection angle φ. This is done by determining whether the anomaly point falls within the sensor's field of view cone; only when |φ|≤45° and the depression angle is between 15°–35° is the distance included in the effective distance calculation. In cases of obstruction, a line-of-sight projection analysis can be performed using a 3D environmental model of the mining area. Ray detection can be used to confirm the presence of rock walls, equipment, or support structures along the path. Even if the distance meets the requirements, obstructed areas should be marked as invisible. All effective distances are integrated into a sensor-anomaly spatial distance matrix, which can be used for subsequent calculations such as coverage distribution, sensor performance degradation analysis, and deployment point updates. This method, based on a combined geometric and visibility assessment, ensures that the response range of each sensor to mining track anomalies is accurately quantified.

[0041] In this embodiment, a railway sensor deployment system is provided for performing the railway sensor deployment method described above, including: The inspection information analysis unit is used to perform multi-location point maintenance analysis on railway inspection records and generate the criticality of maintenance at multiple location points. The structural criticality assessment unit is used to obtain railway facility design drawings; based on the railway facility design drawings, it performs structural criticality assessment and generates a structural criticality score; The full-coverage deployment unit is used to conduct full-coverage deployment analysis based on the criticality of maintenance and structural criticality scores of multiple location points, thereby constructing a full-coverage deployment plan; The detection quality analysis unit is used to process sensor deployments on the railway based on a full-coverage deployment plan and to collect sensor detection signals; it performs quantitative analysis of the signal detection quality of the sensor detection signals and generates the detection quality of each sensor. The position adjustment unit is used to identify and update the position of the performance bottleneck based on the detection quality.

[0042] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0043] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein are implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present 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 of the invention herein.

Claims

1. A method for deploying sensors on a railway, characterized in that, Includes the following steps: Step S1: Perform multi-location point maintenance analysis on railway inspection records to generate the maintenance criticality of multiple location points; Step S2: Obtain railway facility design drawings; Based on the railway facility design drawings, a structural criticality assessment is conducted, and a structural criticality score is generated. Step S3: Conduct a full-coverage deployment analysis based on the criticality scores of maintenance at multiple locations and the criticality scores of the structure, thereby constructing a full-coverage deployment plan; Step S4: Based on the full coverage deployment plan, perform sensor deployment processing on the railway and collect sensor detection signals; Perform quantitative analysis of the signal detection quality of the sensor detection signals to generate the detection quality of each sensor; Step S5: Based on the detection quality, identify and update the location of the performance bottleneck.

2. The sensor deployment method for railways according to claim 1, characterized in that, The specific steps of step S1 are as follows: Deep semantic analysis of railway inspection records is performed to extract railway maintenance information; Identify all fault repair events based on railway maintenance information; The maintenance frequency and number of regional inspections are calculated based on the aforementioned fault maintenance events at different railway locations. The location failure rate is calculated based on the maintenance frequency and the number of regional maintenance visits to obtain the failure probability for multiple locations; The fault repair events are analyzed for the fault time distribution at each fault point, and the fault time sequence distribution data of each fault point is extracted. Identify the severity of the fault in the fault repair event; The maintenance dependency quantification is performed on the severity of the fault, the time-series distribution data of the fault, and the probability of the fault to generate maintenance criticality at multiple location points.

3. The sensor deployment method for railways according to claim 1, characterized in that, The specific steps of step S2 are as follows: Obtain railway facility design drawings; perform topology modeling based on the railway facility design drawings to construct a railway structure model; The key structural points of the railway structural model are identified in depth and marked; the key structural points include main connections, transition points and load-bearing structural points. Railway network connectivity analysis is performed on key structural points to generate structural point connectivity features; Structural criticality is assessed based on the connectivity features of structural points, and a structural criticality score is generated. The signal propagation quality of the key structural points is calculated based on the railway facility design drawings.

4. The sensor deployment method for railways according to claim 3, characterized in that, The specific steps for calculating the signal propagation quality of the key structural points based on railway facility design drawings are as follows: The physical structural parameters of the key structural points are extracted based on the railway facility design drawings; Based on the physical structure parameters, multi-band signal propagation simulation of structural points is performed to generate multi-band signal propagation simulation data. Calculate the signal transmission delay, distortion, and propagation attenuation characteristics of multi-band signal propagation simulation data; The signal propagation quality is calculated based on the signal transmission delay, distortion, and propagation attenuation characteristics to generate the signal propagation quality of key structural points.

5. The sensor deployment method for railways according to claim 1, characterized in that, Step S3 is as follows: Multi-objective solution is performed based on the criticality of maintenance at multiple locations, the criticality score of the structure, and the signal propagation quality to extract the first sensor deployment point; Define the sensor's detection range; The physical detection boundary is calculated and extracted based on the sensor's detection range. Greedy calculations are performed on the railway structure model based on the physical detection boundaries to generate a greedy deployment scheme; Based on the greedy deployment scheme, a full-coverage deployment analysis is performed on the initial sensor deployment points to construct a full-coverage deployment scheme.

6. The sensor deployment method for railways according to claim 1, characterized in that, The specific steps of step S4 are as follows: Set a maintenance cycle window; deploy and process sensors on the railway based on a full-coverage deployment plan; collect signals according to the maintenance cycle window; and extract sensor detection signals. The sensor detects signals and performs environmental background noise identification to extract the environmental background noise. Calculate the signal-to-noise ratio and total harmonic distortion of the ambient background noise to generate the noise interference intensity; The accuracy of each sensor signal is calculated to generate the signal accuracy. The noise interference intensity and signal accuracy are quantitatively analyzed to determine the signal detection quality of each sensor.

7. The sensor deployment method for railways according to claim 1, characterized in that, The specific steps of step S5 are as follows: Based on sensor detection signals, abnormal location is located and sensor spatial distance is calculated to obtain the spatial distance between multiple abnormal locations and sensors. Based on the detection quality, the location of the performance bottleneck is identified, and the performance bottleneck sensor node is marked. Based on the spatial distance, the position adjustment analysis of the performance bottleneck sensor node is performed to generate the second sensor deployment point; The railway coverage rate is calculated based on the second sensor deployment point. If the railway coverage rate exceeds the preset railway coverage rate threshold, the other sensor locations are retained. When the railway coverage rate does not exceed the preset railway coverage rate threshold, the location update calculation of the remaining sensors is performed for the full coverage deployment scheme, and the deployment point of the third sensor is output.

8. The sensor deployment method for railways according to claim 7, characterized in that, The specific steps for locating abnormal locations and calculating the spatial distances between sensors based on sensor detection signals to obtain the spatial distances between multiple abnormal locations and sensors are as follows: Identify and mark abnormal railway parameters based on sensor-detected signals; Based on the anomaly detection parameters, the anomaly location is determined, and the coordinates of the anomaly location are extracted. The spatial distance between the sensor and the abnormal location is calculated based on the coordinates of the abnormal location, and the spatial distance between the sensor and the abnormal location is obtained.

9. A sensor deployment system for railways, characterized in that, A sensor deployment method for a railway as described in claim 1, comprising: The inspection information analysis unit is used to perform multi-location point maintenance analysis on railway inspection records and generate the criticality of maintenance at multiple location points. The structural criticality assessment unit is used to obtain railway facility design drawings; based on the railway facility design drawings, it performs structural criticality assessment and generates a structural criticality score; The full-coverage deployment unit is used to conduct full-coverage deployment analysis based on the criticality of maintenance and structural criticality scores of multiple location points, thereby constructing a full-coverage deployment plan; The detection quality analysis unit is used to process sensor deployments on the railway based on a full-coverage deployment plan and to collect sensor detection signals; it performs quantitative analysis of the signal detection quality of the sensor detection signals and generates the detection quality of each sensor. The position adjustment unit is used to identify and update the position of the performance bottleneck based on the detection quality.

Images