Axle pulse based optical fiber road traffic structure decoupling monitoring method and system

Abstract

The application discloses a kind of based on axle pulse's optical fiber road traffic structure decoupling monitoring method and system, belong to distributed optical fiber sensing, road traffic perception and pavement structure monitoring technical field.The method is set along road and is coupled with pavement vibration sensing optical fiber, continuous collection natural vehicle passage vibration response, constructs distance-time two-dimensional signal matrix, extracts vehicle trajectory, axle pulse, stay and impact event, forms axle pulse fingerprint and equivalent moving load coding.According to similar vehicle cluster, establish self historical reference response library and neighborhood counterfactual reference response library, calculate counterfactual normalized pavement response index;When the same section is continuously abnormal in multiple vehicle clusters, time window or environmental state, and consistent with long-term baseline drift, impact trigger or repeated trajectory residual, output traffic event and suspected pavement structure response abnormal candidate area, and update reference response library and abnormal threshold through review backfill.

Description

Technical Field

[0001] This invention relates to the fields of distributed optical fiber sensing, road traffic condition perception, pavement structure health monitoring, intelligent transportation and digital operation and maintenance of road infrastructure, specifically an optical fiber road traffic structure decoupling monitoring method and system based on axle pulses. Background Technology

[0002] Distributed fiber optic sensing technology uses ordinary optical fibers as continuous linear sensing units, offering advantages such as long sensing distance, resistance to electromagnetic interference, corrosion resistance, continuous coverage, and ease of long-term online monitoring. It has been applied in scenarios such as road settlement monitoring, traffic flow detection, vehicle identification, vehicle trajectory tracking, and road infrastructure safety monitoring. The tire-road interaction, vehicle vibration, braking impact, and road structure response generated during vehicle movement are transmitted to the sensing optical fiber through the road structure, causing changes in the phase, amplitude, or equivalent effect of backscattered light.

[0003] Existing distributed fiber optic solutions for road settlement typically focus on static or quasi-static deformation parameters. For example, they use longitudinal and transverse fibers to form a mesh structure and combine this with deformation surface reconstruction or deflection inversion to obtain the settlement location and deformation amount. The core data objects of this type of solution are settlement amount, deflection information, or structural deformation surfaces. Vehicle traffic is often regarded as interference or additional loads, making it difficult to directly provide dynamic traffic events such as vehicle passage, vehicle stoppage, congestion queues, and impact accidents.

[0004] Existing fiber optic traffic detection solutions typically use distributed fiber optic vibration signals as input for identifying vehicle presence, speed, or traffic flow, and often employ thresholds, trajectory slopes, spectral characteristics, and time-frequency characteristics to output traffic status. While these solutions can identify vehicle events to some extent, they are prone to misinterpreting changes in traffic load as road surface anomalies when faced with differences in response caused by heavy or light vehicles, varying speeds, different lateral distances, different tire conditions, or different fiber optic coupling conditions. This can mask actual road surface anomalies with vehicle variations and environmental noise.

[0005] Existing intersection detection schemes mostly focus on intersection channelized lanes, queuing areas, stop lines, signal phases, or intersection vehicle trajectories. Their output results are usually concentrated on intersection queue length, signal control assistance, and vehicle traffic status, which are not suitable for direct application to traffic-structure joint monitoring of ordinary long-distance road sections.

[0006] Therefore, simply summarizing distributed fiber optic road monitoring as "vibration detection," "traffic flow identification," or "settlement monitoring" is easily confused with existing solutions. A more specific technological chain is needed: one that can convert long-distance fiber optic signals into interpretable road events, perform self-calibration using moving load samples generated by natural vehicles, and further mitigate the impact of vehicle differences, lateral distance differences, and environmental differences on structural anomaly judgments through axle pulse fingerprints and counterfactual response libraries. Summary of the Invention

[0007] The purpose of this invention is to address the disconnect between traffic detection and structural anomaly identification in existing distributed fiber optic road monitoring schemes, the tendency for vehicle load differences to lead to misjudgments of structural anomalies, and the difficulty of interpreting dynamic traffic events through static settlement monitoring. This invention proposes a fiber optic road traffic structure decoupling monitoring method and system based on axle pulses. This method and system can reduce misjudgments caused by heavy vehicles, speed, lateral distance, and environmental changes. Through equivalent moving load coding, similar vehicle clusters, environmental state nodes, and counterfactual reference responses, it can weaken the impact of vehicle differences on the judgment of road structure anomalies, possessing an interpretable chain of evidence. The event map unifies vehicle trajectories, axle pulses, stops, impacts, long-term baselines, and structural anomaly candidate nodes, facilitating the provision of event location, causal evidence, and verification basis. It is suitable for long-term engineering deployment; the shallow, flush-layout design avoids forming obvious protrusions or steps. The system can be continuously updated by combining contradictory samples and on-site verification labels, adapting to changes in road structure, traffic composition, seasonal environment, and equipment status.

[0008] The technical solution to achieve the objective of this invention is:

[0009] The fiber optic road traffic structure decoupling monitoring method based on axle pulses includes the following steps:

[0010] At least one sensing optical fiber is installed along a typical continuous road section so that the sensing optical fiber forms a vibration coupling with the road surface structure;

[0011] A distributed fiber optic demodulation device connected to the sensing fiber continuously collects road vibration responses distributed along the fiber optic distance direction under natural vehicle traffic, and constructs a distance-time two-dimensional signal matrix S(x) based on the spatial sampling location and sampling time. i ,t j );

[0012] The distance-time two-dimensional signal matrix is ​​subjected to distance calibration, time synchronization, background baseline subtraction, filtering and denoising, amplitude normalization and outlier removal to obtain a standardized signal matrix;

[0013] Vehicle trajectory strips, axle pulse sequences, dwell response segments, and impact response segments are extracted from the standardized signal matrix, and an axle pulse fingerprint is formed based on the axle pulse sequence.

[0014] Equivalent moving load codes are generated based on vehicle trajectory strips and axle pulse fingerprints, and vehicle events with similar codes are grouped into similar vehicle clusters.

[0015] A road event map is constructed based on vehicle trajectory strips, axle pulse fingerprints, dwell response segments, impact response segments, equivalent moving load codes, and environmental state information during the monitoring period;

[0016] Based on the vehicle trajectory nodes, stopping nodes and impact nodes in the road event map, vehicle passing, vehicle stopping, slow traffic congestion and abnormal impact are identified to obtain traffic event identification results; the road is divided into multiple road segment response fingerprint units along the optical fiber direction, and a historical reference response library and a neighborhood counterfactual reference response library are established for each road segment response fingerprint unit and each equivalent moving load code or similar vehicle cluster.

[0017] The current response of a real-time vehicle event at the road segment response fingerprint unit is compared with the self-historical reference response library and the neighborhood counterfactual reference response library to calculate the counterfactual normalized pavement response index CC-NRI (CC-NRI is a structural response anomaly evaluation index. It robustly normalizes and weights the current response relative to the amplitude, frequency band, propagation, and baseline of various residuals generated by the self-historical reference and the neighborhood counterfactual reference).

[0018] When the CC-NRI of the response fingerprint unit of the same road segment repeatedly exceeds the abnormal threshold in multiple similar vehicle clusters, multiple discontinuous time windows or multiple environmental conditions, and is consistent with at least one piece of evidence in the event map, such as long-term baseline drift, impact triggering relationship or repeated trajectory residual, the response fingerprint unit of the road segment is marked as a suspected candidate area for road structure response anomaly, and the traffic event, the candidate area for structural response anomaly, the evidence map and the warning level are output.

[0019] The sensing optical fiber is installed in a shallow groove below the lane divider, in a shallow groove at the lane edge, in a shallow groove on the shoulder, or in a protective groove near the central median. When installed in a shallow groove below the lane divider, the groove has a depth of 6 mm to 30 mm and a width of 6 mm to 25 mm. From bottom to top, the groove contains an adhesive leveling layer, a sensing optical fiber or a micro-protective tube containing the sensing optical fiber, a flexible coupling filling layer, and a marking compatible surface layer. After curing, the marking compatible surface layer is flush with the adjacent road surface or has a height difference of no more than 3 mm, thereby obtaining the natural vehicle dynamic load response without significantly changing the road surface structure or affecting vehicle crossing or lane changing.

[0020] The distance-time two-dimensional signal matrix S(x) i ,t jIn a traffic jam, a vehicle passing normally is typically represented by a diagonal strip that moves over time; a vehicle standing still is represented by a continuous response near a fixed spatial location; slow traffic congestion is represented by the aggregation of multiple low-speed trajectories and standing responses in space and time; and a shock event is represented by a sudden high-amplitude, wide-bandwidth, and spatially diffuse response.

[0021] The standardized signal matrix is ​​decomposed into at least two frequency bands to obtain a multi-band energy matrix B. m(xi,tj) The system calculates local energy using sliding spatial windows and sliding temporal windows to extract valid event fragments. The axle pulse sequence is obtained by sorting the envelope peaks in the neighborhood of the vehicle trajectory strip by time. Information such as the number of axles, the time interval between axles, and the wheelbase are obtained from the vehicle vibration response. The axle pulse fingerprint formed by the axle pulse sequence includes a combination of pulse number, time interval between adjacent pulses, equivalent wheelbase sequence converted from vehicle speed, relative amplitude of each pulse, pulse width, and pulse frequency band energy ratio, which is used for subsequent equivalent moving load encoding. The axle pulse fingerprint can reflect the number of vehicle axles, equivalent wheelbase, relative pulse intensity, and axle group vibration characteristics, so that the natural vehicle is no longer just a detection target, but a reusable non-dedicated moving load probe.

[0022] The equivalent moving load code Q is represented as Q=[v,E] v ,f m B r ,L s ,T d ,L f D l A p ], where v is the trajectory velocity, E v For event energy, f m Main frequency, B r For the frequency band energy ratio, L s T represents the spatial coverage length. d For duration, L f For the proportion of low-frequency energy, D l For lateral distance proxy, A p The axle pulse fingerprint is used; the lateral distance proxy quantity is determined by at least two of the following: trajectory strip width, peak spatial offset, energy ratio of adjacent spatial sampling points, low-frequency and mid-to-high-frequency energy attenuation ratio, or relative relationship with lane geometry; when the standardized distance between the equivalent moving load code of a vehicle event and the center of an existing coding cluster is less than the cluster matching threshold, the vehicle event is assigned to the corresponding similar vehicle cluster; otherwise, a new coding cluster is formed or stored in the vehicle cluster to be reviewed.

[0023] The effective event fragments and equivalent moving load codes construct a road event map G=(V,E), where nodes V include at least four types of nodes: vehicle trajectory nodes, axle pulse nodes, stationary nodes, impact nodes, long-term baseline nodes, environmental state nodes, and structural anomaly candidate nodes; edges E include at least four types of edges: temporally adjacent edges, spatially adjacent edges, trajectory continuation edges, axle-related edges, same-load reference edges, baseline drift edges, and anomaly triggering edges; the length of the road segment response fingerprint unit is determined based on the distributed optical fiber spatial resolution, road structure layer thickness, and vehicle wheel track influence width, and is between 0.5 m and 5 m.

[0024] The historical reference response library includes R self(u,q,e) MAD self(u,q,e) n self(u,q,e) And update time, the neighborhood counterfactual reference response library includes R cf(u,q,e) MAD cf(u,q,e) n cf(u,q,e) The neighborhood weights are defined as follows: u is the road segment response fingerprint unit, q is the equivalent moving load code or similar vehicle cluster, and e is the environmental state. The "historical reference response library" adopts the basic idea of ​​comparing historical health baselines in structural health monitoring, representing the historical response set accumulated when the same road segment response fingerprint unit is confirmed to be in a normal state under the same or similar equivalent moving load and environmental state. The "neighborhood counterfactual reference response library" is a reference response construction method proposed for structural anomaly judgment. It represents the estimate of "what kind of response should be generated if the location remains in a normal state"; the neighborhood counterfactual reference response library is generated by weighting the responses of the same equivalent moving load code in adjacent road segment response fingerprint units, historical health periods, or reviewed normal road segments, and samples marked as abnormal, construction, strong noise, or equipment abnormality are removed; the CC-NRI is generated according to CC-NRI. (u,q,e) =Σw m ·ρ m (D m(u,q,e) / Scale m(u,q,e) The calculation is performed, where Dm includes the amplitude residual D. A Band residual D f Propagation residual D p and baseline drift residual D B At least two of the following: Scale m For the median absolute deviation, interquartile range, or preset noise reference, w m For the weight of evidence, ρ m This is either a truncation function or a robust normalization function.

[0025] When the number of valid samples in the historical reference response library or the neighborhood counterfactual reference response library is lower than the preset minimum sample size, or when there is a contradiction between the traffic event identification result and the CC-NRI structural anomaly judgment result, the system will store the relevant original signal segments, standardized signal segments, axle pulse fingerprints, equivalent moving load codes, event maps, CC-NRI, and identification results into the sample library to be reviewed. After obtaining at least one review label from video surveillance, traffic checkpoints, dynamic weighing, manual inspection, pavement defect detection, or road maintenance records, the system will incrementally update the vehicle cluster center, the historical reference response library, the neighborhood counterfactual reference response library, and the anomaly threshold with version numbers.

[0026] A fiber optic road traffic structure decoupling monitoring system based on axle pulses is disclosed. This system executes the aforementioned fiber optic road traffic structure decoupling monitoring method based on axle pulses. The system includes shallow-groove flush-mount sensing fiber, distributed fiber optic demodulation equipment, roadside edge computing units or monitoring platforms, and modules for constructing a distance-time matrix, standardization preprocessing, event fragment extraction, axle pulse fingerprint extraction, equivalent moving load coding, road surface event map generation, road segment response fingerprint database, neighborhood counterfactual reference response generation, CC-NRI calculation, traffic-structure decoupling determination, evidence map output, and verification and update.

[0027] The shallow-groove flush-mount sensing fiber optic cable forms vibration coupling with the road surface structure and is optically connected to a distributed fiber optic demodulation device. The distributed fiber optic demodulation device is communicatively connected to a roadside edge computing unit or monitoring platform and sends the road vibration response distributed along the fiber optic distance direction to the distance-time matrix construction module. The distance-time matrix construction module, standardized preprocessing module, event fragment extraction module, axle pulse fingerprint extraction module, and equivalent moving load encoding module are sequentially connected. The road event map generation module receives vehicle trajectory, axle pulse, stationary events, impact, equivalent moving load encoding, and... Structural anomaly candidate information is used to generate a road event map for evidence association; the road segment response fingerprint database module is connected to the neighborhood counterfactual reference response generation module, and together they provide historical reference responses and neighborhood counterfactual reference responses to the CC-NRI calculation module; the CC-NRI calculation module is connected to the traffic-structure decoupling determination module, the traffic-structure decoupling determination module is connected to the evidence map output module, and the review and backfeed update module is connected to the vehicle cluster center, the road segment response fingerprint database module, the neighborhood counterfactual reference response generation module, and the anomaly threshold, and is used to perform reverse updates based on the review labels.

[0028] For specific types of defects such as voids, cracks, subsidence, loosening, or local stiffness degradation, the system outputs candidate suggestions and verification criteria, and does not use a single fiber vibration response as the final diagnostic basis.

[0029] This technical solution achieves the following effects:

[0030] The system elevates natural vehicles from traffic detection targets to non-dedicated mobile load probes. Through axle pulse fingerprinting and equivalent mobile load coding, the system can establish similar mobile load samples without dedicated loading vehicles or strong reliance on fixed weighing equipment.

[0031] Unlike simple settlement or deflection monitoring, this system does not rely on static settlement or the reconstruction of interwoven fiber optic mesh surfaces as the primary basis for judgment. Instead, it uses the continuous deviation of the response residuals under the same natural moving load as evidence of structural anomalies.

[0032] Unlike simple traffic flow detection, this system not only outputs vehicle flow, speed, dwell time, and congestion, but also uses a dual reference response library and CC-NRI to identify candidate areas with suspected abnormal road surface structure responses, providing evidence maps and verification criteria.

[0033] Reduce misjudgments caused by heavy vehicles, speed, lateral distance, and environmental changes. By using equivalent moving load coding, similar vehicle clusters, environmental state nodes, and counterfactual reference responses, the impact of vehicle differences on the judgment of pavement structure anomalies can be mitigated.

[0034] It possesses an interpretable chain of evidence. The event map unifies vehicle trajectory, axle pulses, stops, impacts, long-term baselines, and candidate nodes for structural anomalies, facilitating the identification of event locations, causal evidence, and verification criteria.

[0035] Suitable for long-term engineering deployment. The shallow, flush layout does not create obvious protrusions or steps. The system can be continuously updated by combining conflict samples and on-site verification labels, adapting to changes in road structure, traffic composition, seasonal environment, and equipment status. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the system structure in the embodiment;

[0037] Figure 2 This is a flowchart illustrating the self-calibration process of the axle pulse fingerprint and counterfactual response library in the embodiment.

[0038] Figure 3 This is a schematic diagram of the relationship between nodes and edges in the road event graph G=(V,E) in the embodiment.

[0039] Figure 4 This is a schematic diagram illustrating the calculation of the dual reference response library and the counterfactual normalization index CC-NRI in the embodiment;

[0040] Figure 5 This is a schematic diagram of the traffic-structure decoupling determination, evidence output, and review and recharge closed loop in the embodiment. Detailed Implementation

[0041] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. These embodiments are only used to explain the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0042] Example 1: Shallow Groove Flush-Type Sensor Fiber Laying and Signal Acquisition

[0043] like Figure 1 As shown, in a test section of urban road, expressway, or ordinary trunk road with a length of 1000 m to 5000 m, shallow trenches are dug below the lane dividers or along the lane edges. The depth of the shallow trenches can be 6 mm to 30 mm, and the width can be 6 mm to 25 mm. An adhesive leveling layer is installed at the bottom of the shallow trench, a sensing optical fiber or a miniature protective tube containing a sensing optical fiber is installed in the middle, and a flexible coupling filling layer and a marking-compatible surface layer are installed on top.

[0044] One or both ends of the sensing optical fiber are connected to the distributed optical fiber demodulation equipment in the roadside monitoring station. Vibrations caused by vehicles passing by, stopping, braking, colliding, impacting, or local stiffness changes in the road structure are transmitted to the sensing optical fiber through the road surface structure, causing changes in the phase, amplitude, or equivalent effect of backscattered light. The demodulation equipment converts these changes into vibration sampling data continuously distributed along the road direction.

[0045] The shallow groove flush layout in this embodiment is mainly used to obtain the dynamic response under natural vehicle moving loads, which ensures vibration coupling and avoids the formation of protrusions or steps at the lane divider that affect lane crossing, lane changing or straddling.

[0046] Example 2: Extraction of normalized matrix, event fragments, and axle pulse fingerprints, such as Figure 2 As shown,

[0047] The system arranges vibration data from different spatial locations along the sensing fiber at the same sampling time into a column, and arranges data from consecutive sampling times in chronological order to construct a distance-time two-dimensional signal matrix S(x). i ,t j ). Where x i Let t represent the i-th spatial sampling position. j Let S(x) represent the j-th sampling time. i ,t j This indicates the vibration amplitude, phase change, or equivalent strain change at that location and time.

[0048] The system for S(x) i ,t jThe system performs distance calibration, time synchronization, background baseline subtraction, bandpass filtering, wavelet denoising, amplitude normalization, and outlier removal. For each spatial sampling location, the system sets an adaptive threshold T based on the mean and standard deviation of the background signal when there are no significant vehicle events. i =μ i +k·σ i The local energy E is calculated using a sliding spatial window and a sliding time window. w .

[0049] Candidate event regions are analyzed for spatial connectivity, temporal continuity, and trajectory continuity to form valid event segments. For vehicle passage events, the system extracts the envelope peak sequence within the neighborhood of the vehicle trajectory strip to obtain the axle pulse sequence. The time intervals between adjacent pulses are converted into an equivalent axle distance sequence based on the vehicle speed, and the axle pulse fingerprint is formed by combining the number of pulses, relative amplitude, pulse width, and frequency band energy ratio.

[0050] Example 3: Equivalent moving load coding and pavement event map construction, such as Figure 3 As shown,

[0051] The system will use vehicle trajectory speed v and event energy E v , main frequency f m Bandwidth energy ratio B r Spatial coverage length L s Duration T d Low-frequency energy ratio L f Lateral distance proxy quantity D l And axle pulse fingerprint A p The equivalent moving load encoding is Q=[v,E] v ,f m B r ,L s ,T d ,L f D l A p The lateral distance proxy can be determined by the statistical relationship between the energy attenuation ratio of different frequency bands, the width of the trajectory strip, the peak spatial offset, the energy ratio of adjacent spatial sampling points, or the relative position with the lane divider.

[0052] When the normalized distance between the equivalent moving load code of a vehicle event and the existing cluster center is less than the cluster matching threshold T c When the vehicle event occurs, the system categorizes it into a corresponding similar vehicle cluster; otherwise, it forms a new similar vehicle cluster or stores it in a cluster awaiting review. In this way, natural traffic flow is organized into samples of different equivalent moving loads for subsequent road segment response self-calibration. The vehicle cluster matching threshold T... cThis is a standardized distance threshold used to determine whether the current equivalent moving load code matches an existing vehicle cluster. The T... c The vehicle cluster matching threshold T can be determined based on the intra-cluster standardized distance distribution of the reviewed vehicle samples, for example, based on the distance quantile values ​​of historical samples, or through offline calibration using labeled samples; c Compared with the adaptive event threshold T used for candidate event extraction in Example 2 i Each can be set independently.

[0053] The system constructs a road event map G=(V,E) based on valid event fragments. Vehicle trajectory nodes record location intervals, time intervals, trajectory slopes, travel direction, and speed; axle pulse nodes record the number of axles, equivalent wheelbase, and relative amplitude; dwell nodes record dwell location, duration, location fluctuation range, and dwell index; impact nodes record peak value, kurtosis, high-frequency energy proportion, and spatial diffusion width; long-term baseline nodes record the median response, dispersion, and long-term drift of the road segment response fingerprint units.

[0054] Example 4: Dual reference response library, CC-NRI calculation and structural anomaly determination, such as Figure 4 As shown,

[0055] The system divides the road into several segments along the direction of the sensing fiber, responding to the fingerprint unit U. r Each road segment response fingerprint unit can correspond to one or more fiber optic spatial sampling points. Its length can be determined based on spatial resolution, road structure layer thickness, and vehicle wheel track influence width, preferably 0.5 m to 5 m.

[0056] For each road segment response fingerprint unit u and equivalent moving load code q, the system establishes a historical reference response library and a neighborhood counterfactual reference response library. self(u,q,e) Responses from the same road segment during historical healthy periods or in previously verified normal states; neighborhood counterfactual reference response library R cf(u,q,e) The system generates a weighted combination of responses from adjacent road sections, historically healthy road sections, or verified normal road sections under the same or similar equivalent moving loads. Samples marked as abnormal, under construction conditions, experiencing strong noise, or malfunctioning equipment are removed, and the number of samples, update time, and reference library version are recorded. Environmental condition e may include temperature, rainfall, road surface slipperiness, construction status, and equipment noise level.

[0057] When a real-time vehicle event passes through the road segment response fingerprint unit u, the system obtains the current response R. now(u,q,e) The current response R now(u,q,e)The road segment response fingerprint unit forms a set of response features under the current vehicle event, including at least two of the following: amplitude features, frequency band energy features, spatial propagation features, and baseline features. Robust center values ​​and dispersion parameters of these response features are stored in the historical reference response database and the neighborhood counterfactual reference response database, respectively, and the amplitude residual D is calculated for each. A Band residual D f Propagation residual D p and baseline drift residual D B The propagation residual may include changes in the correlation coefficient between adjacent spatial sampling points, changes in spatial diffusion width, or changes in propagation velocity. The system conforms to CC-NRI standards. (u,q,e) =Σw m ·ρ m (D m(u,q,e) / Scale m(u,q,e) ) Calculate the counterfactual normalized pavement response index, where Scale m The median absolute deviation, interquartile range, or a preset noise reference can be used, ρ m This is either a truncation function or a robust normalization function.

[0058] If the CC-NRI of a road segment's response fingerprint unit increases only during a single heavy vehicle passage and does not reappear in other equivalent moving load codes or other time periods, the system treats it as a vehicle load difference or an observation sample. If the CC-NRI remains consistently high in at least two similar vehicle clusters, at least two discontinuous time windows, or at least two environmental conditions, and this spatial location is consistent with long-term baseline node drift, abnormal triggering of impact nodes, or repeated vehicle trajectory residuals, the system identifies it as a candidate region for suspected pavement structure response anomalies.

[0059] Example 5: Traffic incident identification, evidence map output and verification feedback, such as... Figure 5 As shown,

[0060] Vehicle passage events are identified by recognizing the oblique trajectory formed in a distance-time two-dimensional signal matrix based on the continuous change of the event center position over time. Vehicle speed can be determined by v=(x... end -x start ) / (t end -t start )Calculation. Vehicle dwelling events are identified based on the fact that the change in the event center location within a preset time is less than a spatial threshold and the duration is greater than a dwelling time threshold. The dwelling index can be calculated as R. s =N(|x c(t) -x bar | <delta x ) / N totalCalculation. Slow-moving congestion events are identified by the aggregation of multiple low-speed vehicle trajectory nodes or stationary nodes within a continuous spatial and temporal range. Suspected accidents are identified by trajectory termination after an impact node, prolonged vehicle stagnation, or queuing behind the vehicle.

[0061] The system outputs the event location, event time, vehicle speed, dwell time, congestion range, suspected accident location, candidate areas for suspected pavement structure response anomalies, evidence map, and warning level. The evidence map includes at least the vehicle trajectory node that triggered the warning, axle pulse node, equivalent moving load code, reference response library version, CC-NRI curve, and review status; for specific defect types, the system outputs candidate types and review suggestions.

[0062] When the number of valid samples in the historical reference response library or the neighborhood counterfactual reference response library is lower than the preset minimum sample size, or when there is a contradiction between the traffic event identification result and the CC-NRI structural anomaly judgment result, the system stores the relevant original signal segments, standardized signal segments, axle pulse fingerprints, equivalent moving load codes, event maps, CC-NRI, and identification results into the sample library to be reviewed. Review data can come from video surveillance, traffic checkpoints, dynamic weighing, manual inspections, pavement defect detection, or road maintenance records. After obtaining the review label, the system incrementally updates the vehicle cluster center, the historical reference response library, the neighborhood counterfactual reference response library, and the anomaly threshold, and retains the version number for engineering traceability.

[0063] The complete working process of this example

[0064] In this example, the sensing fiber continuously collects the road vibration response, and the distributed fiber demodulation equipment converts it into a signal distributed along the road direction; the matrix construction module generates a distance-time two-dimensional signal matrix; the preprocessing module removes background noise, light attenuation differences, and anomalies; the event fragment extraction module determines candidate events and valid event fragments; the axle pulse fingerprint module extracts pulse sequences from vehicle trajectory strips; the equivalent moving load encoding module forms clusters of similar vehicles; the event map generation module organizes vehicle trajectories, axle pulses, stops, impacts, long-term baselines, and structural anomaly candidates into a graph structure; the counterfactual reference response generation module establishes a dual reference response library; the CC-NRI calculation module calculates multi-evidence structural residuals; the traffic-structure decoupling judgment module outputs traffic events and suspected structural anomalies; the evidence map output module outputs location, time, category, warning level, and review basis to the management platform; and the review and feedback module continuously updates the model and reference library based on the on-site review results.

[0065] From the perspective of the module connections, the sensing fiber, as the front-end sensing unit, is buried or fixed in shallow trenches or protective trenches. Its output optical response is converted into computable vibration data by a distributed fiber demodulation device. The vibration data first enters the matrix construction module and the standardization preprocessing module to form a standardized signal matrix under unified spatial and temporal coordinates, and then is transmitted to the event fragment extraction module. The event fragment extraction module sends vehicle passing, stopping, and impact fragments to the axle pulse fingerprint extraction module and the equivalent moving load encoding module, respectively, to obtain similar vehicle clusters used to distinguish differences in vehicle loads. The similar vehicle clusters are written into the road event map generation module to form spatiotemporal correlations between vehicle trajectory nodes, axle pulse nodes, stopping nodes, and impact nodes. On the other hand, they are written as indexes into the road segment response fingerprint database module and the neighborhood counterfactual reference response generation module. The CC-NRI calculation module simultaneously reads the real-time response, the historical reference response, and the neighborhood counterfactual reference response, calculates the multi-evidence structural residual, and outputs it to the traffic-structure decoupling judgment module. The judgment result is then transmitted to the evidence map output module to generate warnings, locations, and verification basis. After receiving verification tags such as video, checkpoint, dynamic weighing, manual inspection or maintenance records, the verification and recharge update module updates the vehicle cluster center, reference response library, anomaly threshold and model version in reverse, thus forming a closed loop from perception, processing, judgment, output to feedback update.

Claims

1. A fiber optic road traffic structure decoupling monitoring method based on axle pulses, characterized in that, Includes the following steps: At least one sensing optical fiber is installed along a typical continuous road section so that the sensing optical fiber forms a vibration coupling with the road surface structure; A distributed fiber optic demodulation device connected to the sensing fiber continuously collects road vibration responses distributed along the fiber optic distance direction under natural vehicle traffic, and constructs a distance-time two-dimensional signal matrix S(x) based on the spatial sampling location and sampling time. i ,t j ); The distance-time two-dimensional signal matrix is ​​subjected to distance calibration, time synchronization, background baseline subtraction, filtering and denoising, amplitude normalization and outlier removal to obtain a standardized signal matrix; Vehicle trajectory strips, axle pulse sequences, dwell response segments, and impact response segments are extracted from the standardized signal matrix, and an axle pulse fingerprint is formed based on the axle pulse sequence. Equivalent moving load codes are generated based on the vehicle trajectory strips and axle pulse fingerprints, and vehicle events with similar codes are grouped into similar vehicle clusters. A road event map is constructed based on vehicle trajectory strips, axle pulse fingerprints, dwell response segments, impact response segments, equivalent moving load codes, and environmental state information during the monitoring period; Based on the vehicle trajectory nodes, stopping nodes, and impact nodes in the road event map, vehicle passage, vehicle stopping, slow traffic congestion, and abnormal impact are identified to obtain traffic event identification results. The road is divided into multiple road segment response fingerprint units along the fiber optic direction. For each road segment response fingerprint unit and each equivalent moving load code or similar vehicle cluster, a historical reference response library and a neighborhood counterfactual reference response library are established. The current response of a real-time vehicle event at the road segment response fingerprint unit is compared with the historical reference response library and the neighborhood counterfactual reference response library to calculate the counterfactual normalized road response index CC-NRI. When the CC-NRI of the response fingerprint unit of the same road segment repeatedly exceeds the abnormal threshold in multiple similar vehicle clusters, multiple discontinuous time windows or multiple environmental conditions, and is consistent with at least one piece of evidence in the event map, such as long-term baseline drift, impact triggering relationship or repeated trajectory residual, the response fingerprint unit of the road segment is marked as a suspected candidate area for road structure response anomaly, and the traffic event, the candidate area for structural response anomaly, the evidence map and the warning level are output.

2. The fiber optic road traffic structure decoupling monitoring method based on axle pulses according to claim 1, characterized in that, The sensing optical fiber is installed in a shallow groove below the lane divider, in a shallow groove at the lane edge, in a shallow groove on the shoulder, or in a protective groove near the central median. When installed in a shallow groove below the lane divider, the groove has a depth of 6 mm to 30 mm and a width of 6 mm to 25 mm. From bottom to top, the groove is provided with an adhesive leveling layer, a sensing optical fiber or a micro-protective tube containing the sensing optical fiber, a flexible coupling filling layer, and a marking compatible surface layer. After curing, the marking compatible surface layer is flush with the adjacent road surface or has a height difference of no more than 3 mm.

3. The fiber optic road traffic structure decoupling monitoring method based on axle pulses according to claim 1, characterized in that, The standardized signal matrix is ​​decomposed into at least two frequency bands to obtain a multi-band energy matrix B. m(xi,tj) The standardized signal matrix is ​​decomposed into at least two frequency bands according to a preset frequency range. For the m-th frequency band, the frequency band signal S is obtained. m (x i ,t j ), and calculate the local energy of the signal in that frequency band, forming the multi-band energy matrix Bm(x) corresponding to the m-th frequency band. i ,t j The system calculates local energy using sliding spatial windows and sliding time windows to extract valid event fragments. The axle pulse sequence is obtained by sorting the envelope peaks in the neighborhood of the vehicle trajectory strip by time. The number of axles, the time interval between axles, and the wheelbase information are obtained from the vehicle vibration response. Based on this information, the number of pulses, the time interval between adjacent pulses, the equivalent wheelbase sequence converted from the vehicle speed, the relative amplitude of each pulse, the pulse width, and the frequency band energy ratio are further defined as the axle pulse fingerprint, which is used for subsequent equivalent moving load coding.

4. The fiber optic road traffic structure decoupling monitoring method based on axle pulses according to claim 3, characterized in that, The equivalent moving load code Q is represented as Q=[v,E] v ,f m B r ,L s ,T d ,L f D l A p ], where v is the trajectory velocity, E v For event energy, f m Main frequency, B r For the frequency band energy ratio, L s T represents the spatial coverage length. d For duration, L f For the proportion of low-frequency energy, D l For lateral distance proxy, A p The equivalent moving load coding, based on axle pulse fingerprints, utilizes parameters obtained from distributed fiber optic responses, including speed, event energy, frequency band characteristics, axle pulse fingerprints, and lateral distance surrogate quantity, to equivalently characterize the moving load effect generated by natural vehicles. This allows for the selection of vehicle events with similar characteristics for structural response comparison. The lateral distance surrogate quantity is determined by at least two of the following: trajectory strip width, peak spatial offset, energy ratio of adjacent spatial sampling points, low-frequency to mid-to-high-frequency energy attenuation ratio, or relative relationship with lane geometry. When the standardized distance between the equivalent moving load code of a vehicle event and the center of an existing coding cluster is less than the cluster matching threshold, the vehicle event is assigned to the corresponding similar vehicle cluster; otherwise, a new coding cluster is formed or stored in a vehicle cluster awaiting review.

5. The fiber optic road traffic structure decoupling monitoring method based on axle pulses according to claim 4, characterized in that, The effective event fragments and equivalent moving load codes construct a road event map G=(V,E), where nodes V include at least four types of nodes: vehicle trajectory nodes, axle pulse nodes, stationary nodes, impact nodes, long-term baseline nodes, environmental state nodes, and structural anomaly candidate nodes; edges E include at least four types of edges: temporally adjacent edges, spatially adjacent edges, trajectory continuation edges, axle-related edges, same-load reference edges, baseline drift edges, and anomaly triggering edges; the length of the road segment response fingerprint unit is determined based on the distributed optical fiber spatial resolution, road structure layer thickness, and vehicle wheel track influence width, and is between 0.5 m and 5 m.

6. The fiber optic road traffic structure decoupling monitoring method based on axle pulses according to claim 1, characterized in that, The historical reference response library includes R self(u,q,e) MAD self(u,q,e) n self(u,q,e) And update time, the neighborhood counterfactual reference response library includes R cf(u,q,e) MAD cf(u,q,e) n cf(u,q,e) The neighborhood weights are defined as follows: u is the road segment response fingerprint unit, q is the equivalent moving load code or similar vehicle cluster, and e is the environmental state. The neighborhood counterfactual reference response library is generated by weighting the responses of the same equivalent moving load code in adjacent road segment response fingerprint units, historical healthy periods, or reviewed normal road segments, and samples marked as abnormal, construction, strong noise, or equipment abnormality are removed. The CC-NRI is based on CC-NRI (u,q,e) =Σw m ·ρ m (D m(u,q,e) / Scale m(u,q,e) The calculation is performed, where Dm includes the amplitude residual D. A Band residual D f Propagation residual D p and baseline drift residual D B At least two of the following: Scale m For the median absolute deviation, interquartile range, or preset noise reference, w m For the weight of evidence, ρ m This is either a truncation function or a robust normalization function.

7. The fiber optic road traffic structure decoupling monitoring method based on axle pulses according to claim 6, characterized in that, When the number of valid samples in the historical reference response library or the neighborhood counterfactual reference response library is lower than the preset minimum sample size, or when there is a contradiction between the traffic event identification result and the CC-NRI structural anomaly determination result, the original signal segment, standardized signal segment, axle pulse fingerprint, equivalent moving load code, road event map, CC-NRI, identification result and evidence map are stored in the sample library to be reviewed; after obtaining at least one review label from video surveillance, traffic checkpoints, dynamic weighing, manual inspection, road surface defect detection or road maintenance records, the similar vehicle clusters, the historical reference response library, the neighborhood counterfactual reference response library and the anomaly threshold are incrementally updated with version numbers.

8. A fiber optic road traffic structure decoupling monitoring system based on axle pulses, characterized in that, The system is used to execute the fiber optic road traffic structure decoupling monitoring method based on axle pulses as described in any one of claims 1-7. The system includes a shallow-groove flush-mount sensing fiber, a distributed fiber optic demodulation device, a roadside edge computing unit or monitoring platform, and a distance-time matrix construction module, a standardization preprocessing module, an event fragment extraction module, an axle pulse fingerprint extraction module, an equivalent moving load coding module, a road surface event map generation module, a road segment response fingerprint database module, a neighborhood counterfactual reference response generation module, a CC-NRI calculation module, a traffic-structure decoupling determination module, an evidence map output module, and a review and recharge update module. The shallow groove flush sensing optical fiber forms vibration coupling with the road surface structure and is optically connected to the distributed optical fiber demodulation equipment. The distributed optical fiber demodulation device is communicatively connected to the roadside edge computing unit or monitoring platform, and sends the road vibration response distributed along the optical fiber distance direction to the distance-time matrix construction module; The distance-time matrix construction module, the standardized preprocessing module, the event fragment extraction module, the axle pulse fingerprint extraction module, and the equivalent moving load encoding module are connected in sequence. The road event map generation module receives vehicle trajectory, axle pulse, stop, impact, equivalent moving load code and structural anomaly candidate information respectively, and generates a road event map for evidence association. The road segment response fingerprint database module is connected to the neighborhood counterfactual reference response generation module, and together they provide the historical reference response and the neighborhood counterfactual reference response to the CC-NRI calculation module. The CC-NRI calculation module is connected to the traffic-structure decoupling determination module, the traffic-structure decoupling determination module is connected to the evidence map output module, and the review and recharge update module is connected to the vehicle cluster center, road segment response fingerprint database module, neighborhood counterfactual reference response generation module, and anomaly threshold, and is used to perform reverse updates based on the review labels.

Images