A bridge dynamic response monitoring and early warning method

Abstract

The application relates to the technical field of bridge structure health early warning, and discloses a bridge dynamic response monitoring and early warning method, which comprises the following steps: identifying a single heavy-load vehicle and extracting physical space geometric data of the vehicle; performing phase alignment processing on original dynamic response time sequences collected by different observation sections based on the physical space geometric data, so as to obtain dynamic response time sequences; calculating dynamic strain transfer functions between adjacent observation sections based on the dynamic response time sequences and generating a longitudinal space transfer state vector; performing probability distance measurement on the longitudinal space transfer state vector and a health benchmark, and determining whether to output a trigger instruction based on the measurement result. The application verifies and regularizes the response signal by using the physical space geometric data, and performs state evaluation in combination with the probability distance measurement, so that the monitoring data distortion problems caused by uncertain excitation sources and vehicle speed change are solved, and the accuracy and robustness of early warning decision are improved.

Description

Technical Field

[0001] This invention relates to the field of bridge structural health early warning technology, specifically a method for monitoring and early warning of bridge dynamic response. Background Technology

[0002] As a critical node in transportation infrastructure, the structural safety of bridges is of paramount importance. Monitoring the dynamic response of bridges under vehicle loads using sensors is a common technique for assessing their health. In practical applications, the method of using normally passing vehicles as natural vibration sources for monitoring has attracted considerable attention due to its low cost and lack of traffic disruption.

[0003] However, implementing such monitoring in open traffic environments presents challenges. Actual traffic flow is complex and variable; multiple vehicles simultaneously on the bridge generate coupled vibrations, making it difficult to correlate the sensor-collected response signals with a single, clearly defined excitation source, leading to uncertainty in the data source. Furthermore, the speed of vehicles traveling on the bridge is not constant; acceleration or deceleration causes nonlinear stretching or compression of the structural response at different spatial locations along the time axis. This phase distortion makes direct alignment and comparison of signals from different measurement points difficult, introducing significant errors if monitoring is based on calculating structural transmission characteristics.

[0004] Therefore, the characteristic parameters extracted from the above-mentioned contaminated and phase-distorted signals have poor stability. When used directly for condition assessment, they are difficult to effectively distinguish between the actual attenuation of structural stiffness and the instantaneous data jumps caused by accidental factors such as vehicle driving posture and environmental noise. This leads to a high risk of misjudgment in existing early warning methods. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for monitoring and early warning of bridge dynamic response, which solves the problems of distortion of dynamic response data due to uncertain excitation sources and vehicle speed changes, and the resulting low reliability of early warning decisions.

[0006] To achieve the above objectives, the present invention provides a method for monitoring and early warning of bridge dynamic response, comprising the following steps: A single heavy-duty vehicle is identified as a mobile excitation source, and its physical wheelbase sequence is extracted. Simultaneously, the original dynamic response time sequences of multiple observation sections of the bridge are acquired, and transient elastic strain sequences are obtained based on a zero-phase-shift bandpass filter. Peak time nodes corresponding to the axles are extracted, and spatiotemporal self-consistency is verified based on the physical wheelbase sequence to establish a set of rigid anchor points. Based on this set of rigid anchor points, constrained dynamic time warping is performed on the transient elastic strain sequences of adjacent observation sections to generate dynamic response time sequences. Based on the peak time nodes in the dynamic response time sequences, effective integration intervals are determined, and the dynamic strain transfer function between adjacent observation sections is calculated and combined into a longitudinal spatial transfer state vector. The Mahalanobis distance between this longitudinal spatial transfer state vector and the health baseline is calculated, and a trigger command is output in conjunction with continuous event judgment logic.

[0007] Preferably, the steps for identifying a single heavy-duty vehicle include: determining whether the total weight of the vehicle exceeds a preset weight threshold; and checking whether there are any other vehicle passage records within the time buffer zone before and after the time the vehicle enters the measuring point.

[0008] Preferably, the value of the preceding and following time buffer zone is calculated based on the total longitudinal span of the strain sensor array and the driving speed limited for that road segment.

[0009] Preferably, the zero-phase-shift bandpass filter employs a zero-phase-shift filtering algorithm, which performs forward and reverse filtering on the original dynamic response time series to offset the phase delay generated during the filtering process.

[0010] Preferably, the step of extracting the peak time node includes: combining the number of physical axes, locating an equal number of local maxima points in each transient elastic strain sequence as peak time nodes.

[0011] Preferably, the spatiotemporal self-consistency verification step includes: within a single observation section, verifying whether the time interval ratio between the peak time nodes is consistent with the physical axis distance ratio of the physical axis distance sequence within a preset tolerance range.

[0012] Preferably, the step of performing constrained dynamic time warping includes: dividing the warping process into multiple independent optimization sub-intervals based on the set of rigid anchor points, and calculating local warping paths in each optimization sub-interval.

[0013] Preferably, for the data segments in the transient elastic strain sequence that are located before the first rigid anchor point and after the last rigid anchor point, linear time axis scaling is used for alignment.

[0014] Preferably, the step of determining the effective integration interval includes: extracting the peak time nodes corresponding to the first axle and the last axle, and adding a pre-time margin and a post-time margin to set the start and end times of the effective integration interval.

[0015] Preferably, the step of calculating the dynamic strain transfer function includes: taking the absolute values ​​of the transient elastic strain sequence of the upstream observation section and the dynamic response time sequence of the downstream observation section as a reference, and integrating them within the effective integration interval to obtain the absolute value integral; dividing the absolute value integral of the downstream observation section by the absolute value integral of the upstream observation section to obtain the dynamic strain transfer function.

[0016] Preferably, before performing the division, the method further includes: checking whether the absolute value integral of the upstream observation section is greater than a preset minimum energy threshold.

[0017] Preferably, the health benchmark includes a health benchmark mean vector and a health benchmark covariance matrix established based on historical health sample data.

[0018] Preferably, the health baseline covariance matrix is ​​regularized to maintain numerical stability during the inversion operation.

[0019] Preferably, the step of the determination logic outputting the trigger instruction includes: storing the calculated Mahalanobis distance in a fixed-length first-in-first-out queue.

[0020] Preferably, a trigger command is output when all Mahalanobis distances in the first-in-first-out queue are greater than a preset control upper limit threshold.

[0021] Preferably, the upper limit threshold is determined based on the chi-square distribution critical value at a preset confidence level.

[0022] This invention provides a method for monitoring and early warning of bridge dynamic response. It has the following beneficial effects: 1. This invention identifies a single heavy-duty vehicle as a mobile excitation source and uses its physical wheelbase sequence to verify the spatiotemporal consistency of the extracted peak time nodes, establishing a data filtering mechanism based on physical constraints. This effectively interferes with the signal, ensuring a high degree of consistency between the data entering subsequent analysis and the mobile excitation source, thereby improving the reliability of monitoring data from the source.

[0023] 2. This invention uses the self-consistency verified peak time node as a rigid anchor point to impose boundary constraints on the dynamic time warping operation, thus solving the problem of nonlinear phase distortion of the cross-sectional response signal caused by the vehicle's speed change while traveling on the bridge deck. This ensures that the signal alignment on the time axis conforms to physical laws, enabling the subsequently calculated dynamic strain transfer function to more accurately reflect the true relative stiffness relationship between the structural sections.

[0024] 3. This invention evaluates the longitudinal spatial transmission state vector, which characterizes the overall state of the structure, using Mahalanobis distance. This Mahalanobis distance considers the correlation between the dimensions of the longitudinal spatial transmission state vector, effectively suppressing the interference of normal data fluctuations caused by different vehicle loads on the evaluation results. Simultaneously, combined with the continuous over-limit judgment logic of the first-in-first-out queue, a warning is only triggered after multiple consecutive events indicate anomalies, avoiding misjudgments caused by single accidental factors and significantly improving the robustness and accuracy of the warning decision. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the early warning system architecture of the present invention; Figure 2 This is a flowchart of the bridge dynamic response monitoring and early warning method of the present invention; Figure 3 This is a timing diagram of event triggering and spatial feature extraction in this invention; Figure 4 This is a schematic diagram of the data acquisition and decoupling filtering principle of the present invention; Figure 5 This is a schematic diagram illustrating the principle of peak topology extraction and self-consistency verification in this invention. Figure 6 This is a schematic diagram illustrating the constrained dynamic time warping operation principle of the present invention. Figure 7 This is a schematic diagram of the principle of dynamic truncation of the integral domain and construction of the state vector in this invention; Figure 8 This is a flowchart of the distance feature calculation and continuous judgment early warning process of the present invention; Figure 9 This is a schematic diagram illustrating the effect of vehicle speed change on the phase of the cross-section signal according to the present invention; Figure 10 This is a diagram illustrating the constrained dynamic time warping alignment effect of the present invention. Figure 11 The diagram shows a comparison of the early warning effects of the present invention, wherein (a) is a schematic diagram of the comparison effect of the method of the present invention, and (b) is a schematic diagram of the comparison effect of the traditional method.

[0026] Among them, 10 is the weighing module; 20 is the sensing module; 30 is the edge gateway; and 40 is the alarm module. Detailed Implementation

[0027] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example 1

[0028] See attached document Figure 1 This invention provides an early warning system for implementing the bridge dynamic response monitoring and early warning method, the early warning system comprising: Weighing module 10 is deployed on the surface of the main lane at the bridge entrance to collect data on the load of passing vehicles and the physical spatial geometry of the vehicles. The sensing module 20 is used to be installed sequentially along the longitudinal direction of the bridge at designated observation sections of the bridge's load-bearing structure to collect the original dynamic response time series of each observation section. Edge gateway 30 is used to connect to weighing module 10 and sensing module 20 through communication network respectively to obtain real-time traffic data and structural response data, and to perform data alignment, state calculation and early warning judgment logic; The alarm module 40 is used to execute the indicated action according to the trigger command output by the edge gateway 30.

[0029] See attached document Figure 2 This invention provides a method for monitoring and early warning of bridge dynamic response, which includes the following steps: Step S10: The edge gateway 30 receives the data packet uploaded by the weighing module 10, determines whether a single heavy-load vehicle has entered the bridge based on the weight threshold, and establishes a trigger time window if the determination condition is met, and extracts the physical axle number and physical axle distance sequence. Step S20: The edge gateway 30 controls the sensing module 20 to synchronously acquire the original dynamic response time series of adjacent observation sections within the trigger time window, calls the zero phase shift bandpass filter to filter the acquired data, removes temperature drift and high-frequency noise signals, and outputs the transient elastic strain sequence. Step S30: The edge gateway 30 extracts the peak time nodes of the axle excitation corresponding to each observation section in the transient elastic strain sequence, and verifies the self-consistency of the time distribution of the peak time nodes within a single observation section based on the physical axle spacing sequence. After the verification is passed, the coordinate pairs of the associated peak time nodes are converted into a set of rigid anchor points. Step S40: The edge gateway 30 sets boundary constraints between the transient elastic strain sequences of adjacent observation sections based on the set of rigid anchor points, performs constrained dynamic time warping operation, eliminates cross-section phase distortion, and outputs dynamic response time sequence. Step S50: The edge gateway 30 dynamically determines the effective integration interval using the aligned peak time nodes, calculates the dynamic strain transfer function between adjacent observation sections within the effective integration interval, and combines the calculation results of each adjacent observation section into a longitudinal spatial transfer state vector. In step S60, the edge gateway 30 obtains the historical baseline probability distribution parameters, calculates the Mahalanobis distance between the longitudinal spatial transmission state vector and the health baseline, and stores it in the first-in-first-out queue. When the Mahalanobis distance values ​​of multiple consecutive events in the first-in-first-out queue meet the continuity limit condition, it is determined that the bridge has structural stiffness decay, and a trigger command is sent to the alarm module 40.

[0030] To further clarify the implementation of each technical aspect of the present invention, the following will provide a detailed description of the implementation of each functional module involved above and its internal processing flow.

[0031] See attached document Figure 3 , Figure 3 This is a timing diagram of event triggering and spatial feature extraction according to an embodiment of the present invention. Specifically, step S10 is executed by the edge gateway 30 in conjunction with the weighing module 10.

[0032] The edge gateway 30 continuously receives data packets broadcast by the weighing module 10 through a communication interface. In this embodiment, the data packets contain real-time load and geometric information of vehicles entering the main lane of the bridge. To avoid structural response interference caused by the superposition of excitation signals from multiple vehicles, the edge gateway 30 needs to filter out independent moving excitation sources from the continuous traffic flow. Therefore, step S10 may include the following sub-steps: S101, Edge Gateway 30 performs a validity determination of heavy-load vehicles. Specifically, Edge Gateway 30 parses the data packet to obtain the total vehicle weight W of the current vehicle. total and the total weight of the vehicle W total Compared with the preset weight threshold W th A comparison is performed. As a preferred method, this weight threshold W... th The value can be set based on the standard weight of large trucks as defined in relevant traffic regulations or the design live load standard of the corresponding bridge, for example, a value of 30 tons. When W is met... total >W th At this time, the edge gateway 30 determines that the current vehicle meets the heavy load condition and classifies it as a heavy load vehicle. To ensure the spatial uniqueness of the mobile excitation source, after confirming the heavy load condition, an exclusivity test needs to be performed.

[0033] Edge gateway 30 checks the time buffer ΔT before the time T0 when the heavy-duty vehicle enters the measurement point. pre and the subsequent time buffer band ΔT postWithin the test section, the weighing module 10 records data from other vehicles. Typically, the value of the front and rear time buffer zone is calculated based on the ratio of the total longitudinal span of the sensing module 20 (e.g., a strain sensor array) to the minimum driving speed limit for that section, ensuring that no other vehicles enter the test section before or after the heavy-duty vehicle. If the time interval [T0-ΔT]... post , T0+ΔT post Since there are no other vehicle passage records within the area, the edge gateway 30 determines that the heavy-duty vehicle is a single heavy-duty vehicle without interference.

[0034] After the exclusivity test is passed, the edge gateway 30 establishes a trigger time window T. w The trigger time window T w The start time is the measurement point time T0, and the trigger time window T w The cutoff time is calculated based on the latest estimated time when the vehicle leaves the end sensor, thus covering the complete cycle of the single heavy-duty vehicle from the bridge entrance to its departure from the test section, and is used to activate the data acquisition actions of subsequent nodes. If a data packet is missing or the exclusivity check fails, the edge gateway 30 discards the current data packet and resets the listening state to avoid entering invalid calculation logic.

[0035] S102, the edge gateway 30 performs the extraction of physical space geometric data. The edge gateway 30 parses the physical axle number K of a single heavy-duty vehicle from the data message of the effectively triggered weighing module 10, where K≥2.

[0036] Simultaneously, the edge gateway 30 extracts the physical wheelbase between adjacent axles of a single heavy-duty vehicle, generating a physical wheelbase sequence. This physical wheelbase sequence is defined as set L, and the mathematical expression for set L is: L={L1,L2,…L m ,…,L K-1}; In the formula, L m This represents the physical wheelbase between the m-th and (m+1)-th axles of the target vehicle, where m is a positive integer in the range [1, K-1].

[0037] Furthermore, vehicles often accelerate or decelerate while driving on bridges, causing nonlinear tensile or compressive distortions in their dynamic strain response across various observation sections over time. However, the inherent physical wheelbase spatial structure of the vehicle chassis is rigid and constant. This step extracts this physical wheelbase sequence, essentially providing a physical reference scale that does not deform with vehicle speed for subsequent irregular transient elastic strain sequences.

[0038] Edge gateway 30 caches the physical wheelbase sequence in local memory as prior parameters for subsequent self-consistency verification and phase alignment boundary constraints. Regarding the underlying data calculation process by which the weighing module 10 uses piezoelectric crystals or bending plate sensors to obtain the vehicle's total weight and the physical wheelbase between axles, those skilled in the art can refer to the operating specifications of existing road traffic monitoring equipment. The mechanism for acquiring these physical parameters is well-known in the field and will not be elaborated upon here.

[0039] See attached document Figure 4 , Figure 4 This is a schematic diagram of data acquisition and decoupling filtering according to an embodiment of the present invention. Specifically, step S20 is executed by the edge gateway 30 in conjunction with the sensing module 20.

[0040] After establishing the triggering time window for effective excitation events, the system needs to acquire dynamic response data reflecting the bridge structure under moving loads. In actual monitoring environments, there are often slow-change temperature interferences and high-frequency electromagnetic noise from equipment. Directly extracting features from the original sampled signals may lead to deviations in subsequent spatiotemporal calculations. Therefore, step S20 may include the following sub-steps: S201, the edge gateway 30 sends a synchronization acquisition command to the sensing modules 20 distributed longitudinally along the bridge within the trigger time window. In this embodiment, to ensure that the dynamic response data of different observation sections have a strict comparison benchmark on the time axis, the sensing modules of each observation section built into the sensing module 20 use the system global clock for timing and synchronization triggering. For adjacent observation sections longitudinally along the bridge, taking observation section A and observation section B as examples, the sensing module 20 uses a set sampling frequency f s Data acquisition is performed to generate the original dynamic response time series S corresponding to observation section A. A,raw (t) and the original dynamic response time series S of the corresponding section B. B,raw (t).

[0041] To prevent sequence misalignment across cross-sections due to packet loss in the communication link, the edge gateway 30 will collect the S... A,raw (t) and S B,raw (t) Perform data integrity and length checks. If the lengths of the two sets of original dynamic response time series are inconsistent or there are null value breakpoints, discard the data within the trigger time window and terminate the current calculation. As a preferred method, set the sampling frequency f. s It is necessary to cover the frequency band of the structural response caused by vehicle excitation. Typically, f... s The frequency is set between 100Hz and 500Hz to ensure that the Nyquist sampling theorem condition is met, so that the strain wave peak morphology characteristics generated by axle rolling can be restored in the time domain.

[0042] S202, the edge gateway 30 invokes a zero-phase-shift bandpass filter to perform frequency domain filtering on the acquired raw dynamic response time series. The raw strain signal output by the bridge structure is the result of the superposition of multiple physical field responses, mainly including static drift caused by temperature changes, transient elastic strain caused by vehicle rolling, and high-frequency noise interference caused by the external electromagnetic environment. The zero-phase-shift bandpass filter is used for bandpass filtering primarily to remove low-frequency and high-frequency interference terms.

[0043] The edge gateway 30 sets the lower cutoff frequency f of the zero-phase-shift bandpass filter based on the structural dynamic characteristics of the bridge. low With upper limit cutoff frequency f high Lower cutoff frequency f low Used to block low-frequency drift components caused by changes in ambient temperature gradients, its value is typically set to 0.1Hz to 0.5Hz; upper cutoff frequency f high It is used to filter out high-frequency electromagnetic noise and high-frequency local flutter interference, and its value is generally set to 20Hz to 50Hz.

[0044] Considering that conventional digital filters introduce phase shifts during filtering, causing signal peaks to shift on the time axis and affecting the accuracy of subsequent time node extraction, this step employs a zero-phase-shift filtering algorithm for bandpass filtering. Specifically, the edge gateway 30 passes the original dynamic response time sequence forward through a preset zero-phase-shift bandpass filter, then reverses the time of the resulting intermediate sequence, passes it backward through the same zero-phase-shift bandpass filter, and finally reverses the time of the second filtering result again. This dual forward and reverse filtering mechanism completely cancels out the phase delay caused by filtering.

[0045] After bandpass filtering, the edge gateway 30 outputs a transient elastic strain sequence with environmental coupling interference removed. Corresponding to the aforementioned observation sections A and B, the filtered transient elastic strain sequences are denoted as S0 and S1, respectively. A '(t) and S B The transient elastic strain sequence obtained at this time is in a zero-mean distribution state, which characterizes the elastic dynamic response of the bridge structure under the action of heavy vehicles.

[0046] For the deformation photoelectric signal conversion mechanism of the sensing module 20 based on fiber Bragg gratings or resistance strain gauges, and the specific differential equation derivation of the zero phase shift filtering algorithm, those skilled in the art can consult technical manuals in the field of structural health monitoring and digital signal processing. Its underlying signal conversion and filtering implementation logic is a well-known technology in this field, and will not be elaborated here.

[0047] See attached document Figure 5 , Figure 5This is a schematic diagram illustrating the principle of peak topology extraction and self-consistency verification according to an embodiment of the present invention. Specifically, step S30 is executed by the edge gateway 30.

[0048] After acquiring the transient elastic strain sequence after filtering out environmental interference, the system needs to extract physical features that characterize the dynamic response of a single heavy-duty vehicle. In complex traffic environments, sensor signals may be mixed with interference from light vehicles in adjacent lanes or secondary vibrations caused by uneven road surfaces. The edge gateway 30 screens for abnormal signals in both time and space dimensions to prevent misaligned or contaminated data from entering the subsequent evaluation process. Therefore, step S30 may include the following sub-steps: S301, the edge gateway 30 locates the peak time node of the excitation response of each axle in the transient elastic strain sequence. For the transient elastic strain sequence S of the observation section A within the same trigger time window... A '(t) and the transient elastic strain sequence S of the observation section B B At step S10, edge gateway 30 performs a local maximum search. To avoid mistaking minute fluctuations in the signal for axle excitation peaks, edge gateway 30 combines the number of physical axles K of a single heavy-duty vehicle obtained in step S10, sets a sliding time window in the transient elastic strain sequence, and extracts the K local maxima with the largest values.

[0049] As a preferred approach, the length of the sliding time window is determined based on the ratio of the minimum physical wheelbase of a single heavy-duty vehicle to the maximum speed limit of the target road segment. Properly setting the sliding time window length can avoid missing adjacent dual-axle structural peaks and prevent misjudging double burrs generated by the excitation of the same axle as two independent axles.

[0050] Edge gateway 30 arranges the extracted local maxima points in chronological order to obtain the peak time node sequence P generated from the peak time nodes of observation section A. A And the peak time node sequence P generated from the peak time node of the observation section B. B The formula is expressed as: P A ={t A,1 ,t A,2 ,...,t A,K}; P B ={t B,1 ,t B,2 ,...,t B,K}; In the formula, t A,k This represents the time coordinate of the strain peak occurring in the transient elastic strain sequence of the observed section A, corresponding to the compression of the k-th axle by a single heavy-duty vehicle, i.e., P. AThe k-th peak time node, t B,K P represents the time coordinate of the strain peak generated by the k-th axle rolling in the transient elastic strain sequence of the observed section B. B The k-th peak time node is defined, where k ranges from [1, K] and is a positive integer. Within the set sliding time window, if the number of extracted local maxima is not equal to K, or the time interval between any two adjacent extracted local maxima is less than the set sensor sampling period, the edge gateway 30 determines that the signal within the sliding time window is distorted or aliased, discards the event data, and terminates the current calculation process.

[0051] S302, the edge gateway 30 performs spatiotemporal self-consistency verification of a single observation section based on the physical wheelbase sequence. After extracting the peak time nodes, the system needs to verify whether these time features correspond to a single heavy-duty vehicle. Considering that the time a single heavy-duty vehicle spends passing through a single sensor section is relatively short, the driving state of the single heavy-duty vehicle within this time period can be approximated as uniform speed. Based on this feature, the time interval ratio of each peak time node measured within the same observation section should correspond to the physical wheelbase ratio of the vehicle's physical wheelbase sequence.

[0052] The edge gateway 30 uses the physical wheelbase sequence L={L1,L2,...,L...} obtained by the weighing module 10. K-1 The verification is performed independently within observation sections A and B, respectively. The constraint formulas for its spatiotemporal self-consistency verification are as follows: ; ; In the formula, m is a positive integer in the range [1, K-1]; L m t is the physical axis distance between the m-th axis and the (m+1)-th axis; A,m+1 -t A,m t represents the time interval between the peak times of the m-th axis and the (m+1)-th axis on observation section A; B,m+1 -t B,m t represents the time interval between the peak times of the m-th axis and the (m+1)-th axis on the observation section B; B,K -t B,1 ε represents the time interval between the peak times of the first and last axles in observation section B; t The time measurement tolerance constant set for the system. To represent the total wheelbase of a single heavy-duty vehicle from the first axle to the last axle, t A,K -t A,1 This represents the total time span between the peak time nodes of the first and last axles on observation section A, where K is the number of physical axles of a single heavy-duty vehicle.

[0053] In this embodiment, by using the total wheelbase and total time span of a single heavy-duty vehicle as the normalization benchmark, rather than the wheelbase of the first segment, the benchmark instability caused by excessively short or long local wheelbases of a single heavy-duty vehicle can be effectively avoided. This constrains the proportional relationship between the time intervals of all peak time node sequences and the physical wheelbases of the physical wheelbase sequences within a more stable range, effectively establishing a comparison benchmark for the extracted signal nodes. The time measurement tolerance constant ε t The sampling frequency accuracy of sensor module 20 is determined based on a combination of the estimated maximum local vehicle acceleration fluctuation of the target road segment, typically set within the range of 0.05 to 0.15 to accommodate errors caused by sampling discretization. If any proportional difference calculated from observation section A or observation section B is greater than ε... t This indicates that the peak time node sequence failed to reflect the physical structure of a single heavy-duty vehicle, and the edge gateway 30 determined that the event data was invalid and discarded.

[0054] S303, after successful verification, edge gateway 30 establishes a set of rigid anchor points. When the spatiotemporal self-consistency verification of both observation sections meets the constraints, edge gateway 30 confirms that the extracted peak time node sequence matches the corresponding single heavy-duty vehicle feature. Edge gateway 30 combines the peak time nodes corresponding to the same axle in the two observation sections to form a two-dimensional coordinate pair, constructing a set of rigid anchor points Anchor containing K elements: Anchor={(t A,1 ,t B,1 ),(t A,2 ,t B,2 ),…,(t A,K ,t B,K )}; The coordinate pairs (t) in this set of rigid anchor points A,K ,t B,K The value represents the temporal correlation of the extreme values ​​of the structural response induced by the corresponding physical axle at different spatial cross sections. The edge gateway 30 uses this set of rigid anchor points as the boundary nodes to be penetrated in the subsequent sequence feature alignment algorithm, providing a spatial reference for cross-section signals.

[0055] For algorithms that use a sliding time window to search for local maxima in transient elastic strain sequences, those skilled in the art can use conventional peak detection processing functions, the basic logic of which is well-known in the field and will not be elaborated here.

[0056] See attached document Figure 6 , Figure 6 This is a schematic diagram of a constrained dynamic time warping operation according to an embodiment of the present invention. Specifically, step S40 is executed by the edge gateway 30.

[0057] In real-world traffic scenarios, heavy-duty vehicles often accelerate, decelerate, or brake while traveling on a bridge. This macroscopic speed nonstationarity causes the dynamic strain response generated by vehicle excitation at the downstream observation section to undergo nonlinear time stretching or compression relative to the upstream observation section. Directly using the traditional time translation method for cross-section alignment makes it difficult to address this nonlinear phase distortion, leading to systematic errors in subsequent state calculations. Introducing a dynamic time warping algorithm can address this problem through nonlinear bending mapping of the time axis. Conventional dynamic time warping, lacking physical constraints, is prone to mismatches that deviate from the actual structural response characteristics when searching for matching paths. In this embodiment, the edge gateway 30 performs constrained optimization calculations based on the previously extracted set of rigid anchor points; therefore, step S40 may include the following sub-steps: S401, Edge Gateway 30 performs spatial mesh generation on the transient elastic strain sequence based on the rigid anchor point set. Edge Gateway 30 acquires the transient elastic strain sequence S of observation section A. A '(t) and the transient elastic strain sequence S of the observation section B B '(t), and retrieve the set of rigid anchor points Anchor={(t) containing K coordinate pairs. A,1 ,t B,1 ),...,(t A,K ,t B,K The edge gateway 30 divides the main space of the regularization operation into K-1 independent optimization sub-intervals. The boundary of each optimization sub-interval is defined by two adjacent rigid anchor points (t). A,K ,t B,K ) and (t A,k+1 ,t B,k+1 This partitioning mechanism forces the mapping path to pass through the time anchor point representing the physical axle excitation state, transforming the engineering physical correspondence into the optimization boundary constraint of the algorithm.

[0058] For data segments within the sliding time window that fall before the first rigid anchor point and after the last rigid anchor point, the system does not perform nonlinear normalization. The edge gateway 30 uses the beginning and end time spans of observation section A as a reference and performs proportional time axis scaling mapping on the corresponding edge data segments of observation section B using linear interpolation to preserve the structural response energy during the entry and free decay phases.

[0059] S402, the edge gateway 30 calculates the cumulative distance matrix and performs local path optimization in each optimization sub-interval. In the k-th optimization sub-interval, the edge gateway 30 establishes the local distance matrix. For the transient elastic strain sequence S... A The time index i and transient elastic strain sequence S in '(t) BIn time index j in '(t), edge gateway 30 calculates the Euclidean distance between corresponding points of two time indices as the local regularization cost d(i,j), denoted as d(i,j)=|S A '(t)- S B '(t)|, all d(i,j) form a local distance matrix.

[0060] To initiate dynamic programming iteration, edge gateway 30 initializes the cumulative distance to the starting point of the optimization sub-interval, setting D... k (t A,k ,t B,k )=d(t A,k ,t B,k The edge gateway 30 updates the cumulative distance D based on the local regularization cost and the initial state, and assigns infinity to other invalid meshes corresponding to non-starting boundary boundaries. k (i,j): D k (i,j)=d(i,j)+min(D k (i-1,j),D k (i,j-1),D k (i-1,j-1)); In the formula, k takes the value of a positive integer in the range [1, K-1]; the range of i is limited to (t A,k ,t A,k+1 The range of values ​​for j is limited to (t). B,k ,t B,k+1 ]; min() represents the minimum value function; D k (i-1,j) represents the cumulative distance in the horizontal direction; D k (i,j-1) represents the cumulative distance in the vertical direction; D k (i-1, j-1) represents the cumulative distance along the diagonal, and all D k (i,j) form the cumulative distance matrix.

[0061] As a preferred approach, by restricting the values ​​of i and j, the calculation of the cumulative distance is constrained within the local grid formed by adjacent rigid anchor points, avoiding the risk of the algorithm jumping to unrelated peak regions. After completing the cumulative distance update, the edge gateway 30 starts from the rigid anchor point (t) A,k+1 ,t B,k+1 To the rigid anchor point (t) A,k ,t B,k Perform backtracking to find the locally regularized path that minimizes the total cumulative cost.

[0062] S403, the edge gateway 30 splices locally regularized paths to generate a cross-sectional dynamic response time series. The edge gateway 30 splices the linearly mapped beginning and end edge data paths with the optimization backtracking paths of K-1 optimization sub-intervals in chronological order, forming a global mapping path that runs through the entire excitation response process. Based on this global mapping path, the edge gateway 30 generates the transient elastic strain sequence S of the observed section B. B '(t) is mapped onto the time axis of the observation section A for reconstruction.

[0063] The reconstruction process uses the time series of the observed section A as a unified benchmark, and then reconstructs S. B The nonlinear phase distortion in '(t) is compensated, and the output is a dynamic response time series aligned with the observation section A in terms of time nodes and phase. This dynamic response time series reduces the signal misalignment caused by vehicle speed changes, making the response waveforms of adjacent observation sections comparable on the same time scale, thus providing a basis for calculating the dynamic strain transfer function.

[0064] For the underlying computational mechanism of finding the minimum cost backtracking path in the cumulative distance matrix using dynamic programming, those skilled in the art can refer to the conventional dynamic time warping optimization theory in discrete time series analysis. Its basic optimization method is a well-known technology in this field and will not be elaborated here.

[0065] See attached document Figure 7 , Figure 7 This is a schematic diagram illustrating the principle of dynamic truncation of the integral domain and construction of state vectors according to an embodiment of the present invention. Specifically, step S50 is executed by the edge gateway 30.

[0066] After completing the nonlinear phase alignment of the cross-section signal, the system needs to quantify the relative dynamic response relationship between adjacent observation sections when excited by the same moving load. Since the original trigger signal within the entire time window contains redundant data and background noise before the vehicle enters the core influence line section of the bridge, directly extracting features from the dynamic response time series within the entire time window may reduce the signal-to-noise ratio of the core excitation response. Therefore, the edge gateway 30 dynamically defines the effective data interval containing the main dynamic energy and constructs parameters that reflect the overall longitudinal relative stiffness distribution characteristics of the bridge based on this. Thus, step S50 may include the following sub-steps: S501, Edge Gateway 30 performs adaptive truncation of the effective integration interval. The excitation energy of the vehicle on the bridge structure is mainly concentrated in the decay period from when the front of the vehicle passes over the measuring point to when the rear of the vehicle leaves the measuring point. Edge Gateway 30 acquires the transient elastic strain sequence of the observation section A generated in the previous step, and extracts the first peak time node t corresponding to the first axle. A,1 and the Kth peak time node t corresponding to the last axle A,K.

[0067] Based on the above peak time node t A,1 and t A,K The edge gateway 30 sets the start time t of the effective integration interval. start With the deadline t end The calculation formula is as follows: t start =max(t A,1 -Δt pre ,t win_start ); t end =min(t A,K +Δt post ,t win_end ); In the formula, Δt pre This represents the lead time margin, used to cover the initial bending pre-strain generated when the first axle of a single heavy-duty vehicle reaches the front axle; Δt post This represents the post-time margin, used to include the residual free-decaying oscillation energy of the structure after the vehicle leaves the measuring point; t win_start and t win_end These represent the start and end times of the aforementioned trigger time window, respectively; the max() and min() functions are used to provide boundary overrun protection to prevent memory access errors caused by the truncation time exceeding the actual boundary of the sampling array.

[0068] As a preferred approach, the lead time margin Δt pre With post-time margin Δt post The value of Δt is determined based on a combination of the bridge span and the length of the local stress influence line of the sensor. For medium-span bridges, Δt is typically set. pre Set Δt between 0.5 seconds and 1.0 seconds. post The interval is between 1.5 seconds and 3.0 seconds to ensure that the integral domain can fully cover the main working cycle of the structure.

[0069] S502, Edge Gateway 30 calculates the dynamic strain transfer function. After establishing the effective integration interval with a high signal-to-noise ratio, Edge Gateway 30 calculates the dynamic strain transfer function within this effective integration interval [t]. start ,t end Within [the observation section A], the transient elastic strain sequence S A The dynamic response time series of '(t) and the phase-aligned observation section B Summation is achieved by performing absolute value integrals separately.

[0070] Furthermore, based on physical principles, structural strain can be positive or negative, and direct integration may fail to reflect the true total deformation due to the cancellation of positive and negative values. By taking the absolute value of the signal and then performing time integration, it is equivalent to accumulating the deformation effects in all directions. The resulting absolute value integral is physically proportional to the total elastic potential energy absorbed and released by the observed section during the excitation event.

[0071] Edge gateway 30 calculates the dynamic strain transfer function H between two observation sections. AB The calculation formula is as follows: ; In the formula, t is the discrete-time index; The amplitude of the dynamic response time series of the phase-aligned observation section B at time t; Let t be the amplitude of the transient elastic strain sequence at time t, which serves as the reference observation section A. start Let t be the starting time of the effective integration interval. end This is the cutoff time for the effective integration interval.

[0072] In this embodiment, due to the dynamic response time series The time distortion caused by macroscopic vehicle speed non-stationarity has been eliminated, and the energy ratio H AB Ideally, this mainly reflects the ratio of the local relative bending stiffness of the structure at observation section B to that at observation section A. If a minor damage occurs at observation section B, leading to a decrease in stiffness, its local strain will be amplified, thereby causing H... AB The numerical value increases. To prevent sensor failure from causing the denominator to be calculated as zero, thus triggering a program exception, the edge gateway 30 verifies the integral value of the denominator before performing the division. As a preferred method, the minimum energy threshold for this verification can be determined based on the background noise integral value of the sensing module 20 when no vehicle is passing by, for example, taking 5 to 10 times the background noise integral value to ensure that the signal used for calculation has a sufficient signal-to-noise ratio.

[0073] In S503, edge gateway 30 generates the longitudinal spatial transfer state vector. In engineering practice, bridges typically have N sequentially adjacent observation sections longitudinally, where N ≥ 3. Based on this method, edge gateway 30 traverses all observation section pairs, and according to the spatial topology order from upstream to downstream, that is, the role positioning of observation section A with the nth observation section as the reference and observation section B with the (n+1)th observation section as the downstream, it repeatedly executes the calculation logic from S501 to S502, thereby sequentially calculating H. 1,2 H 2,3 Until H N-1,N This yields N-1 local dynamic strain transfer functions.

[0074] Edge gateway 30 combines the dispersed dynamic strain transfer functions into a one-dimensional array to construct a longitudinal spatial transfer state vector V that reflects the overall spatial stiffness gradient of the bridge under the current excitation event: V=[H 1,2 H 2,3 ,...,H n,n+1 ,...,H N-1,N ]; In the formula, H n,n+1 V represents the dynamic strain transfer function between the nth and (n+1)th observation sections, where n is a positive integer in the range [1, N-1], and N is the total number of observation sections arranged along the longitudinal direction of the bridge. This longitudinal spatial transfer state vector V couples the dynamic characteristics of each isolated observation section into a continuous topological dimension, providing a standardized data benchmark for subsequent multidimensional probability distribution assessment and macroscopic early warning decision-making.

[0075] For the software code implementation of absolute value summation and basic array concatenation operations for discrete time series, those skilled in the art can use conventional high-level programming language array processing functions. The underlying data operation methods are well-known technologies in this field and will not be elaborated here.

[0076] See attached document Figure 8 , Figure 8 This is a flowchart of distance feature calculation and continuous early warning determination according to an embodiment of the present invention. Specifically, step S60 is executed by the edge gateway 30 in conjunction with the alarm module 40.

[0077] After obtaining the longitudinal spatial transfer state vector characterizing the longitudinal spatial stiffness gradient of the bridge, the system needs to quantify the degree of deviation between the current state and the initial health baseline of the structure. Considering that a single excitation event is easily affected by accidental factors such as abnormal vehicle trajectories or instantaneous road debris, the results of a single evaluation are prone to fluctuation. To filter out such non-structural disturbances, the system uses a logic of multi-dimensional spatial distance measurement combined with time series statistics for verification. Therefore, step S60 may include the following sub-steps: S601, the edge gateway 30 calculates the Mahalanobis distance between the current longitudinal spatial transmission state vector and the health state statistical distribution. The edge gateway 30 retrieves a pre-calibrated health baseline parameter set from local memory, which includes the health baseline mean vector μ. H Covariance matrix with health baseline .

[0078] In one embodiment, the process of establishing the health baseline mean vector and the health baseline covariance matrix is ​​as follows: When the bridge is confirmed to be in good working order during the initial operational phase, the system continuously collects and extracts Q valid historical longitudinal spatial transmission state vectors corresponding to single heavy-load vehicle passage events, forming a historical sample set {V1, V2, ..., V...}.Q To ensure the full-rank property of the subsequent health baseline covariance matrix, the total amount Q of the collected historical sample set must be strictly greater than the dimension N-1 of the historical longitudinal spatially transmitted state vector. Based on this historical sample set, the health baseline mean vector μ H According to the formula The calculated health baseline covariance matrix is ​​obtained using the unbiased estimation formula. The calculation shows that q is the sample index, and V is... q The state vector is passed to the historical longitudinal space corresponding to the q-th historical sample. This is to prevent multicollinearity among historical samples. The singularity is irreversible; the system performs regularization correction on the calculated health baseline covariance matrix, i.e. Where I is the same as Identity matrices of the same order A very small positive constant (e.g., 10⁻⁶) is set for the system to ensure the numerical stability of subsequent matrix inversion operations. After the baseline parameters are retrieved, the edge gateway 30 substitutes the newly generated longitudinal spatial transfer state vector V into the distance formula for calculation: ; In the formula, D M The Mahalanobis distance is the calculated value; V is the longitudinal spatial propagation state vector corresponding to the current event; μ H The aforementioned mean vector of health benchmarks; is the inverse of the health baseline covariance matrix after regularization correction; T denotes matrix transpose.

[0079] In this embodiment, since the axle load and wheelbase distribution of heavy vehicles are discrete, the simple spatial stiffness index often exhibits wide fluctuations. The Mahalanobis distance introduces the inverse matrix of the healthy baseline covariance matrix as a weight, which can automatically absorb and offset the normal physical fluctuations caused by different vehicle total weight distributions, so that the calculation results purely reflect the abnormal offset of the local stiffness of the structure.

[0080] S602, the edge gateway 30 performs continuous limit violation determination based on a first-in-first-out (FIFO) queue. To reduce the false alarm rate during the anomaly detection process, the edge gateway 30 maintains a fixed-length FIFO queue of length M in memory. Each time a new Mahalanobis distance D is obtained... M Then, the edge gateway 30 pushes the Mahalanobis distance into the tail of the first-in-first-out queue; when the first-in-first-out queue reaches its storage limit, the first Mahalanobis distance stored is simultaneously removed.

[0081] After updating the FIFO queue, the edge gateway 30 performs a traversal comparison to check whether all M existing Mahalanobis distances in the queue exceed the set control upper limit threshold γ. As a preferred method, the control upper limit threshold γ is determined based on a chi-square distribution table. Since the square of the Mahalanobis distance under normal conditions approximately follows a chi-square distribution with N-1 degrees of freedom, the system sets a confidence level (usually 0.99 or 0.999), obtains the critical value of the chi-square distribution for the corresponding degrees of freedom by looking up the table, and takes the square root of this value to obtain the control upper limit threshold γ. The queue length M is typically a positive integer between 3 and 5. The system determines that the bridge has experienced substantial stiffness reduction if and only if all elements in the FIFO queue are greater than γ, i.e., if M consecutive independent heavy vehicle vibration events all point to structural anomalies.

[0082] S603, the edge gateway 30 generates a trigger command and links with the alarm module 40. When the confirmation condition of continuous over-limit is met, the edge gateway 30 immediately generates a linkage trigger command. On one hand, the edge gateway 30 encapsulates the warning time, abnormal section number, and Mahalanobis distance for trigger determination into an industrial communication message, which is transmitted to the remote central monitoring server via the network to provide data verification support. On the other hand, the edge gateway 30 triggers the inversion of the switch state of the local hardware interface, outputting an execution level to the alarm module 40.

[0083] After receiving the signal level, the alarm module 40 connects the associated audible and visual alarm device circuit, triggering the warning light to flash and the buzzer to sound, issuing a physical-level intervention signal to the on-site road administration or bridge maintenance personnel.

[0084] For the underlying programming methods for establishing the covariance matrix based on statistical principles and implementing first-in-first-out queue data updates in microcontrollers, those skilled in the art can refer to the basic literature on data structures and applied statistics. The conventional code implementation methods are well-known technologies in this field and will not be elaborated here.

[0085] To enable those skilled in the art to more clearly understand the purpose, technical solution, and advantages of this invention, the present invention will be further described in detail below with reference to specific application embodiments, real experimental test data, and corresponding drawings. It should be noted that the embodiments described herein are only for explaining the present invention and are not intended to limit the scope of protection of the present invention. Example 2

[0086] To more clearly illustrate the technical concept and implementation effects of this invention, the following will use a monitoring scenario of an actual bridge as an example to explain the entire process of this invention.

[0087] 1. Scene setting This embodiment applies to a three-span continuous prestressed concrete box girder bridge with a total length of 120 meters. To implement this invention, the system hardware deployment is as follows: Weighing Module 10: A dynamic axle load cell (WIM) is installed 50 meters upstream of the bridge entrance under the main lane surface to obtain the total weight of passing vehicles, the number of physical axles, and the sequence of physical axle distances.

[0088] Sensing Module 20: Along the longitudinal direction of the main girder of the bridge, five fiber Bragg grating (FBG) strain sensors are installed at the bottom of the web at key stress locations such as mid-span and quarter-span points. Their spatial coordinates are x1=25m, x2=40m, x3=55m, x4=70m, and x5=85m, respectively. The total number of observation sections is N=5.

[0089] Edge gateway 30 and alarm module 40: installed in a device box near the bridgehead and connected to each module via optical fiber.

[0090] 2. Process Execution At a certain moment, a six-axle heavy truck approaches the bridge, and the system executes the following steps: Step S10: Weighing module 10 measures the total weight W of the truck. total =45 tons, which is greater than the preset weight threshold W th =30 tons. Edge gateway 30 determines it to be a heavy-duty vehicle. By checking the 15-second time buffer before and after, it is confirmed that no other vehicles have passed, constituting a valid single-vehicle excitation event, and the heavy-duty vehicle is determined to be a single heavy-duty vehicle. Edge gateway 30 establishes a trigger time window and extracts the physical axle number K=6 and the physical wheelbase sequence L={1.5m,6.5m,1.3m,1.3m,1.3m} of the single heavy-duty vehicle from the data packet.

[0091] Step S20: Edge gateway 30 instructs 5 FBG sensors to use f s The original dynamic response time series was synchronously acquired at a sampling frequency of 200Hz. Subsequently, a zero-phase-shift Butterworth bandpass filter with a passband of [0.3Hz, 30Hz] was used to perform bandpass filtering on the five acquired original dynamic response time series, resulting in five transient elastic strain sequences S1'(t) to S5'(t).

[0092] Step S30: In each set of transient elastic strain sequences, the edge gateway 30 searches for and extracts the K (K=6) peaks with the largest amplitudes, obtaining the peak time node sequences P1 to P5. Then, taking the third observation section as an example, for P3={t 3,1 ,...,t 3,6 To verify its spatiotemporal self-consistency, calculate whether the time interval ratio of its peak time node sequence matches the physical axis spacing ratio of the physical axis spacing sequence L in terms of the time measure tolerance constant ε. t=0.08 matching. Verification showed that the peak time node sequences of all five observation sections passed the self-consistency test. Edge gateway 30 then randomly paired to establish four sets of rigid anchor points; for example, the rigid anchor point set between observation section 3 and observation section 4 is Anchor. 3,4 ={(t 3,1 ,t 4,1 ),...,(t 3,6 ,t 4,6 )}.

[0093] Step S40: The edge gateway 30 performs constrained dynamic time warping on the transient elastic strain sequence S4'(t) of the observation section 4, using the transient elastic strain sequence S3'(t) of the observation section 3 as a reference. Anchor is used. 3,4 The warping space is divided into K-1 (K-1=5) optimization sub-intervals. Dynamic time warping is performed independently in each optimization sub-interval. Finally, the sub-intervals are concatenated to form a global mapping path, and a dynamic response time series that is perfectly aligned with the phase of S3'(t) is reconstructed. This process is performed once for all adjacent observation sections (observation section 1 and observation section 2, observation section 2 and observation section 3, observation section 3 and observation section 4, observation section 4 and observation section 5).

[0094] Step S50: Taking observation section 3 and observation section 4 as examples, the edge gateway 30 is based on the rigid anchor point t 3,1 and t 3,6 And combined with the pre-set time margin Δt pre =0.8 seconds and post-time margin Δt post =2.0 seconds, dynamically determining the effective integration interval. Within this effective integration interval, the dynamic strain transfer function H is calculated. 3,4 By combining the dynamic strain transfer functions calculated from all N-1=4 adjacent observation sections, the longitudinal spatial transfer state vector V=[H] for this event is obtained. 1,2 H 2,3 H 3,4 H 4,5 The longitudinal spatial state vector is a 4-dimensional vector.

[0095] Step S60: Edge gateway 30 retrieves the stored health baseline mean vector μ H (4-dimensional) and covariance matrix (4x4 dimensional). Calculate the Mahalanobis distance D between the current longitudinal spatially transferred state vector V and the health baseline. M D M Store the data in a first-in, first-out (FIFO) queue of length M=4. After checking, the D values ​​of the four most recent events in the FIFO queue are... MNone of the values ​​exceeded the upper control threshold γ calculated based on the chi-square distribution (degrees of freedom N-1=4) and a 99.9% confidence level. The system determined that the bridge was in normal condition, did not issue an alarm, and continued monitoring for the next event.

[0096] To verify the effectiveness and advancement of the present invention in eliminating operational environment interference and improving the sensitivity of early damage identification, a bridge finite element simulation environment consistent with the scenario of the embodiment was constructed based on the "MATLABR2025a" software in the matrix laboratory, and comparative verification experiments were carried out.

[0097] Comparison settings: Experimental group: The technical solution disclosed in this invention is adopted. Specifically, spatiotemporal self-consistency verification and constrained dynamic time warping are performed through steps S30 to S40 to eliminate phase distortion caused by vehicle speed non-stationarity; and longitudinal spatial transfer state vector is constructed and Mahalanobis distance is calculated through steps S50 to S60 to eliminate the coupling effect of vehicle load difference and perform multidimensional statistical early warning.

[0098] Control group: A simplified traditional monitoring method was used. This method directly performs a rigid time shift on the signal at the downstream observation section (aligning only with the first peak time node), and then calculates the dynamic strain energy ratio within the entire time window as a damage index, without considering nonlinear phase distortion or using multidimensional correlation analysis.

[0099] The test results are as follows: According to the appendix Figure 9 It is evident that under the condition of acceleration by a single heavy-duty vehicle, the bridge structure response signal exhibits significant nonlinear phase distortion. The transient elastic strain sequence signals (solid line) at observation section 3 and (dashed line) at observation section 4 show a clear difference in morphology, with the peak time node spacing between the downstream and upstream observation sections being significantly compressed. This indicates that without effective phase alignment correction, directly comparing the transient elastic strain sequences at observation sections 3 and 4 will introduce a large calculation error caused by vehicle speed changes, which is not due to structural state changes.

[0100] According to the appendix Figure 10 It can be seen that the dynamic time warping algorithm used in the experimental group of this invention can perfectly correct the aforementioned phase distortion. After processing in step S40, the transient elastic strain sequence of the aligned observation section 4 corresponds one-to-one with and precisely coincides with the peaks of each physical axle in the transient elastic strain sequence (solid line) of the observation section 3 on the time axis. This proves that the method of this invention can effectively decouple the interference of vehicle speed non-stationarity on the structural response signal, providing a clean data foundation for subsequent calculation of accurate indicators that only reflect the relative stiffness characteristics of the structure.

[0101] See attached document Figure 11 The Mahalanobis distances calculated for sample points representing the healthy state (hollow circles) stably cluster within a low value range, while the Mahalanobis distances for sample points representing the 5% stiffness damage state (solid cubes) consistently jump to a significantly higher value range. There is a clear distinction between the two states, and the preset upper control threshold can distinguish between them 100% without any false alarms or false negatives. This demonstrates that the method of this invention has extremely high sensitivity to minor damage and very strong index stability.

[0102] Furthermore, according to the control group results, the calculated conventional energy ratios for both healthy and damaged states exhibited drastic and irregular fluctuations. The sample point distribution ranges for the two states largely overlapped, making it impossible to find a suitable upper control threshold to effectively distinguish them. If the upper control threshold is set too low, a large number of healthy samples will be misclassified as damaged; if the upper control threshold is set too high, all damaged samples will be missed. This demonstrates that conventional methods are highly susceptible to interference from operational environment noise and cannot reliably identify early structural performance degradation.

[0103] In summary, through direct comparison with traditional methods, and in conjunction with supplementary... Figures 9 to 11 The intuitive demonstration and experimental results fully demonstrate that the present invention has successfully solved the key technical problems in mobile load monitoring through multi-stage refined processing, and its final output trigger command has accuracy, stability and reliability that are unmatched by traditional methods.

[0104] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring and early warning of bridge dynamic response, characterized in that, Includes the following steps: Identify a single heavy-duty vehicle as a mobile excitation source and acquire its physical space geometric data. Establish a trigger time window based on the passage of the single heavy-duty vehicle and simultaneously collect the original dynamic response time series at multiple observation sections of the bridge distributed longitudinally within the trigger time window. The original dynamic response time sequence was bandpass filtered using a zero-phase-shift bandpass filter to remove low-frequency drift and high-frequency noise interference, thus obtaining the transient elastic strain sequence. Using the physical space geometric data as a physical reference, the transient elastic strain sequences of different observation sections are phase aligned to obtain the dynamic response time series; The phase alignment process specifically includes the following steps: Multiple peak time nodes were located in the transient elastic strain sequence of each observation section; Based on the physical axis distance sequence contained in the physical space geometric data, the time distribution of the wave peak time nodes within a single observation section is verified for self-consistency. If the verification is successful, the peak time nodes corresponding to the same axle in different observation sections will be combined to establish a set of rigid anchor points as phase alignment boundary constraints. Based on the set of rigid anchor points, the transient elastic strain sequence of adjacent observation sections is divided into multiple optimization sub-intervals; Perform dynamic time warping independently within each optimization sub-interval to generate locally warped paths; By splicing the local regularized paths, a dynamic response time series aligned with the phase of the reference observation section is reconstructed; Based on the dynamic response time series, the dynamic strain transfer function between adjacent observation sections is quantified and combined to generate a longitudinal spatial transfer state vector characterizing the overall longitudinal stiffness distribution of the bridge. The longitudinal spatial transfer state vector is subjected to a probabilistic distance measurement with the health benchmark representing the structural health status, and a trigger command is output based on whether the measurement result satisfies the continuity limit condition.

2. The bridge dynamic response monitoring and early warning method according to claim 1, characterized in that, The steps of identifying a single heavy-duty vehicle as a mobile excitation source and acquiring its physical spatial geometric data include: Determine if the total weight of the vehicle exceeds a weight threshold; Check whether there are records of other vehicles passing through within the time buffer zone before and after the moment the vehicle enters the measuring point; When the total weight of the vehicle is greater than the weight threshold and there are no other vehicles within the time buffer zone, the vehicle is determined to be a single heavy-duty vehicle, and the sequence of physical axles and physical wheelbase of the single heavy-duty vehicle is extracted as physical space geometric data.

3. The bridge dynamic response monitoring and early warning method according to claim 1, characterized in that, The step of quantifying the dynamic strain transfer function between adjacent observation sections includes: Based on the peak time nodes in the dynamic response time series, the effective integration interval for calculation is adaptively determined; Within the effective integration interval, the dynamic response time series of adjacent observation sections are integrated to obtain the dynamic strain transfer function.

4. The bridge dynamic response monitoring and early warning method according to claim 3, characterized in that, The steps for quantifying the dynamic strain transfer function between adjacent observation sections are as follows: Within the effective integration interval, the absolute integral of the transient elastic strain sequence of the upstream observation section, which serves as the reference, and the absolute integral of the dynamic response time sequence of the downstream observation section after phase alignment processing are calculated respectively. The ratio of the absolute integral of the downstream observation section to the absolute integral of the upstream observation section is used as the dynamic strain transfer function.

5. The bridge dynamic response monitoring and early warning method according to claim 1, characterized in that, The step of performing a probabilistic distance measurement between the longitudinal spatially transferred state vector and the health benchmark representing the structural health state includes: Obtain the health baseline, which includes the health baseline mean vector and the health baseline covariance matrix; Calculate the Mahalanobis distance between the longitudinal spatially transmitted state vector and the health baseline.

6. The bridge dynamic response monitoring and early warning method according to claim 5, characterized in that, The step of determining whether to output a trigger command based on whether the measurement result meets the continuity limit condition includes: Store the calculated Mahalanobis distance into a fixed-length first-in-first-out queue; When all Mahalanobis distance values ​​in the first-in-first-out queue are greater than the preset control upper limit threshold, a trigger command is output.

7. The bridge dynamic response monitoring and early warning method according to claim 5, characterized in that, The health benchmarks are established in the following ways: Under the condition of bridge health, collect and construct multiple historical longitudinal spatial transmission state vector samples; Based on the historical longitudinal spatial transmission state vector samples, the health baseline mean vector and health baseline covariance matrix are calculated.

Images