Bridge health monitoring method, device and readable medium based on multi-sensor fusion

Abstract

This invention discloses a bridge health monitoring method, equipment, and readable medium based on multi-sensor fusion, belonging to the field of bridge health monitoring technology. The method includes the following steps: S1, multi-source sensor deployment across the entire domain; S2, sensor fault self-diagnosis and self-calibration; S3, multi-source data preprocessing and fusion; S4, intelligent data calculation and real-time response; S5, structural assessment and damage diagnosis. This invention employs the aforementioned bridge health monitoring method, equipment, and readable medium based on multi-sensor fusion. Through multi-modal full-domain perception, a three-level full-link fusion algorithm, and a cloud-edge-device collaborative architecture, it addresses pain points such as limitations of single monitoring dimensions, insufficient heterogeneous data fusion, and poor real-time performance. It adapts to the all-scenario, all-weather operation and maintenance needs of various bridges, achieving high-precision, highly robust intelligent health monitoring and safety early warning.

Description

Technical Field

[0001] This invention belongs to the field of bridge health monitoring technology, specifically relating to a bridge health monitoring method, equipment, and readable medium based on multi-sensor fusion. Background Technology

[0002] As a crucial component of transportation infrastructure, bridges are susceptible to the effects of vehicle loads, environmental factors, material aging, and structural damage during their long-term service life. Their structural safety directly impacts the safety of public transportation operations. Traditional bridge inspection methods rely heavily on manual patrols and periodic checks, which suffer from low efficiency, high subjectivity, difficulty in providing real-time feedback on structural condition, and difficulty in detecting early-stage, hidden damage.

[0003] However, most existing bridge health monitoring technologies rely on single-type sensors for data acquisition, resulting in limited monitoring dimensions, susceptibility to environmental interference, and insufficient data reliability. While some employ multiple sensor types, the lack of a unified spatiotemporal benchmark and effective fusion mechanism for multi-source heterogeneous data leads to data silos, weak anti-interference capabilities, and a difficulty in balancing monitoring accuracy and robustness. Furthermore, existing systems primarily focus on data acquisition and simple display, lacking intelligent assessment, damage identification, and early warning decision-making capabilities. The centralized cloud processing model also suffers from high data transmission pressure and poor real-time performance, making it difficult to meet the long-term, stable, and high-precision health monitoring and maintenance needs of large bridges.

[0004] Therefore, a new method is urgently needed. Summary of the Invention

[0005] The purpose of this invention is to provide a bridge health monitoring method, equipment, and readable medium based on multi-sensor fusion. This method solves the pain points of single monitoring dimension limitations, insufficient heterogeneous data fusion, and poor real-time performance through multimodal full-domain perception, three-level full-link fusion algorithm, and cloud-edge-device collaborative architecture. It is suitable for the operation and maintenance needs of various bridges in all scenarios and all weather, and achieves high-precision and highly robust intelligent health monitoring and safety early warning.

[0006] To achieve the above objectives, the present invention provides a bridge health monitoring method, device, and readable medium based on multi-sensor fusion, comprising the following steps: S1. Establish a basic sensing system covering the entire bridge area, unify the spatiotemporal reference of all monitoring devices, and output and transmit the original heterogeneous sensor dataset after spatiotemporal alignment to S2 in real time. S2 receives the raw heterogeneous sensor data set transmitted by S1, and uses the Grubbs criterion combined with... The criteria are to identify sensor faults at a 95% confidence level, perform online self-calibration for sensors with minor faults, mark and remove the data of sensors with serious faults, and output the valid raw data after verification to S3. S3 receives the valid raw data transmitted by S2, performs full-link fusion processing on it, and outputs standardized and structured monitoring data to S4; S4 receives the standardized feature vector and structured monitoring parameter set transmitted by S3, realizes the off-flow processing of monitoring data based on the cloud-edge-device collaborative architecture, controls the total latency of the control end-edge-cloud within the safe threshold, realizes millisecond-level response to abnormal data and local preliminary early warning, and outputs abnormal early warning markers, global feature inference results and initial status judgment data vectors to S5. S5 receives the abnormal warning markers, global feature inference results and initial status judgment data vectors transmitted by S4, completes multi-source data cross-validation, bridge structure health index calculation, damage location, deterioration trend prediction and safety warning level classification, generates operation and maintenance suggestions and pushes them to the terminal.

[0007] Preferably, S1 specifically includes: S101. Through bridge structural mechanics simulation analysis and field survey, the key stress sections and high-incidence areas of bridge defects are determined, and the bridge is divided into mechanical monitoring area, environmental monitoring area and appearance defect monitoring area. S102. Differentiatedly deploy corresponding sensors in each zone: deploy contact mechanical sensors in the mechanical monitoring zone, deploy environmental parameter sensors in the environmental monitoring zone, and deploy visual and lidar non-contact sensors in the appearance defect monitoring zone. S103. Power on and test all sensors and perform signal acquisition tests to eliminate equipment and signal transmission faults. S104. Configure a BeiDou / GPS dual-mode industrial-grade clock synchronization module for the entire sensor group, and execute... Nanosecond-level time synchronization calibration to unify spatiotemporal reference; S105. Build a dual transmission network of wired fiber optic and 5G Mesh to transmit the original heterogeneous sensor dataset after spatiotemporal alignment to S2 in real time.

[0008] Preferably, S3 specifically includes: S301. Divide the valid raw data into three categories: mechanical response, environmental parameters, and appearance monitoring data, convert them into a unified standardized format, and establish a data classification index library. S302. Outliers are removed using a combination of the Laida criterion and box plot method. Noise is reduced using a combination of adaptive Kalman filtering and wavelet denoising. Finally, multi-source similar data calibration is completed using an adaptive weighted average fusion based on an improved entropy-weighted analytic hierarchy process. The fusion formula is as follows: ; In the formula, The calibration value is obtained after fusing data from multiple sources of the same type. The number of similar sensors participating in the fusion; For the first A collection of all sample data from a single type of sensor; For the first The combined dynamic weighting coefficients of the sensors satisfy the constraints. It is obtained by weighting subjective weights and objective weights, and the calculation formula is: ; In the formula, Subjective weighting; For objective weighting; S303. Extract core characteristic parameters reflecting the health status of the bridge from the calibration data to form a set of core characteristic parameters; S304. Decision-level complementary fusion of multi-source heterogeneous data is achieved through adaptive unscented Kalman filtering. The Kalman filter update formula is: ; In the formula, for The state value updated after filtering at each time step; for Time based The predicted state value at time; for Kalman gain at time step; for Sensor observations at any given time; for The observation matrix at time points; S305. Perform redundancy analysis on the fused data, remove invalid and redundant data, and output standardized feature vectors and structured monitoring parameter sets.

[0009] Preferably, the adaptive weighted average fusion algorithm of the improved entropy weight-analysis method in S302 specifically includes: Construct the original data matrix: ; In the formula, The number of similar sensors participating in the fusion; Number of samples per sensor; The sample number; The original data matrix; Subjective weights are calculated using the analytic hierarchy process. Calculating objective weights using an improved entropy weight method : The raw data is standardized, and the formula for standardizing positive indicators is: ; In the formula, For the first The first sensor Standardized values ​​of group samples; For the first The first sensor Group sample monitoring values; For the first The minimum value of sample data from a group of similar sensors; For the first The maximum value of sample data from a group of similar sensors; The standardized formula for negative indicators is: ; Calculate the first The first sensor Standardized proportion of group samples The formula is: ; In the formula, For the first The first sensor The standardized proportion of each sample data satisfies ; The formula for calculating information entropy is: ; In the formula, For the first Information entropy of sensor data; The coefficient of difference is calculated using the following formula: ; In the formula, For the first The coefficient of difference between individual sensor data; Normalization yields the objective weights, as shown in the formula: ; Calculate the dynamic weighting of the combination based on the optimized allocation ratio of 6:4 for bridge scenarios: ; In the formula, For the first The combined dynamic weighting coefficients of the sensors satisfy the constraints. Substitute the values ​​into the fusion formula to obtain the fused calibration values.

[0010] Preferably, S4 specifically includes: S401 pushes key parameters with high real-time requirements to edge computing nodes and synchronously uploads all fused data to the cloud server. S402. Edge computing nodes preprocess key parameters locally and compare them with preset security thresholds in real time. The total latency between the edge and cloud meets the following requirements: ; In the formula, Total latency for edge cloud; Data preprocessing latency; For data transmission delay; This is to account for data computation and processing latency; When parameters exceed limits, a local preliminary warning is immediately triggered. S403: The cloud server completes the aggregation and association of all data, establishes a global database, dynamically trains and optimizes the health assessment model, and dynamically calibrates the security threshold at the edge. S404: The edge device completes the initial judgment of the local state of the bridge, and the cloud device completes the initial judgment of the overall state. The abnormal warning marker, the global feature inference result, and the initial state judgment data vector are transmitted to S5.

[0011] Preferably, S5 specifically includes: S501. Integrate contact and non-contact monitoring data to conduct two-way cross-validation and eliminate misjudgment based on single data. S502. Based on the valid data after cross-validation, substitute the data into the formula for calculating the structural health index with structural importance coefficient to calculate the overall health score of the bridge. The formula is as follows: ; In the formula, The health index of bridge structure; To monitor the total number of core parameters; For the first The structural importance coefficient of the term parameter satisfies Key section parameters Non-critical parameters ; For the first Measured values ​​of the parameters; For the first The baseline value of the parameter; S503. Compare the measured data with the benchmark data of the bridge digital twin model to locate the specific location of the damage and generate the three-dimensional coordinates of the damage location; S504. Based on the LSTM temporal neural network, a deterioration trend prediction model is constructed to predict the structural health index change trend in the next 6-12 months. S505, combining structural health index, damage characteristics, and deterioration trends, classifies bridge safety early warning into four levels: Level 1 Early Warning Level II warning Level III early warning Level IV warning ; S506. Based on the warning level and damage characteristics, generate targeted operation and maintenance suggestions and specific operation and maintenance instructions; S507: The structural health index, damage location, deterioration trend, early warning report and operation and maintenance instructions are simultaneously pushed to the bridge health monitoring platform and multi-terminal operation and maintenance terminals.

[0012] Therefore, the present invention employs the above-mentioned bridge health monitoring method, equipment, and readable medium based on multi-sensor fusion. Compared with the prior art, the technical solution of the present invention has the following beneficial effects: (1) This invention adapts to all types of bridges through differentiated partitioning and nanosecond-level spatiotemporal synchronization; it adds a fault self-diagnosis and self-calibration link to eliminate / correct fault data from the source, ensure the validity of basic data, and solve the problem of data distortion in the prior art. (2) This invention is designed for bridge monitoring scenarios. It makes customized improvements to the general fusion algorithm and adopts an adaptive weighted average fusion method based on the improved entropy weight-analysis method to achieve calibration of similar data, which solves the problem of low accuracy of traditional fixed weight calibration. It also adopts an adaptive unscented Kalman filter to achieve heterogeneous data fusion, which adapts to the nonlinear structural response of bridges and eliminates data conflicts. (3) This invention realizes data diversion processing based on cloud-edge-end collaborative architecture. The edge end is equipped with a lightweight algorithm chip to complete local fast processing, and the cloud completes global data aggregation and model optimization. It strictly controls the total latency of the end-edge-cloud, realizes millisecond-level response to abnormal data and local preliminary warning, solves the problems of high transmission latency, delayed local warning and excessive computing load of traditional monitoring, and adapts to the long-term unattended operation and maintenance needs of bridges. (4) This invention achieves accurate quantitative assessment of bridge health status through the structural health index calculation formula with structural importance coefficient; based on the LSTM time-series neural network, it can predict potential risks 6-12 months in advance, realizing the upgrade from post-diagnosis to pre-warning, and filling the technical gap that the existing technology cannot achieve the prediction of bridge deterioration throughout its entire life cycle. (5) The present invention pushes the monitoring and diagnosis results and the deterioration trend prediction results to the bridge health monitoring platform and multi-terminal operation and maintenance terminal simultaneously, realizing data visualization, real-time early warning and precise instructions. Operation and maintenance personnel can obtain the bridge health status and operation and maintenance suggestions at the first time, greatly improve the bridge operation and maintenance efficiency, reduce operation and maintenance costs, and ensure the traffic safety and operational life of the bridge throughout its entire life cycle.

[0013] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating an embodiment of the bridge health monitoring method, device, and readable medium based on multi-sensor fusion of the present invention; Figure 2 This is a flowchart illustrating the multi-source data preprocessing and fusion process of an embodiment of the bridge health monitoring method, device, and readable medium based on multi-sensor fusion of the present invention. Figure 3 This is a flowchart of the fusion algorithm in an embodiment of the bridge health monitoring method, device and readable medium based on multi-sensor fusion of the present invention; Figure 4 This is an LSTM-based bridge structure deterioration trend prediction diagram, which is an embodiment of the bridge health monitoring method, equipment and readable medium based on multi-sensor fusion of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Unless otherwise defined, the technical or scientific terms used in the present invention should have the ordinary meaning understood by those skilled in the art.

[0016] Example 1 like Figures 1-4 As shown, this embodiment provides a bridge health monitoring method, device, and readable medium based on multi-sensor fusion. It should be understood that the specific parameters, models, and protocols mentioned in this embodiment are merely examples to help those skilled in the art understand the present invention, and are not intended to limit the present invention.

[0017] The bridge health monitoring method, device, and readable medium based on multi-sensor fusion of the present invention include the following steps: S1. Establish a basic sensing system that provides full coverage of the bridge and eliminates blind spots in monitoring, unify the spatiotemporal reference of all monitoring devices, and output and transmit the spatiotemporally aligned raw heterogeneous sensor dataset in real time. This includes the following steps: S101. First, through bridge structural mechanics simulation analysis and on-site surveys, key stress sections such as the mid-span / support of the main girder, the top / bottom of the piers, the anchorage zone / tower column of the tower, and the pot / plate bearings are identified. Simultaneously, high-incidence areas of bridge defects such as cracks, corrosion, settlement, and concrete spalling are marked. Based on monitoring functions, the bridge is divided into three major zones: a mechanical monitoring zone, an environmental monitoring zone, and a visual defect monitoring zone. S102. In the mechanical monitoring area, contact mechanical sensors such as fiber optic strain gauges, MEMS accelerometers, vibrating wire displacement gauges, and tilt sensors are deployed at key stress sections. The deployment spacing is adjusted according to the stress distribution density of the section. The deployment spacing in stress concentration areas is ≤2m, and the deployment spacing in ordinary areas is ≤5m. The deployment is denser in stress concentration areas. These sensors are used to collect mechanical response data such as strain, vibration, displacement, and tilt angle of the bridge structure. In the environmental monitoring area, temperature and humidity sensors and wind speed and direction sensors are deployed at locations such as the guardrails on both sides of the bridge and the top of the towers. At least one set is deployed every 500 meters, and an additional 2-3 sets are added to the main span of long-span bridges. These sensors are used to collect environmental parameter data such as temperature, humidity, wind speed, and wind direction of the bridge's service environment. In the appearance defect monitoring area, high-definition vision cameras (resolution ≥4K, frame rate ≥25fps), lidar (point cloud accuracy ≥1mm), and acoustic sensors are deployed at locations such as the bridge facade, the inner side of the box girder, and the perimeter of the supports. The monitoring view is unobstructed. These sensors are used to collect data such as bridge surface images, 3D point clouds, and acoustic signals of structural crack propagation. S103. After all sensors have been physically deployed, power-on debugging and signal acquisition tests are performed on each sensor to confirm that the sensors are working properly and the signal transmission is stable. Equipment faults and signal transmission faults are eliminated to ensure the validity of the sensor data. S104. Configure a dual-mode industrial-grade clock synchronization module (BDS / GPS) for the entire sensor group to perform... Nanosecond-level time synchronization calibration unifies the spatiotemporal reference of all monitoring devices, ensuring that the timing of data collected by sensors at different locations and of different types is completely matched, and avoiding data timing misalignment; S105. Establish a dual transmission network consisting of wired fiber optic and 5GMesh, with the wired fiber optic serving as the backbone network to ensure stable transmission of large volumes of data, and the 5GMesh serving as a backup network to enable wireless data retransmission at remote locations on the bridge, ensuring no dead zones in the network; transmit the original heterogeneous sensor dataset after spatiotemporal alignment to the S2 sensor fault self-diagnosis and self-calibration stage in real time and without loss through the dual transmission network. S2 receives the spatiotemporally aligned original heterogeneous sensor dataset transmitted by S1; and uses the Grubbs criterion combined with... The criteria are to identify sensor drift, jamming, and disconnection faults at a 95% confidence level; among them, sensor data drift ≤ 5% of full scale is judged as a minor fault, and sensor data drift > 5% of full scale, sensor jamming, or disconnection is judged as a serious fault.

[0018] For sensors with minor faults, online self-calibration is performed using effective data from redundant sensors with the same cross-section to correct drift data; for sensors with severe faults, they are marked and removed from the dataset to avoid fault data interfering with subsequent analysis.

[0019] The verified valid raw data is transmitted to the S3 multi-source data preprocessing and fusion stage.

[0020] S3. Perform customized end-to-end fusion processing on the effective raw data for bridge scenarios to output standardized and structured monitoring data. This includes the following steps: S301: Receive valid raw data transmitted by S2, classify it into three categories according to data type: mechanical response data, environmental parameter data, and appearance monitoring data, convert raw data of different formats and dimensions into a unified standardized format, establish a data classification index library, and realize standardized data management. S302. Outliers are removed by a combination of the Laida criterion and box plot method. Noise is reduced by a combination of adaptive Kalman filtering and wavelet denoising algorithm. Finally, multi-source similar data calibration is completed by adaptive weighted average fusion based on the improved entropy weight-analysis method.

[0021] Data collected from the same monitoring point at the same time A similar sensor Group sample data, construct the original data matrix: ; In the formula, The number of similar sensors participating in the fusion; Number of samples per sensor; The sample number; The original data matrix; The subjective weights are determined using the analytic hierarchy process (AHP) based on the structural importance of the sensor placement locations and measurement accuracy. Specifically: Based on the structural criticality of the sensor deployment location and the equipment calibration accuracy, Each sensor is compared and scored pairwise (using a 1-9 scale) to obtain a judgment matrix. ,in To determine the dimension of a matrix using the analytic hierarchy process.

[0022] Calculate the largest eigenvalue of the judgment matrix The subjective weight vector is obtained by normalizing the corresponding feature vectors and their features. .

[0023] Calculate the consistency index Combined with the average random consistency index To obtain the consistency ratio If CR < 0.1, then the test is passed and the weight allocation is reasonable.

[0024] The objective weights are determined using the entropy weight method based on the stability and dispersion of historical sensor data. Specifically: To eliminate the influence of dimensions, the original data Convert to standardized values: Positive indicators (the larger the better, such as signal strength) are represented as: ; In the formula, For the first The first sensor Standardized values ​​of group samples; For the first The first sensor Group sample monitoring values; For the first A collection of all sample data from a single type of sensor; Negative metrics (lower is better, such as noise levels) are represented as: ; Calculation percentage The formula is: ; In the formula, For the first The first sensor The standardized proportion of each sample data satisfies ; The formula for calculating the information entropy of each sensor's data is: ; In the formula, For the first Entropy of sensor data, value range ; The smaller the value, the greater the data dispersion and the higher the information content; The coefficient of difference is calculated using the following formula: ; In the formula, For the first The coefficient of difference between individual sensor data; The larger the value, the higher the discriminative power of the sensor data, and the greater its weighting value. Normalization yields objective weights The formula is: ; The subjective weights and objective weights are optimized in a 6:4 ratio for the bridge scenario to obtain the final combined weights: ; In the formula, For the first The combined dynamic weighting coefficients of the sensors satisfy the constraints. ; Substituting the combined weights into the adaptive weighted average fusion formula, we obtain the final calibration data for this monitoring point: ; In the formula, This is the calibration value after fusing data from multiple sources of the same type.

[0025] S303. Extract core characteristic parameters that reflect the health status of the bridge from the calibration data after data-level preprocessing, including mechanical characteristic parameters (peak strain, natural frequency of vibration, displacement amplitude, rate of change of tilt angle), environmental characteristic parameters (extreme wind speed, temperature and humidity change gradient), and appearance defect characteristic parameters (crack length / width, corrosion area, surface deformation, frequency of acoustic anomaly signal), to form a core characteristic parameter set, thereby realizing data dimensionality reduction and feature extraction. S304. Decision-level complementary fusion of multi-source heterogeneous data is achieved through adaptive unscented Kalman filtering. The Kalman filter update formula is: ; In the formula, for The state value updated after filtering at each time step; for Time based The predicted state value at time; for Kalman gain at time step; for Sensor observations at any given time; for The observation matrix at time points; Kalman gain With an adaptive noise covariance adjustment factor, it can adjust in real time according to the vibration amplitude of the bridge structure and the ambient wind speed, adapting to the nonlinear structural response of bridges under strong wind and heavy traffic conditions. It globally and collaboratively integrates mechanical response data, environmental parameter data, and appearance monitoring data, eliminating conflicts and biases between different types of data, achieving complementary advantages, and improving data reliability.

[0026] S305. Perform redundancy analysis on the fused full data, remove redundant data that is duplicated, featureless, or meaningless for bridge health assessment, compress the data volume, and improve subsequent calculation efficiency.

[0027] The standardized feature vectors after noise reduction and calibration, and the fused structured monitoring parameter set are transmitted to the S4 data intelligent computing and real-time response stage via the cloud-edge-device transmission link. Data feature tags and fusion accuracy indicators are attached during the transmission process.

[0028] S4. Based on a cloud-edge-device collaborative architecture, monitoring data is processed rapidly by offloading, ensuring the total latency between the control device, edge, and cloud is within a safe threshold. This enables millisecond-level response to abnormal data and initial local warnings. The specific steps are as follows: S401 receives the standardized feature vector and structured monitoring parameter set transmitted by S3, performs data diversion based on the cloud-edge-device collaborative architecture, pushes key parameters with high real-time requirements such as strain peak value, vibration natural frequency mutation value, and displacement over-limit value to the edge computing node, and synchronously uploads the full fused data to the cloud server in time series. S402 edge computing nodes are equipped with lightweight algorithm chips such as FPGA chips or embedded AI chips to meet local fast computing needs.

[0029] The received key parameters undergo local lightweight preprocessing, including data smoothing, feature value comparison, and time-series trend analysis. The preprocessed key parameters are then compared in real-time with a preset safety threshold, which is determined based on bridge design values, historical health data, and a structural mechanics model. Response latency is controlled via edge-cloud total latency, calculated using the following formula: ; In the formula, Total latency for edge cloud; Data preprocessing latency; For data transmission delay; This is to account for data computation and processing latency; Precise latency control is achieved through algorithm optimization and network speed enhancement; when parameters exceed limits are detected, a local preliminary warning is immediately triggered, including audible and visual warning signals, real-time annotation of abnormal parameters, and local caching of warning information.

[0030] S403 The cloud server receives valid data uploaded by edge computing nodes, including normal parameter data and abnormal parameter data, completes the time series summary and spatial location correlation of the full bridge monitoring data, and establishes a global database for bridge health monitoring.

[0031] Based on historical and real-time data from a global database, the bridge health assessment model is dynamically trained and optimized. It is updated every 24 hours under normal operating conditions and in real-time under abnormal operating conditions, improving the model's diagnostic accuracy. Simultaneously, the safety thresholds at the edge points are dynamically calibrated according to the long-term aging of the bridge structure and environmental changes, ensuring the reasonableness and adaptability of the thresholds.

[0032] S404: The edge-end data, combined with threshold comparison results, makes a preliminary judgment on the local state of the bridge. The cloud-based data, combined with global data, makes a preliminary inference on the overall state of the bridge, forming an initial state judgment conclusion. The edge-end anomaly warning markers (including anomaly locations, anomaly parameters, out-of-limit values, and warning trigger times), global feature inference results, and initial state judgment data vectors are synchronously and in real time transmitted to the S5 structural assessment and damage diagnosis stage via an encrypted transmission link. Among them, the anomaly warning markers are high-priority data, achieving priority transmission and priority analysis.

[0033] S5. Complete multi-source data cross-validation, quantitative calculation of bridge structural health index, precise damage location, structural deterioration trend prediction, and safety early warning level classification; generate targeted operation and maintenance suggestions and instructions; and push them to the terminal to achieve seamless integration of monitoring and operation and maintenance. This includes the following steps: S501 receives edge-end anomaly warning markers, global feature inference results, and initial state judgment data vectors transmitted from S4. It integrates contact-based mechanical sensing data with non-contact visual / LiDAR / acoustic monitoring data for bidirectional cross-validation. Contact-based mechanical sensing data verifies internal stress anomalies in the bridge structure, while non-contact monitoring data verifies external structural defects. This mutual corroboration eliminates misjudgments caused by single data sources, ensuring data authenticity and diagnostic accuracy.

[0034] S502. Based on the valid data after cross-validation, substitute the data into the formula for calculating the structural health index with structural importance coefficient to calculate the overall health score of the bridge. The formula is as follows: ; In the formula, The health index of bridge structure; To monitor the total number of core parameters; For the first The structural importance coefficient of the term parameter satisfies Key section parameters Non-critical parameters ; For the first Measured values ​​of the parameters; For the first The baseline value of the parameter; S503. The measured data is accurately compared with the baseline data of the digital twin model that restores the bridge's physical structure at a 1:1 scale. Point cloud matching and stress field mapping algorithms are used. The digital twin model includes the bridge's geometric features, material properties, stress characteristics, and design parameters.

[0035] By using the model's 3D visualization function, the specific locations of hidden bridge damage (such as internal structural microcracks and internal bearing aging) and surface defects (such as external cracks, corrosion, and concrete spalling) can be located, generating 3D coordinates for damage location and clarifying the structural part of the bridge to which the damage belongs.

[0036] S504. Based on an LSTM (Long Short-Term Memory) temporal neural network, using historical monitoring data of the bridge over the past 12 months as the training set (normalized to the [0,1] interval, divided into training and validation sets at a 9:1 ratio), a deterioration trend prediction model is constructed using two LSTM hidden layers (64 neurons per layer) and an Adam optimizer (learning rate 0.001). The convergence criterion is the minimum mean square error of the validation set. This model predicts the structural health index changes over the next 6-12 months, allowing for early warning of potential bridge structural deterioration risks and upgrading from post-construction diagnosis to pre-construction early warning.

[0037] S505, combining structural health index, damage type, damage degree, importance of damage location, and deterioration trend prediction results, classifies bridge safety early warning into four levels: Level 1 warning ; Level II warning ; Level 3 warning ; Level IV warning ; To achieve quantitative classification of bridge health status.

[0038] S506. Based on the bridge's safety warning level, damage location results, damage type, and deterioration trend prediction results, generate targeted operation and maintenance suggestions and specific operation and maintenance instructions: Level 1 warning recommends conducting routine inspections in accordance with regulations; The Level 2 warning recommends localized repairs for minor damage and regular retesting of relevant parameters; The Level 3 warning recommends immediately closing some lanes, conducting comprehensive structural inspections, and completing repairs within a specified timeframe. A Level IV warning recommends immediately closing the bridge and activating the emergency repair plan to ensure bridge traffic safety.

[0039] S507. The complete structural health index, three-dimensional coordinates of damage location, deterioration trend prediction results, early warning level reports and operation and maintenance instructions are simultaneously pushed to the bridge health monitoring platform and operation and maintenance terminal. The bridge health monitoring platform realizes data visualization, three-dimensional model linkage and online report generation. The operation and maintenance terminal includes the operation and maintenance personnel's mobile APP, the monitoring center's large screen and the maintenance unit's management system, ensuring that the operation and maintenance personnel can obtain monitoring and diagnosis results as soon as possible.

[0040] Therefore, the present invention adopts the above-mentioned bridge health monitoring method, equipment and readable medium based on multi-sensor fusion. The method solves the pain points such as the limitation of single monitoring dimension, insufficient heterogeneous data fusion and poor real-time performance through multimodal full-domain perception, three-level full-link fusion algorithm and cloud-edge-device collaborative architecture. It is adapted to the all-weather operation and maintenance needs of various bridges in all scenarios and achieves high-precision and robust intelligent health monitoring and safety early warning.

[0041] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0042] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A bridge health monitoring method based on multi-sensor fusion, characterized in that, Includes the following steps: S1. Establish a basic sensing system covering the entire bridge area, unify the spatiotemporal reference of all monitoring devices, and output and transmit the original heterogeneous sensor dataset after spatiotemporal alignment to S2 in real time. S2 receives the raw heterogeneous sensor data set transmitted by S1, and uses the Grubbs criterion combined with... The criteria are to identify sensor faults at a 95% confidence level, perform online self-calibration on sensors with minor faults, mark and remove the data of sensors with serious faults, and output the valid raw data after verification to S3. S3 receives the valid raw data transmitted by S2, performs full-link fusion processing on it, and outputs standardized and structured monitoring data to S4; S4 receives the standardized feature vector and structured monitoring parameter set transmitted by S3, realizes the off-flow processing of monitoring data based on the cloud-edge-device collaborative architecture, controls the total latency of the control end-edge-cloud within the safe threshold, realizes millisecond-level response to abnormal data and local preliminary early warning, and outputs abnormal early warning markers, global feature inference results and initial status judgment data vectors to S5. S5 receives the abnormal warning markers, global feature inference results and initial status judgment data vectors transmitted by S4, completes multi-source data cross-validation, bridge structure health index calculation, damage location, deterioration trend prediction and safety warning level classification, generates operation and maintenance suggestions and pushes them to the terminal.

2. The bridge health monitoring method based on multi-sensor fusion according to claim 1, characterized in that, S1 specifically includes: S101. Through bridge structural mechanics simulation analysis and field survey, the key stress sections and high-incidence areas of bridge defects are determined, and the bridge is divided into mechanical monitoring area, environmental monitoring area and appearance defect monitoring area. S102. Differentiatedly deploy corresponding sensors in each zone: deploy contact mechanical sensors in the mechanical monitoring zone, deploy environmental parameter sensors in the environmental monitoring zone, and deploy visual and lidar non-contact sensors in the appearance defect monitoring zone. S103. Power on and test all sensors and perform signal acquisition tests to eliminate equipment and signal transmission faults. S104. Configure a BeiDou / GPS dual-mode industrial-grade clock synchronization module for the entire sensor group, and execute... Nanosecond-level time synchronization calibration to unify spatiotemporal reference; S105. Establish a dual transmission network of wired fiber optic and 5G Mesh to transmit the original heterogeneous sensor dataset after spatiotemporal alignment to S2 in real time.

3. The bridge health monitoring method based on multi-sensor fusion according to claim 2, characterized in that, S3 specifically includes: S301. Divide the valid raw data into three categories: mechanical response, environmental parameters, and appearance monitoring data, convert them into a unified standardized format, and establish a data classification index library. S302. Outliers are removed using a combination of the Laida criterion and box plot method. Noise is reduced using a combination of adaptive Kalman filtering and wavelet denoising. Finally, multi-source similar data calibration is completed using an adaptive weighted average fusion based on an improved entropy-weighted analytic hierarchy process. The fusion formula is as follows: ； In the formula, The calibration value is obtained by fusing data from multiple sources of the same type. The number of similar sensors participating in the fusion; For the first A collection of all sample data from a single type of sensor; For the first The combined dynamic weighting coefficients of the sensors satisfy the constraints. It is obtained by weighting subjective weights and objective weights, and the calculation formula is: ； In the formula, Subjective weighting; For objective weighting; S303. Extract core characteristic parameters reflecting the health status of the bridge from the calibration data to form a core characteristic parameter set; S304. Decision-level complementary fusion of multi-source heterogeneous data is achieved through adaptive unscented Kalman filtering. The Kalman filter update formula is: ； In the formula, for The state value updated after filtering at each time step; for Time based The predicted state value at time; for Kalman gain at time step; for Sensor observations at any given time; for The observation matrix at each time point; S305. Perform redundancy analysis on the fused data, remove invalid and redundant data, and output standardized feature vectors and structured monitoring parameter sets.

4. The bridge health monitoring method based on multi-sensor fusion according to claim 3, characterized in that, The improved entropy-weighted analytic hierarchy process (AHP) adaptive weighted average fusion algorithm in S302 specifically includes: Construct the original data matrix: ； In the formula, The number of similar sensors participating in the fusion; This refers to the number of samples per sensor. The sample number; The original data matrix; Subjective weights are calculated using the analytic hierarchy process. Calculating objective weights using an improved entropy weight method : The raw data is standardized, and the formula for standardizing positive indicators is: ； In the formula, For the first The first sensor Standardized values ​​of group samples; For the first The first sensor Group sample monitoring values; For the first The minimum value of sample data from a group of similar sensors; For the first The maximum value of sample data from a group of similar sensors; The standardized formula for negative indicators is: ； Calculate the first The first sensor Standardized proportion of group samples The formula is: ； In the formula, For the first The first sensor The standardized proportion of each sample data satisfies ; The formula for calculating information entropy is: ； In the formula, For the first Information entropy of sensor data; The coefficient of difference is calculated using the following formula: ； In the formula, For the first The coefficient of difference between individual sensor data; Normalization yields the objective weights, as shown in the formula: ； Calculate the dynamic weighting of the combination based on the optimized allocation ratio of 6:4 for bridge scenarios: ； In the formula, For the first The combined dynamic weighting coefficients of the sensors satisfy the constraints. ; Substitute the values ​​into the fusion formula to obtain the fused calibration values.

5. The bridge health monitoring method based on multi-sensor fusion according to claim 4, characterized in that, S4 specifically includes: S401 pushes key parameters with high real-time requirements to edge computing nodes and synchronously uploads all fused data to the cloud server. S402. Edge computing nodes preprocess key parameters locally and compare them with preset security thresholds in real time. The total latency between the edge and cloud meets the following requirements: ； In the formula, Total latency for edge cloud; Data preprocessing latency; For data transmission delay; This is to account for data computation and processing latency; When parameters exceed limits, a local preliminary warning is immediately triggered. S403: The cloud server completes the aggregation and association of all data, establishes a global database, dynamically trains and optimizes the health assessment model, and dynamically calibrates the security threshold at the edge. S404: The edge terminal completes the initial judgment of the local state of the bridge, and the cloud completes the initial judgment of the overall state. The abnormal warning mark, global feature inference results, and initial state judgment data vector are transmitted to S5.

6. The bridge health monitoring method based on multi-sensor fusion according to claim 5, characterized in that, S5 specifically includes: S501. Integrate contact and non-contact monitoring data to conduct two-way cross-validation and eliminate misjudgment based on single data. S502. Based on the valid data after cross-validation, substitute the data into the formula for calculating the structural health index with structural importance coefficient to calculate the overall health score of the bridge. The formula is as follows: ； In the formula, The health index of bridge structure; To monitor the total number of core parameters; For the first The structural importance coefficient of the term parameter satisfies Key section parameters Non-critical parameters ; For the first Measured values ​​of the parameters; For the first The baseline value of the parameter; S503. Compare the measured data with the benchmark data of the bridge digital twin model to locate the specific location of the damage and generate the three-dimensional coordinates of the damage location; S504. Construct a degradation trend prediction model based on LSTM temporal neural network to predict the structural health index change trend in the next 6-12 months. S505, combining structural health index, damage characteristics, and deterioration trends, classifies bridge safety early warning into four levels: Level 1 Early Warning Level II warning Level III early warning Level IV warning ; S506. Based on the warning level and damage characteristics, generate targeted operation and maintenance suggestions and specific operation and maintenance instructions; S507: The structural health index, damage location, deterioration trend, early warning report and operation and maintenance instructions are simultaneously pushed to the bridge health monitoring platform and multi-terminal operation and maintenance terminals.

7. A computer device, characterized in that, include: A processor configured to be coupled to memory, read and execute instructions and / or program code in the memory to perform the method as described in any one of claims 1-6.

8. A computer-readable medium, characterized in that, The computer-readable medium stores computer program code that, when executed on a computer, causes the computer to perform the method as described in any one of claims 1-6.

Images