Health monitoring system based on civil engineering structure

Abstract

The invention relates to the technical field of civil engineering monitoring, and discloses a structure health monitoring system based on civil engineering. A high-density sensing network of the system collects strain field distribution, vibration spectrum and environmental corrosion parameters of a structure through a distributed multi-mode sensor array; performing clock drift compensation and space coordinate normalization on the asynchronous sampling data by a space-time alignment engine to generate an original feature tensor of a unified space-time reference; the semantic modeling unit is combined with a design drawing and a material parameter library, the original feature tensor is mapped to a component semantic space, and hierarchical structure features with topological marks are output; the adaptive fusion core executes dynamic weight distribution, eliminates sensor conflict data and generates an anti-interference fusion diagnosis index; the edge computing node operates a lightweight damage detection model according to the fusion diagnosis index, and outputs a local component health state level; and the cloud collaborative analyzer aggregates multiple edge node results and predicts the overall residual life of the structure in combination with historical degradation data.

Description

Technical Field

[0001] This invention relates to the field of civil engineering monitoring technology, specifically to a health monitoring system for civil engineering structures. Background Technology

[0002] In the field of civil engineering, large structures such as bridges, tunnels, and high-rise buildings are subject to various factors during long-term service, including loads, environmental erosion, and material aging, leading to a gradual degradation of their structural performance and potentially causing safety hazards or even accidents. Therefore, effective health monitoring of civil engineering structures and timely understanding of changes in their structural condition are crucial means to ensure the safe operation of structures.

[0003] In existing civil engineering structural health monitoring technologies, sensor deployment is mostly single-point or sparsely distributed, making it difficult to comprehensively capture the overall condition information of the structure. Some monitoring systems use only a single type of sensor, which can only acquire data in a single dimension such as strain or vibration, failing to comprehensively reflect the health status of the structure under the influence of multiple factors. At the same time, due to the differences in sampling frequency and data format among different types of sensors, the asynchronous sampling data lacks a unified spatiotemporal reference, making it difficult to effectively correlate the data and complicating subsequent data processing and analysis.

[0004] In the data processing stage, existing technologies often perform simple analysis of raw monitoring data without fully integrating the structure's design information and material properties. This fails to effectively correlate the monitoring data with the physical meaning of structural components, making it difficult for the analysis results to accurately reflect the health status of specific components. Furthermore, data collected by different sensors may conflict or contain errors. Existing fusion algorithms mostly use fixed weight allocation methods, making it difficult to dynamically adjust according to data reliability. This results in weak anti-interference capabilities of the fused diagnostic indicators, making it impossible to accurately identify structural damage.

[0005] Regarding the application of monitoring results, existing systems mostly focus on detecting damage to local components, lacking the ability to predict the overall remaining life of the structure. Although some systems attempt to conduct overall analysis, they do not fully integrate the results of multiple local monitoring nodes, nor do they combine historical degradation data to establish scientific predictive models. This results in an incomplete assessment of the overall safety status of the structure, making it difficult to meet the needs of long-term structural safety management in engineering practice. Summary of the Invention

[0006] The purpose of this invention is to provide a health monitoring system for civil engineering structures to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a health monitoring system for civil engineering structures, the system comprising: High-density sensor networks are used to collect strain field distribution, vibration spectrum and environmental corrosion parameters of structures through distributed multimodal sensor arrays; The spatiotemporal alignment engine performs clock drift compensation and spatial coordinate normalization on the asynchronous sampling data of the multimodal sensor array, generating an original feature tensor with a unified spatiotemporal reference. The semantic modeling unit maps the original feature tensors to a preset component semantic space based on the structural design drawings and material parameter library, and outputs hierarchical structural features with topological tags. The adaptive fusion core receives hierarchical structural features and performs dynamic weight allocation. It eliminates conflicting data between sensors through iterative optimization algorithms and generates anti-interference fusion diagnostic indicators. Edge computing nodes run lightweight damage detection models in real time based on fused diagnostic indicators and output the health status level of local components. The cloud-based collaborative analyzer aggregates the outputs of multiple edge computing nodes and combines them with historical degradation data to predict the overall remaining lifespan of the structure.

[0008] Preferably, the high-density sensor network includes: The strain field acquisition subnetwork uses a fiber optic grating sensor array to measure the micro-strain on the surface of the structure and calculates the principal stress direction by Bragg wavelength offset. The vibration sensing subnet captures the structure’s natural frequency and damping ratio through a piezoelectric accelerometer array and extracts the frequency band energy features using wavelet packet decomposition. The corrosion monitoring subnetwork uses electrochemical impedance sensors to detect the chloride ion concentration and steel reinforcement polarization resistance in concrete.

[0009] Preferably, the spatiotemporal alignment engine performs: Lagrange interpolation compensation is performed on the wavelength sampling sequence of the fiber Bragg grating sensor array to eliminate the time offset caused by transmission delay; Establish a spatial position relationship diagram of the sensors, and transform the local coordinate system of the piezoelectric accelerometer array to the global reference system through Lie algebra transformation; A sliding window Kalman filter was used to perform time synchronization calibration on the slowly varying signal of the electrochemical impedance sensor.

[0010] Preferably, the semantic modeling unit includes: The component segmentation module divides the original feature tensor into feature blocks corresponding to the physical components based on the beam-column node information in the design drawings. The material property binding module extracts concrete grade and steel reinforcement specifications from the material parameter library and adds elastic modulus and yield strength labels to each feature block. The topology encoder generates the connection stiffness matrix between components based on the structural mechanics model and embeds it into the hierarchical structural features.

[0011] Preferably, the adaptive fusion core performs: Construct a conflict detection model based on Mahalanobis distance to identify anomalous data blocks that exceed the statistical confidence interval in hierarchical structural features; The reliability weights of multimodal sensor data are estimated using a Gaussian mixture model, and abnormal data blocks are reconstructed using weighted methods. A variational autoencoder is used to perform low-dimensional manifold projection on the reconstructed features to generate fusion diagnostic indicators.

[0012] Preferably, the edge computing node includes: The microcrack detection unit analyzes and fuses the strain concentration coefficient in diagnostic indicators using a depth-separable convolutional network. The stiffness degradation assessment unit calculates the equivalent stiffness attenuation rate of the component based on the vibration spectrum characteristics and the connection stiffness matrix. A corrosion risk quantification unit predicts the corrosion rate of steel bars based on the chloride ion diffusion equation and polarization resistance value.

[0013] Preferably, the cloud-based collaborative analyzer performs: Construct a finite element model of the entire structure and use the health status level of local components as boundary condition input; A time series prediction network was used to simulate the stress redistribution process under different loading conditions. The reliability index of the structural system is calculated by Monte Carlo sampling, and the confidence interval of the remaining lifetime is output.

[0014] Preferably, the system further includes: The interface is dynamically updated to automatically correct the material constitutive relations of the finite element model based on newly acquired hierarchical structural features; The feedback adjustment loop compares the residuals between the predicted strain field and the actual measured value, and adjusts the convolution kernel parameters of the depth-separable convolutional network.

[0015] Preferably, the dynamic update interface includes: The parameter sensitivity analysis module identifies the material parameters that have the greatest impact on structural reliability. The incremental learning engine updates the Young's modulus matrix of the finite element model through an online backpropagation algorithm.

[0016] Preferably, the feedback adjustment loop includes: The residual distribution statistics module calculates the root mean square value of the prediction error for each component node; An adaptive learning rate scheduler dynamically adjusts the update stride of the convolution kernel parameters based on the root mean square value.

[0017] Compared with the prior art, the beneficial effects of the present invention are: High-density sensor networks employ distributed multimodal sensor arrays, enabling simultaneous acquisition of strain field distribution, vibration spectrum, and environmental corrosion parameters of structures. This breaks through the limitations of traditional monitoring systems that rely on single sensors or sparse deployments. It can acquire structural status information from multiple dimensions, comprehensively reflecting the health status of structures under the influence of different factors. This results in broader and richer monitoring data coverage, providing a comprehensive data source for subsequent accurate analysis of structural health status.

[0018] The spatiotemporal alignment engine performs clock drift compensation and spatial coordinate normalization on the asynchronous sampling data of the multimodal sensor array, generating a raw feature tensor with a unified spatiotemporal reference. This process solves the problem of data correlation caused by differences in sampling frequency and data format between different sensors, transforming the originally scattered and asynchronous data into a unified standard feature tensor. This eliminates data deviations in the temporal and spatial dimensions, laying a solid foundation for subsequent data transfer and collaborative analysis between modules, and making the data processing of the entire monitoring process more coherent and accurate.

[0019] The semantic modeling unit maps the original feature tensors to a pre-defined component semantic space based on the structure's design drawings and material parameter library, outputting hierarchical structural features with topological tags. This unit fully integrates the structure's design information and material properties, corresponding abstract original feature tensors to specific structural components. This gives the monitoring data clear physical meaning and component attribution, enabling subsequent damage detection and health assessment to accurately locate specific components. This avoids the data-component disconnect problem in traditional monitoring, making the monitoring results more targeted and practical, and allowing engineers to intuitively understand the health status of each component.

[0020] The adaptive fusion core receives hierarchical structural features and performs dynamic weight allocation. It eliminates conflicting data between sensors through iterative optimization algorithms, generating interference-resistant fusion diagnostic indicators. Compared to traditional fixed-weight fusion algorithms, this core can dynamically adjust weights based on the reliability of data from different sensors, effectively identifying and eliminating conflicting data. This minimizes the impact of data errors and interference on diagnostic results, and the generated fusion diagnostic indicators more accurately reflect the structural health, improving the reliability of diagnostic results and providing high-quality data support for subsequent damage detection of edge computing nodes.

[0021] Edge computing nodes run a lightweight damage detection model in real time based on fused diagnostic indicators, outputting the health status level of local components. The application of the lightweight model ensures the real-time nature of damage detection, enabling rapid response to changes in structural condition and timely output of the health status of local components. This allows engineers to grasp the health status of each local component immediately, facilitating timely investigation and handling of components with potential hazards. It effectively avoids safety risks caused by detection delays and improves the timeliness and efficiency of structural safety management.

[0022] The cloud-based collaborative analyzer aggregates the outputs of multiple edge computing nodes and combines them with historical degradation data to predict the overall remaining lifespan of the structure. This module achieves global integration of local monitoring results, moving beyond the assessment of single local components. Instead, it takes a holistic perspective, combining long-term historical degradation data to predict the overall remaining lifespan of the structure. This provides engineers with a comprehensive reference for developing long-term structural maintenance and repair plans, helping to rationally allocate resources, extend the service life of structures, ensure the long-term safe and stable operation of structures, and reduce economic losses and safety risks caused by blind maintenance or neglecting hidden dangers. Attached Figure Description

[0023] Figure 1 This is a timing diagram of the health monitoring system for civil engineering structures described in this invention; Figure 2 A flowchart illustrating the workflow of a high-density sensor network; Figure 3 This is a flowchart of the spatiotemporal alignment engine's workflow. Detailed Implementation

[0024] 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, 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.

[0025] Please see Figure 1 This invention provides a health monitoring system for civil engineering structures, the system comprising a high-density sensor network, a spatiotemporal alignment engine, a semantic modeling unit, an adaptive fusion core, edge computing nodes, and a cloud-based collaborative analyzer working together.

[0026] A high-density sensor network collects strain field distribution, vibration spectrum, and environmental corrosion parameters of the structure through a distributed array of multimodal sensors, including fiber optic grating sensors, piezoelectric accelerometers, and electrochemical impedance sensors, to cover the key physical parameters of the structure. A spatiotemporal alignment engine receives asynchronous sampling data from the multimodal sensor array and performs clock drift compensation and spatial coordinate normalization. Clock drift compensation uses Lagrange interpolation to align the time series, and spatial coordinate normalization uses Lie algebra transformation to map the local coordinate system to the global reference system, ultimately generating an original feature tensor with a unified spatiotemporal reference. The semantic modeling unit maps the original feature tensor to a predefined component semantic space based on the structure's design drawings and material parameter library. This unit analyzes the geometric information in the design drawings, divides the original feature tensor into feature blocks corresponding to physical components, and extracts attributes such as concrete grade and steel reinforcement specifications from the material parameter library. It adds labels such as elastic modulus and yield strength to each feature block and generates a connection stiffness matrix between components based on the structural mechanics model, outputting hierarchical structural features with topological tags. The adaptive fusion core receives hierarchical structural features and performs dynamic weight allocation to eliminate conflicting data between sensors. This core detects anomalous data blocks using a Mahalanobis distance model and estimates the reliability weights of each sensor's data using a Gaussian mixture model. It then employs a variational autoencoder for feature reconstruction, generating anti-interference fusion diagnostic indicators. Edge computing nodes run lightweight damage detection models in real-time based on these fusion diagnostic indicators. These models include microcrack detection, stiffness degradation assessment, and corrosion risk quantification modules. These models analyze the local component state based on deep separable convolutional networks and physical equations, outputting health status levels. A cloud-based collaborative analyzer aggregates the results from multiple edge computing nodes, combines them with historical degradation data to construct a finite element model of the overall structure, simulates stress redistribution using a time-series prediction network, and calculates system reliability using Monte Carlo sampling, ultimately outputting the remaining lifetime confidence interval. The system achieves adaptive optimization through a dynamic update interface and a feedback adjustment loop. The dynamic update interface corrects the constitutive relationship of the finite element model based on new data, while the feedback adjustment loop adjusts model parameters through residual comparison to ensure long-term monitoring accuracy.

[0027] Example 1: See Figure 2The strain field acquisition subnetwork is deployed using a fiber Bragg grating sensor array. Sensors are attached to the concrete surface or embedded in the structural surface in a grid pattern. Each sensing unit uses ultraviolet laser writing technology to create a Bragg grating, with the grating period optimized according to a preset measurement range. The array cabling employs a hybrid star and bus topology architecture. The backbone fiber is laid along the main stress path of the structure, and branch nodes connect individual sensing units to reduce signal attenuation. During data acquisition, the demodulation device emits a broadband light source. The Bragg wavelength shift reflected by the sensor is linearly related to the strain change. The wavelength demodulator records the shift and converts it into micro-strain values. Principal stress direction calculation combines a flower-shaped sensor group composed of strain gauges from multiple directions. The plane stress state is analyzed using the Mohr's circle principle, ultimately generating a strain contour map covering the structural surface. The vibration sensing subnetwork uses a piezoelectric accelerometer array of shear-type triaxial sensors. The installation positions are determined through modal analysis to cover the main vibration mode nodes of the structure. The sensor bases are fixed with epoxy resin to ensure synchronous vibration with the structure, and the signal cables are shielded from electromagnetic interference by metal conduits. The acquisition system is set with an adjustable sampling frequency, typically 200Hz for bridge structures to capture modal information within the 0-100Hz range. The wavelet packet decomposition algorithm is implemented using the Mallat fast algorithm, performing eight-level decomposition of the acceleration signal to obtain 256 sub-bands. The energy value of each sub-band is calculated and normalized to form a feature vector. The correlation between frequency band energy characteristics and structural damage is mapped using a historical database. For example, a decrease in local stiffness can lead to an increase in the proportion of high-frequency band energy. The system identifies abnormal vibration modes by tracking these feature changes in real time.

[0028] The electrochemical impedance spectroscopy (EIS) sensor in the corrosion monitoring subnetwork employs a three-electrode system: a pre-embedded stainless steel rod as the working electrode, an Ag / AgCl electrode as the reference electrode, and a platinum wire electrode as the counter electrode. The sensor is pre-embedded at different depths in the protective layer before concrete pouring, with the electrode spacing set to 2-5 cm based on the rebar mesh size. The excitation signal is generated by a potentiostat, with a frequency scanning range from 10 mHz to 100 kHz. The measured Nyquist plot is fitted to an equivalent circuit model using ZView software. The chloride ion concentration is calculated based on the diameter of the diffusion impedance arc in the impedance spectrum, and the polarization resistance is derived from the real part of the impedance in the low-frequency region. The temperature compensation module integrates a PT100 temperature sensor to correct for the temperature effects on electrochemical parameters in real time. The sensor network's power supply system uses a hybrid solution of solar panels and supercapacitors, while the wireless transmission module uses a hybrid communication protocol combining 4G / LTE and LoRa. The data acquisition terminal is equipped with edge computing capabilities, enabling preliminary filtering and compression of the raw data. The network management system includes automatic diagnostic functions, periodically checking sensor connection status and battery power; abnormal nodes trigger alarm signals. Clock synchronization employs the IEEE 1588 precision time protocol, with all sensor nodes achieving microsecond-level time synchronization via a master clock. The fiber Bragg grating sensor array's wavelength demodulation device utilizes tunable Fabry-Perot filter technology, with a scanning frequency set to 100Hz and a wavelength resolution of 1pm. Each sensing channel is equipped with an optical switch for cyclic detection, and optical path attenuation is monitored in real-time to ensure data reliability. Temperature compensation is achieved by setting a reference grating, which is separately packaged in an unstressed area. Strain transfer efficiency is optimized through finite element analysis, and the sensor packaging structure employs a multi-layered protection design to reduce shear hysteresis. The piezoelectric accelerometer's signal conditioning circuit includes a charge amplifier and an anti-aliasing filter, with the output voltage range adjusted to ±5V to accommodate the data acquisition card's range. The sensor sensitivity coefficient is stored in an embedded chip, automatically read and converted into physical quantities by the system. The array layout employs a strategy combining uniform point distribution with increased density at key locations, increasing the density of measurement points in stress concentration areas such as around openings and near supports. The modal recognition algorithm uses a random subspace method to automatically extract the first 10 frequencies and damping ratio parameters.

[0029] The potentiostat circuit of the electrochemical impedance sensor employs a zero-resistance galvanometer design to eliminate the influence of wire resistance on the measurement results. Shielded twisted-pair cables are used for the electrode leads to reduce AC interference, and contact impedance changes caused by concrete drying shrinkage are compensated for through periodic calibration. The monitoring cycle is set according to environmental corrosivity: once daily in splash zones and once weekly in atmospheric zones. Data validity verification is achieved through repeated measurements, with automatic retesting when the coefficient of variation exceeds 5%. The network communication protocol uses a custom binary format to compress data packets, with the packet header containing a timestamp, node ID, and data type identifier. The routing algorithm dynamically selects the transmission path based on signal strength, switching communication modes when the packet loss rate exceeds a threshold. The data security mechanism uses the AES encryption algorithm, with each sensor node possessing an independent key. Remote configuration allows adjustment of sampling parameters, and the system maintains dual buffers for parallel processing of acquisition and transmission. The self-calibration mechanism is implemented through a built-in standard signal source, periodically injecting electrical signals of known amplitude to verify channel gain. Sensor health monitoring includes impedance detection and baseline drift analysis; abnormal sensor data is marked and excluded from fusion calculations. Network coverage optimization is based on a ray tracing algorithm, deploying relay nodes within complex steel structures to eliminate signal shadowing. The electromagnetic compatibility design includes a feedthrough filter and a shielded enclosure to ensure stable operation in harsh industrial environments.

[0030] Example 2: See Figure 3The spatiotemporal alignment engine and semantic modeling unit components form a key bridge from raw data to engineering semantic transformation. The spatiotemporal alignment engine processes heterogeneous data streams from high-density sensor networks. Its primary task is to time-align the asynchronous sampling sequence of the fiber Bragg grating sensor array. The crystal oscillator built into each sensor node has a slight frequency offset, causing clock drift. The engine uses the standard time provided by the BeiDou time service module as a reference and reconstructs equally spaced sampling points through Lagrange interpolation. The interpolation algorithm adopts a three-point interpolation mode, constructing a quadratic polynomial function with three adjacent raw data points at the target time, and dynamically fitting the theoretical wavelength value at that time. This processing effectively eliminates millisecond-level time jitter caused by differences in transmission paths. To address the issue of unifying the spatial coordinates of the piezoelectric accelerometer array, the engine reads a pre-stored sensor deployment coordinate table. This table records the three-dimensional coordinates of each accelerometer in the global coordinate system and its orientation angle in the local coordinate system. Using the exponential mapping relationship in Lie group theory and Lie algebra theory, the acceleration vector in the local coordinate system is rotated to the global direction. The transformation matrix is ​​calculated from the rotation vector corresponding to the sensor's installation orientation, ensuring that vibration data from different locations can be vector-superimposed and compared. The slowly varying signals from the electrochemical impedance spectroscopy sensor are synchronized using a sliding window Kalman filter. The filter's state variables are set as a binary pair of chloride ion concentration and polarization resistance. The observation equation is established based on the sensor output, and the process noise covariance is adaptively adjusted based on historical data statistics. The filter advances at a fixed time step, independently running a prediction-update loop for each sensor channel, ultimately outputting a parameter sequence with strictly aligned timestamps.

[0031] The semantic modeling unit begins with the semantic parsing of the original feature tensor. The component segmentation module loads the BIM model or CAD design drawings of the structure and parses the geometric information of the solid components. This module extracts the geometric boundaries of components such as beams, columns, and slabs using boundary representation, generating 3D bounding boxes as spatial indexes. Sensor measurement points falling within the same bounding box are automatically clustered and merged into the corresponding components. The material property binding module connects to a material parameter database, which stores parameters such as concrete strength grade, rebar type, and protective layer thickness in a relational table format. This module establishes a connection with the database through component numbers, attaching material constitutive parameters to each component feature block. For example, elastic modulus, Poisson's ratio, and compressive strength labels are added to concrete components, while yield strength and fatigue limit labels are added to steel components. The topology encoder constructs connection relationships based on the stiffness method principle in structural mechanics. It generates a topology graph between components based on the node connection information in the design drawings. In the graph, nodes represent the centroids of components, edges represent connection relationships, and the weight of the edges is assigned by the reciprocal of the connection stiffness, ultimately forming a sparse stiffness matrix and converting it into an adjacency list format to embed feature data.

[0032] The spatiotemporal alignment engine incorporates a detailed anomaly handling mechanism. When a data stream interruption from a sensor is detected, the engine initiates a data reconstruction procedure. Based on spatiotemporal kriging interpolation, it estimates the missing data using measurements from surrounding normal sensors. The interpolation weights are determined by both the distance function and the correlation coefficient. For coordinate deviations caused by sensor deployment errors, the system provides a manual calibration interface, allowing engineers to input actual measured coordinates to override the initial settings. The clock synchronization protocol uses an improved version of IEEE 1588v2, reducing software stack latency through hardware timestamps. The master clock periodically exchanges bidirectional time messages with each acquisition node to calculate and compensate for transmission delays.

[0033] The material parameter library of the semantic modeling unit adopts a version management mechanism. When a structure has undergone reinforcement and repair, a new version of material parameters can be created and its effective date can be marked. The component segmentation algorithm supports multi-scale partitioning. For large continuous structures such as box girder structures, it can be further divided into several virtual sub-components to improve analysis accuracy. The topology encoder also considers nonlinear connections. For special components such as seismic isolation bearings, their connection stiffness is expressed as a function of displacement rather than a fixed value. The unit output adopts the standard JSON-LD format. Each feature block contains three semantic tags: spatial coordinates, material properties, and topological relationships, forming a machine-readable engineering information model. In the specific data processing pipeline, the raw sensor data packets first enter the time alignment queue. The queue manager sorts the packets according to the timestamps and triggers the interpolation recalculation process for late data. The spatial coordinate transformation module uses quaternions instead of Euler angles to avoid gimbal lock problems. The transformed data is uniformly converted to the North-East-Ground coordinate system commonly used in engineering. The semantic mapping process adopts a parallel processing architecture. The feature extraction tasks of different components are distributed to multiple computing cores for synchronous execution. The data consistency check module verifies the geometric relationship logic, such as ensuring that the coordinates of common nodes of connected components are consistent, and automatically triggers a manual review process when a conflict is found.

[0034] The system also includes metadata management functionality, recording parameter configurations and data traceability information for each processing step. The spatiotemporal alignment engine can be configured with different interpolation strategies; spline interpolation is used for vibration data to maintain waveform smoothness, while linear interpolation is chosen for slowly varying parameters to reduce computational overhead. The semantic modeling unit supports incremental update mode; when new sensor nodes are added during monitoring, the system can dynamically expand component feature blocks without re-initializing the entire model. The stiffness matrix output by the topology encoder uses a block storage format, facilitating direct access by subsequent finite element analysis software. When missing coordinate information for certain sensors is detected, the system automatically estimates their approximate location based on signal correlation and prompts for confirmation. The material parameter library has an integrity verification mechanism; missing required parameters prevent feature blocks from being passed downstream. The time synchronization system establishes a hierarchical time synchronization structure, with the regional master clock synchronizing with several sub-domain clocks. Sensors within sub-domains use the simpler NTP protocol to reduce communication overhead. Data timestamps include time zone information and leap second markers, ensuring the temporal continuity of long-term monitoring data. A bidirectional index is established between the semantic model version and monitoring data, supporting backtracking analysis of state at any historical point in time.

[0035] Example 3: The adaptive fusion core receives hierarchical structural features from the semantic modeling unit. This feature data block contains spatiotemporally aligned sensor readings, along with additional component material properties and topological relationships. The first step in the core processing is to construct a conflict detection model based on Mahalanobis distance. This model uses historical datasets collected under structural health baseline conditions as a reference distribution, calculates the Mahalanobis distance between each current feature vector and the mean of the reference distribution, and compares the distance values ​​with a preset threshold after weighting by the inverse covariance matrix. When the distance value of a feature vector exceeds a critical value determined by the chi-square distribution, the data block is marked as potentially anomalous data. The Gaussian mixture model then performs weight estimation on the reliability of the multimodal sensor data. The model treats the data stream of each sensor as a Gaussian distribution component and iteratively solves for the mean, covariance, and mixing coefficients of each component using the expectation-maximization algorithm. The weight of each data point is determined by its posterior probability of belonging to each component. Anomalous data points are assigned lower weight coefficients because they deviate significantly from the main distribution component. Variational autoencoders perform dimensionality reduction on the weighted reconstructed features. The encoder part, composed of fully connected layers, compresses high-dimensional features into low-dimensional latent variables. The decoder part attempts to reconstruct the denoised features from the latent variables. Its loss function... It is composed of the reconstruction error and the KL divergence between the latent variable distribution and the standard normal distribution: in: Representing variables in the latent space, It is the input feature vector. It is the approximate posterior distribution of the encoder output. It is the prior distribution (standard normal distribution). It is the conditional likelihood defined by the decoder. Denotes KL divergence, This is the expectation operator. By minimizing this loss function, the model learns the low-dimensional manifold structure of the data's essence, ultimately outputting a robust fusion diagnostic metric.

[0036] Edge computing nodes are deployed in cabinets near the structures at the monitoring site. Their microcrack detection unit employs a depthwise separable convolutional network architecture. The network's input layer receives strain field distribution maps from fused diagnostic indicators. First, it independently processes strain data from each channel through deep convolutional layers to extract spatial features. Then, it performs point-by-point convolution to fuse cross-channel information, outputting the crack risk probability for each grid cell. The stiffness degradation assessment unit calculates based on vibration spectrum features and the connection stiffness matrix provided by the semantic modeling unit. This unit obtains the theoretical vibration frequency of the structure by solving a generalized eigenvalue problem, compares the measured frequency with the theoretical frequency, and uses the modal participation factor to calculate the equivalent stiffness reduction factor for each component. The corrosion risk quantification unit integrates electrochemical monitoring data and environmental parameters. This unit solves unsteady-state diffusion equations to simulate the transport process of chloride ions in concrete. The boundary conditions of the equations are determined by the measured surface concentration from the sensors. Simultaneously, it derives the corrosion current density using the Stern-Gehry formula based on the polarization resistance value, ultimately outputting a spatiotemporal distribution prediction of the steel corrosion rate. The implementation of the adaptive fusion core includes a dynamic weight update mechanism. After the monitoring system has been running for a period of time, the parameters of the Gaussian mixture model need to be recalibrated. The system periodically collects new health status data and updates the distribution parameters of each component through an online learning algorithm to avoid false alarms caused by the natural degradation of the model's structural performance. The conflict detection module has an adaptive threshold adjustment function. When the ambient temperature changes drastically, the system temporarily relaxes the Mahalanobis distance threshold to reduce false alarms. The variational autoencoder training process uses mini-batch gradient descent. Each time the edge computing nodes are dormant, the network parameters are fine-tuned using local historical data to maintain the model's adaptability to changes in structural state.

[0037] The microcrack detection network on the edge computing nodes employs a transfer learning strategy. Initial weights are pre-trained on strain field data generated from a large amount of finite element simulation, and then adapted to actual structural characteristics through online learning after on-site deployment. The stiffness degradation assessment algorithm considers material nonlinear effects and automatically switches to an iterative algorithm based on tangent stiffness to improve accuracy when large deformations are detected. The corrosion prediction model integrates environmental temperature and humidity sensor data, dynamically adjusting the temperature correction factor in the diffusion coefficient to improve prediction accuracy for different seasons. Data flow processing at the edge follows strict time-series control; the adaptive fusion core performs a complete data fusion cycle every five minutes, including data quality checks, conflict detection, weight calculation, and feature reconstruction. The three evaluation units on the edge computing nodes operate in a pipelined parallel manner. After processing a batch of data, the microcrack detection unit immediately passes it to the stiffness evaluation unit while simultaneously receiving the next batch, maximizing the utilization of computing resources. All intermediate results are timestamped and stored in a local circular buffer for on-demand retrieval by the cloud-based collaborative analyzer.

[0038] The system design takes into account computational resource constraints. The latent variable dimension of the variational autoencoder is dynamically adjusted according to the complexity of the components, using an 8-dimensional latent space for simple components and a 32-dimensional latent space for complex node regions. The number of layers in the depthwise separable convolutional network is also configurable. A lightweight 6-layer network is used for routine monitoring, and when an anomaly is detected, it automatically switches to a 12-layer fine-grained network for detailed analysis. The mesh generation accuracy of the corrosion prediction model is matched with the sensor density to ensure that each mesh contains at least one monitoring point. In terms of fault tolerance, when a sensor fails for a long time, resulting in data loss, the adaptive fusion core automatically adjusts the number of components in the Gaussian mixture model and redistributes the weight coefficients. Edge computing nodes have a degraded operation mode, so that when an evaluation unit fails, the system can still make a status judgment based on the output of the remaining units. All algorithm modules have a watchdog timer to monitor the execution time. Tasks that time out are terminated and restarted to ensure system real-time performance. The data consistency check module verifies the physical rationality of the fusion diagnostic indicators, such as the strain concentration coefficient should not exceed the theoretical value corresponding to the material's yield limit. When an anomaly is detected, a data re-fusion process is triggered.

[0039] Example 4: Taking a long-span steel box girder bridge as an application background, the cloud platform receives component health status data uploaded from edge computing nodes of bridge towers, box girders, and hangers. This data is continuously updated every five minutes. The cloud-based collaborative analyzer first constructs a finite element model of the entire bridge. The model mesh uses a hybrid scheme of tetrahedral and shell elements. The steel box girder is discretized using shell elements, and the bridge tower is simulated using beam elements. The mesh size is set to 0.5 meters according to the required calculation accuracy. The local component health status level is used as a boundary condition input to the model. For example, when an edge node detects that the stiffness degradation level of a bridge deck is level two, the elastic modulus of the corresponding area is reduced by 15%; when the corrosion level of a hanger reaches level three, its element cross-sectional area is reduced by 10% accordingly. The time series prediction network uses a gated cyclic unit architecture to process historical monitoring data. The network input includes a time series of parameters such as stress, strain, and vibration frequency over the past 30 days, and the output is the stress redistribution under different load conditions for the next 7 days. The load case combinations, as required by the specifications, include various combinations of dead load, highway live load, wind load, and temperature effects. During network training, health monitoring data from the first three years after the bridge's construction were used as the training set. A data augmentation strategy involving randomly discarding some sensor channels was employed to improve generalization ability. The Monte Carlo sampling process assesses structural reliability by generating a random set of parameters. Each sampling randomly generates variables such as material strength and load magnitude. Finite element analysis is used to calculate the structural response, and the frequency of stress exceeding limits in 100,000 samplings is statistically analyzed to ultimately obtain a confidence interval estimate of the remaining life.

[0040] The dynamic update interface automatically corrects the finite element model based on newly acquired hierarchical structural features. When a systematic deviation is detected between the model's predicted values ​​and actual measurements, the interface initiates a parameter sensitivity analysis program. This program observes changes in output by perturbing the input parameters of the finite element model, identifying constitutive parameters most sensitive to structural response, such as the elastic modulus of concrete and the yield strength of steel. The incremental learning engine uses an online learning algorithm to adjust model parameters. Each time new monitoring data is received, the engine calculates the residual between the predicted strain field and the actual measured values, and fine-tunes the material parameter matrix of the finite element model using the gradient descent method. The adjustment magnitude is proportional to the residual size. See Table 1, which shows the component health status data uploaded by the edge nodes: Table 1: Health Status Monitoring Data Records of Steel Box Girder Bridge Components The feedback adjustment loop continuously compares the strain distribution predicted by the finite element model with the actual sensor measurements. The residual distribution statistics module calculates the root mean square error at each measurement point, and this value is statistically analyzed weekly. When the residual in a certain region is consistently large, the adaptive learning rate scheduler increases the update step size of the material parameters for the corresponding element in that region, accelerating the model correction process; conversely, for well-fitted regions, the learning rate is decreased to maintain stability. This region-differentiated learning strategy enables the model to quickly adapt to local changes in structural performance.

[0041] In the specific operational process, the cloud platform performs a complete lifespan prediction analysis every morning. First, it aggregates component health data from the past 24 hours from edge nodes to update the boundary conditions of the finite element model. The time series prediction network then runs, simulating stress redistribution under conditions such as standard fleet loads and strong winds, focusing on the changing trends of stress concentration areas. Monte Carlo sampling is executed in parallel on the cloud computing cluster, with each computing node processing one thousand sampling analyses. Finally, the failure probability statistics of all nodes are summarized. A dynamic update interface is activated after each lifespan prediction, using the latest monitoring data to correct model parameters. The corrected model is then used for the next round of prediction, forming a closed-loop optimization.

[0042] The system has a fault-tolerant mechanism for abnormal data. When data uploaded by an edge node significantly contradicts that of other nodes, the cloud platform marks the data and initiates a review process. During the review, the system temporarily uses historical averages to replace outliers for analysis and sends a verification notification to maintenance personnel. The model version management function records detailed information for each parameter adjustment and supports rollback to any historical version for analysis and comparison. The data visualization interface displays the structural stress distribution and remaining life prediction results in the form of a 3D color cloud map, with red areas indicating high-risk areas requiring close attention. During bridge operation, the system detected stress concentration at the junction of the box girder web and bottom plate. The stress level predicted by the finite element model was 8% lower than the measured value. The dynamic update interface, through parameter sensitivity analysis, found that the material constitutive relationship in the weld area significantly affected the calculation results. Therefore, the elastoplastic model parameters were adjusted for the elements in this area. After adjustment, the deviation between the model prediction and the measured value was reduced to within 2%, and the life prediction results showed that the fatigue life in this area was shortened by approximately 15% compared to the initial estimate. This dynamic correction capability enables the system to track gradual changes in structural performance, providing a more accurate basis for maintenance decisions.

[0043] The system also integrates environmental monitoring data, incorporating temperature gradients and humidity changes as additional loads into the analysis. During seasonal transitions, structural deformation caused by temperature changes significantly affects stress distribution; the model considers this effect through thermodynamic coupling analysis. Long-term monitoring has revealed that bridge hanger stress generally increases by 3-5% during the high-temperature period in summer. The system automatically incorporates a temperature compensation factor into its predictions to eliminate the interference of environmental factors on life assessment. All analysis results generate structured reports, which are transmitted to the maintenance management department's information system via a secure interface. The cloud-based collaborative analyzer adopts a microservice architecture, with modules such as finite element analysis, time series forecasting, and reliability calculation deployed independently and exchanging data via message queues. This design allows for dynamic resource expansion based on computational load, automatically launching more computing nodes when large-scale Monte Carlo sampling is required. Data storage employs a hybrid approach of time-series and relational databases; monitoring data is stored in the time-series database for rapid querying, while analysis results and model parameters are stored in the relational database to ensure transactional consistency. The system generates weekly health diagnostic reports, automatically marking components with performance degradation exceeding limits and their impact on overall safety.

[0044] Example 5: The parameter sensitivity analysis module of the dynamically updated interface uses the variance-based Sobol method for global sensitivity analysis. This method generates input parameter combinations through Monte Carlo sampling and observes the changes in the output response variance. In specific implementation, the module sets material parameters of the finite element model, such as the elastic modulus of concrete, the yield strength of steel reinforcement, and Poisson's ratio, as input variables. Each parameter is uniformly sampled within its possible value range. The output variance of the structural reliability index is calculated by performing multiple finite element analyses. Then, the first-order sensitivity index and the overall sensitivity index of each parameter are calculated. The closer the index value is to 1, the greater the influence of the parameter on the output result. The analysis process uses Saltelli sampling sequences to generate input samples. This sampling strategy can obtain stable sensitivity index estimates with a small number of samples. The system automatically performs a full-parameter sensitivity analysis periodically (e.g., monthly) to identify the key material parameters that have the greatest impact on the reliability of the current structural system. The incremental learning engine updates the Young's modulus matrix of the finite element model using an online backpropagation algorithm. This engine constructs the finite element analysis process as a differentiable computational graph, ensuring that the mapping relationship between material parameters and displacement response has a computable gradient. When new monitoring data arrives, the engine first runs finite element analysis to obtain the predicted displacement field, then calculates the loss function between the predicted value and the actual sensor measurement value. This function is typically expressed as mean square error. Next, the backpropagation algorithm is used to calculate the gradient of the loss function with respect to each element value in the Young's modulus matrix. Finally, a stochastic gradient descent algorithm with kinetic momentum is used to update the modulus values. Physical constraints are applied to the update process to ensure that the Young's modulus remains within a reasonable range allowed by materials science. Simultaneously, the update magnitude is limited by a trust region strategy, ensuring that the change in the modulus of each element in a single iteration does not exceed 5% of the current value, preventing over-adjustment that could lead to model instability.

[0045] The residual distribution statistics module of the feedback control loop employs a sliding time window strategy to calculate the prediction error of each component node. The window size is dynamically adjusted according to data characteristics; a shorter 24-hour window is used for rapidly changing data such as vibration frequency, while a 30-day window is used for slowly varying parameters such as corrosion rate. At each time step, the module calculates the residual between the finite element model prediction and the actual sensor measurement, statistically analyzes the residual sequence of each component node within the window period, calculates its root mean square value, and normalizes the result to the 0-1 interval. The statistical process considers the spatiotemporal correlation of the data, performing cluster analysis on spatially adjacent nodes to group nodes with similar residual patterns into the same region, and calculating representative statistics for each region. The adaptive learning rate scheduler dynamically adjusts the update stride of the convolutional kernel parameters based on the root mean square value of the residuals. The scheduler maintains a learning rate matrix corresponding to each finite element mesh node, initially set to the global learning rate. At the beginning of each training cycle, the scheduler reads the normalized root mean square values ​​of each node from the residual statistics module. For node regions with large residuals, the learning rate is increased proportionally to accelerate parameter convergence in that region; for node regions with small residuals, the learning rate is decreased accordingly to improve parameter stability. The learning rate adjustment employs a smooth transition strategy, limiting the learning rate changes between adjacent nodes to a preset range to avoid drastic oscillations. Hard boundaries are set for the upper and lower limits of the learning rate to prevent gradient explosion or vanishing gradient problems.

[0046] In practical applications of bridge monitoring, the system found that the stress prediction residuals in the web region of the box girder were consistently higher than those in other areas. Parameter sensitivity analysis showed that the material constitutive parameters of the welded joints in this region significantly affected the calculation results. The incremental learning engine specifically adjusted the Young's modulus values ​​of the elements in this region, gradually narrowing the gap between prediction and measurement through multiple iterations. The feedback adjustment loop also detected that the root mean square value of the residuals in the web region was 2.3 times that of the bridge deck region. Therefore, it automatically increased the learning rate of this region to 1.8 times the baseline value, while decreasing the learning rate of the well-fitted bridge deck region to 0.7 times the baseline value. The system also has a learning rate decay mechanism. As the training rounds increase, the global base learning rate gradually decays exponentially, avoiding parameter oscillations near the optimal solution due to excessively high learning rates in the later stages of training. The decay rate is adaptively adjusted according to the overall residual change trend. When the system detects that the residual decline has entered a plateau period, it accelerates the decay rate to promote convergence. In special cases, such as after structural repair and reinforcement, the system resets the learning rate to the initial value and restarts to adapt to the changed structural state.

[0047] The dynamic update interface incorporates a parameter update review mechanism. When the parameter modification suggestions proposed by the incremental learning engine exceed the historical variation range, the system pauses automatic updates and generates a task awaiting review. Execution only occurs after engineer confirmation. The review threshold is dynamically set based on the historical variation range of the parameter, typically twice the standard deviation of the parameter's variation range over the past year. This review mechanism prevents erroneous parameter updates due to sensor anomalies or model defects, improving system reliability. The feedback adjustment loop includes an abnormal residual detection function. When the residual of a node suddenly increases sharply, the system marks the node and checks the sensor's operating status, ruling out spurious residuals caused by equipment failure. Simultaneously, the loop analyzes the spatial distribution pattern of the residuals. If adjacent nodes exhibit similar residual characteristics, it indicates that local model parameters need adjustment; if the residual distribution shows a random scattered pattern, it is more likely caused by measurement noise, and the learning rate adjustment range is reduced accordingly.

[0048] During system operation, the parameter update history is fully recorded and visualized, allowing engineers to view the curves of various material parameters over time and determine the trend of structural performance degradation. The learning rate adjustment log is also recorded in detail, providing a data analysis foundation for system optimization. Long-term operational data shows that this adaptive update strategy gradually improves the model's prediction accuracy as monitoring time increases, especially when the structure experiences seasonal temperature changes and traffic load fluctuations, enabling rapid tracking of changes in structural response characteristics. The implementation also considered optimizing computational efficiency. Parameter sensitivity analysis employs a parallel computing strategy, allocating different parameter combinations to multiple computational cores for simultaneous finite element analysis. The incremental learning process uses mini-batch gradient descent, with each update based on data within a time window rather than single-point data, improving the stability of parameter updates. The residual statistics module uses an incremental calculation algorithm, avoiding recalculating statistics for the entire time window each time, reducing computational overhead. These optimizations enable the system to efficiently process monitoring data from large structures on conventional server hardware.

[0049] Example 6: Taking a cross-river immersed tunnel as an application scenario, this tunnel is composed of multiple reinforced concrete immersed tubes joined together. Long-term monitoring of the structural integrity of the tube sections, the sealing status of the joints, and the impact of surrounding water and soil pressure is required. A high-density sensor network is deployed segment by segment. Fiber grating sensors are installed in a matrix along the circumferential and longitudinal directions on the inner wall of the immersed tubes, covering key areas such as the middle, ends, and joints of the tube sections to collect strain field distribution data. Piezoelectric accelerometer arrays are deployed on the water-facing side and bottom of the outer wall of the tube sections to capture the vibration spectrum caused by ship navigation and water flow impact. Electrochemical impedance sensors are pre-embedded at the bottom of the immersed tubes and at the joints to monitor the chloride ion concentration in the concrete and the polarization resistance of the reinforcing steel. Simultaneously, environmental temperature and humidity sensors are installed at the top of the tunnel to assist in analyzing the environmental impact on the structure.

[0050] After receiving asynchronous sampling data from various sensors, the spatiotemporal alignment engine uses Lagrange interpolation to compensate for time offsets caused by differences in signal transmission distance within the immersed tube for the wavelength sampling sequence of the fiber Bragg grating sensor. Based on the coordinate system in the tunnel design drawings, a spatial position relationship diagram of the sensors is established, and the local coordinate system of the piezoelectric accelerometer array is transformed to the tunnel global reference system through Lie algebra transformation to ensure spatial consistency of vibration data at different locations. For the slowly varying signals output by the electrochemical impedance sensor, a sliding window Kalman filter is used for time synchronization calibration to eliminate the interference of ambient temperature fluctuations on the sampling timing, and finally, the original feature tensor of the unified spatiotemporal reference is generated.

[0051] The semantic modeling unit calls upon the BIM design drawings and material parameter library of the immersed tunnel. The component segmentation module divides the original feature tensor into corresponding feature blocks such as single immersed tunnel sections and tunnel joints based on information such as pipe segment division and joint structure in the drawings. The material attribute binding module extracts parameters such as the concrete grade, steel bar diameter and protective layer thickness of the immersed tunnel from the parameter library and adds material attribute labels such as elastic modulus and yield strength to each feature block. The topology encoder calculates the connection stiffness matrix of the joints between pipe sections based on the tunnel structural mechanics model and embeds it into the hierarchical structural features to clarify the mechanical relationship between each pipe section and the joint.

[0052] After receiving hierarchical structural features, the adaptive fusion core constructs a conflict detection model based on Mahalanobis distance to identify anomalous data blocks that exceed the statistical confidence interval, such as abnormal chloride ion concentration readings caused by sensor dampness. The reliability of multimodal sensor data is analyzed using a Gaussian mixture model, and dynamic weights are assigned to different types of data, such as strain, vibration, and corrosion, to perform weighted reconstruction of anomalous data blocks. A variational autoencoder is used to project the reconstructed features onto a low-dimensional manifold, generating anti-interference fusion diagnostic indicators that primarily reflect the strain concentration of pipe sections, vibration frequency changes, and the risk of steel reinforcement corrosion.

[0053] Edge computing nodes are deployed in monitoring rooms along the tunnel, running a lightweight damage detection model based on fused diagnostic indicators. The microcrack detection unit analyzes the strain concentration coefficient of the pipe sections using a depth-separable convolutional network to identify the presence of microcracks in the inner wall; the stiffness degradation assessment unit combines vibration spectrum characteristics with the joint connection stiffness matrix to calculate the equivalent stiffness attenuation rate of the pipe sections and joints, judging the changes in structural bearing capacity; the corrosion risk quantification unit predicts the steel corrosion rate based on the chloride ion concentration and polarization resistance value measured by the electrochemical impedance sensor, using the chloride ion diffusion equation, and outputs the local health status level of each pipe section and joint.

[0054] The cloud-based collaborative analyzer aggregates the health status levels of local components uploaded by various edge computing nodes, and combines this with historical degradation data since the construction of the immersed tunnel to construct an overall finite element model of the tunnel. The health status levels of each pipe section and joint are used as boundary conditions input into the model. A time series prediction network is used to simulate stress redistribution under different working conditions, such as stress changes in the tunnel structure during high water levels in the flood season and when ships are heavily loaded. The reliability index of the tunnel structure system is calculated through Monte Carlo sampling to analyze the probability of risks such as pipe section cracking and joint leakage, and finally outputs the confidence interval of the overall remaining life of the tunnel.

[0055] The system's dynamic update interface periodically corrects the material constitutive relations of the finite element model based on newly acquired hierarchical structural features. The parameter sensitivity analysis module identifies the material parameters most impacting tunnel structural reliability, such as the elastic modulus of concrete and the corrosion resistance of reinforcing steel. The incremental learning engine updates the Young's modulus matrix of the corresponding pipe sections in the finite element model using an online backpropagation algorithm, enabling the model to reflect the gradual changes in structural performance. The feedback adjustment loop continuously compares the residuals between the strain field predicted by the finite element model and the actual measured values. The residual distribution statistics module calculates the root mean square (RMS) value of the prediction error for each pipe section node. The adaptive learning rate scheduler dynamically adjusts the kernel parameters and step size of the depthwise separable convolutional network based on the RMS value. When the residual at the pipe section joint is large, the parameter adjustment amplitude is increased to improve the accuracy of microcrack detection.

[0056] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0057] 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 health monitoring system for civil engineering structures, characterized in that, include: High-density sensor networks are used to collect strain field distribution, vibration spectrum and environmental corrosion parameters of structures through distributed multimodal sensor arrays; The spatiotemporal alignment engine performs clock drift compensation and spatial coordinate normalization on the asynchronous sampling data of the multimodal sensor array, generating an original feature tensor with a unified spatiotemporal reference. The semantic modeling unit maps the original feature tensors to a preset component semantic space based on the structural design drawings and material parameter library, and outputs hierarchical structural features with topological tags. The adaptive fusion core receives hierarchical structural features and performs dynamic weight allocation. It eliminates conflicting data between sensors through iterative optimization algorithms and generates anti-interference fusion diagnostic indicators. Edge computing nodes run lightweight damage detection models in real time based on fused diagnostic indicators and output the health status level of local components. The cloud-based collaborative analyzer aggregates the outputs of multiple edge computing nodes and combines them with historical degradation data to predict the overall remaining lifespan of the structure.

2. The health monitoring system for civil engineering structures according to claim 1, characterized in that, The high-density sensor network includes: The strain field acquisition subnetwork uses a fiber optic grating sensor array to measure the micro-strain on the surface of the structure and calculates the principal stress direction by Bragg wavelength offset. The vibration sensing subnet captures the structure’s natural frequency and damping ratio through a piezoelectric accelerometer array and extracts the frequency band energy features using wavelet packet decomposition. The corrosion monitoring subnetwork uses electrochemical impedance sensors to detect the chloride ion concentration and steel reinforcement polarization resistance in concrete.

3. The health monitoring system for civil engineering structures according to claim 2, characterized in that, The spatiotemporal alignment engine executes: Lagrange interpolation compensation is performed on the wavelength sampling sequence of the fiber Bragg grating sensor array to eliminate the time offset caused by transmission delay; Establish a spatial position relationship diagram of the sensors, and transform the local coordinate system of the piezoelectric accelerometer array to the global reference system through Lie algebra transformation; A sliding window Kalman filter was used to perform time synchronization calibration on the slowly varying signal of the electrochemical impedance sensor.

4. The health monitoring system for civil engineering structures according to claim 3, characterized in that, The semantic modeling unit includes: The component segmentation module divides the original feature tensor into feature blocks corresponding to the physical components based on the beam-column node information in the design drawings. The material property binding module extracts concrete grade and steel reinforcement specifications from the material parameter library and adds elastic modulus and yield strength labels to each feature block. The topology encoder generates the connection stiffness matrix between components based on the structural mechanics model and embeds it into the hierarchical structural features.

5. A health monitoring system for civil engineering structures according to claim 4, characterized in that, The adaptive fusion core execution: Construct a conflict detection model based on Mahalanobis distance to identify anomalous data blocks that exceed the statistical confidence interval in hierarchical structural features; The reliability weights of multimodal sensor data are estimated using a Gaussian mixture model, and abnormal data blocks are reconstructed using weighted methods. A variational autoencoder is used to perform low-dimensional manifold projection on the reconstructed features to generate fusion diagnostic indicators.

6. A health monitoring system for civil engineering structures according to claim 5, characterized in that, The edge computing nodes include: The microcrack detection unit analyzes and fuses the strain concentration coefficient in diagnostic indicators using a depth-separable convolutional network. The stiffness degradation assessment unit calculates the equivalent stiffness attenuation rate of the component based on the vibration spectrum characteristics and the connection stiffness matrix. A corrosion risk quantification unit predicts the corrosion rate of steel bars based on the chloride ion diffusion equation and polarization resistance value.

7. A health monitoring system for civil engineering structures according to claim 6, characterized in that, The cloud-based collaborative analyzer executes: Construct a finite element model of the entire structure and use the health status level of local components as boundary condition input; A time series prediction network was used to simulate the stress redistribution process under different loading conditions. The reliability index of the structural system is calculated by Monte Carlo sampling, and the confidence interval of the remaining lifetime is output.

8. A health monitoring system for civil engineering structures according to claim 7, characterized in that, Also includes: The interface is dynamically updated to automatically correct the material constitutive relations of the finite element model based on newly acquired hierarchical structural features; The feedback adjustment loop compares the residuals between the predicted strain field and the actual measured value, and adjusts the convolution kernel parameters of the depth-separable convolutional network.

9. A health monitoring system for civil engineering structures according to claim 8, characterized in that, The dynamic update interface includes: The parameter sensitivity analysis module identifies the material parameters that have the greatest impact on structural reliability. The incremental learning engine updates the Young's modulus matrix of the finite element model through an online backpropagation algorithm.

10. A health monitoring system for civil engineering structures according to claim 9, characterized in that, The feedback adjustment loop includes: The residual distribution statistics module calculates the root mean square value of the prediction error for each component node; An adaptive learning rate scheduler dynamically adjusts the update stride of the convolution kernel parameters based on the root mean square value.

Images