Civil engineering structure monitoring data fusion processing and visual analysis method based on cloud computing architecture

Abstract

The invention discloses a civil engineering structure monitoring data fusion processing and visual analysis method based on a cloud computing architecture. In the invention, through standardized integration and dynamic cleaning of multi-source heterogeneous data, in a data access stage, an edge node carries out real-time preprocessing on vibration strain displacement and other multi-type sensor data, and in combination with a moving average filtering and abnormal value elimination mechanism, the stability of original data is ensured; after entering the cloud, heterogeneous data output by different devices are integrated into a standard column storage structure through unified format conversion and unit calibration, and meanwhile, missing values and sensor drift errors are processed by adopting linear interpolation and drift correction technologies, so that the data quality is remarkably improved. The hierarchical processing mode not only retains the real-time advantage of the edge end, but also realizes deep cleaning of large-scale data by means of cloud computing power, so that a more reliable data basis is obtained for subsequent fusion analysis, and structural state misjudgment caused by data exception is avoided as much as possible.

Description

Technical Field

[0001] This invention belongs to the field of civil engineering structure monitoring technology, specifically a method for data fusion processing and visualization analysis of civil engineering structure monitoring based on cloud computing architecture. Background Technology

[0002] Civil engineering structural monitoring is a systematic engineering approach that utilizes modern sensing, testing, and data analysis technologies to conduct long-term or periodic measurements and assessments of the stress, deformation, vibration, and environmental effects of critical infrastructure such as bridges, dams, tunnels, and high-rise buildings during construction and operation. By deploying equipment such as strain gauges, displacement sensors, and accelerometers, it acquires structural response data in real time and combines this with theoretical analysis and model simulation to understand the actual working performance of the structure, promptly detect damage and anomalies, and provide early warnings of potential risks. This technology provides a scientific basis for structural safety assessment, maintenance decisions, life prediction, and design verification, and is a core technical means to ensure the safe and reliable operation of major projects throughout their entire life cycle. Civil engineering structural monitoring data processing, through the establishment of data-driven models or comparison with finite element simulation results, allows the processed data to be used to assess the structural health status, provide early warnings of potential damage, verify design theories, and guide maintenance decisions. It is a core technical link in realizing intelligent operation and maintenance and full life-cycle safety management of structures, providing scientific assurance for the long-term safe operation of major infrastructure.

[0003] However, in traditional monitoring methods, data from multiple sensors often face integration difficulties due to format differences and inconsistent units. Noise interference processing relies on simple filtering, which is insufficient to effectively remove outliers, resulting in inadequate data quality and stability. Centralized storage architectures, which store high-frequency real-time data alongside historical data, waste storage resources. Simultaneously, data transmission delays affect real-time monitoring response efficiency, thereby impacting engineers' accurate assessment and rapid decision-making regarding the overall structural condition. Summary of the Invention

[0004] The purpose of this invention is to provide a method for data fusion processing and visualization analysis of civil engineering structure monitoring based on cloud computing architecture in order to solve the problems mentioned above.

[0005] The technical solution adopted in this invention is as follows: a method for data fusion processing and visualization analysis of civil engineering structure monitoring based on cloud computing architecture, characterized in that: the method includes the following steps: S1: Deploy edge computing nodes to connect with various sensors, drone inspection equipment, and manual recording terminals to collect structural vibration, strain, displacement, and environmental temperature and humidity data. Use lightweight filtering algorithms to initially remove extreme noise data and mark timestamps and spatial coordinates according to data type. S2: Based on the timeliness and importance of data, a three-tier storage system is constructed: high-frequency real-time data is temporarily stored on edge nodes, mid-frequency analytical data is stored in a distributed database on the cloud platform, and historical archived data is migrated to a low-cost object storage system; S3: Perform format conversion, unit unification, and missing value imputation on the heterogeneous data stored in S2. Use the sliding window method to identify and correct sensor drift errors. The cleaned standard data is grouped according to the structural feature dimensions required by the S4 fusion module, and the data quality assessment report is fed back to S1. S4: Combine engineering structural design drawings to construct a physical topology network, map the standardized data output by S3 to structural component nodes, and use a spatiotemporal anchoring algorithm to associate the temporal data of different monitoring points of the same component with the spatial response characteristics of different components, and fuse them to generate a feature matrix containing the overall mechanical state of the structure. S5: Based on the fusion feature matrix output by S4, a method combining principal component analysis and structural mechanics feature screening is used to retain key indicators sensitive to structural safety. The data compression rate and feature integrity are balanced by dynamically adjusting the dimensionality reduction. S6: Combining the key features and structural design of the S5 output, a dynamic early warning model is constructed based on the safety threshold. A sliding time window is used to compare the deviation of the current feature vector from the historical health baseline. S7: Based on the monitoring parameters in the S2 metadata table and the anomaly detection results output by S6, design a layered visualization template: the base layer displays the original monitoring curves, the middle layer presents the structural topology heat map after S4 fusion, and the top layer overlays the anomaly warning markers of S6. S8: Initiate the corresponding response process based on the anomaly level detected by S6: Level 1 warning is pushed to the project management terminal, Level 2 warning is linked to the on-site inspection drone, and Level 3 warning triggers the expert remote consultation system. S9: Integrates the visualization results of S7 with the early warning response records of S8 to generate a structural health monitoring report. It uses natural language processing technology to extract typical cases and handling experience from the report, constructs a knowledge graph for structural disease diagnosis, and regularly updates the topology fusion algorithm and feature weight model to form a data-driven continuous optimization closed loop.

[0006] In a preferred embodiment, in step S1, when deploying edge computing nodes, a data acquisition module compatible with mainstream sensor interfaces needs to be configured. The vibration sensor is a piezoelectric accelerometer with a sampling frequency of 50–200 Hz. The strain sensor is a fiber optic grating with a range of -2000–2000 με. The displacement monitoring device is a laser displacement meter with an accuracy controlled at the 0.01 mm level. The sampling interval of the environmental temperature and humidity sensor is set to 10–30 seconds. The edge node hardware configuration needs to meet at least a 4-core processor and 8 GB of memory to support parallel access of multiple devices, with a maximum of 50 devices connected to a single node. In the data preprocessing stage, a moving average filtering algorithm is used, and the window size is dynamically adjusted according to the data type. The vibration data window is set to 5 sampling points, and the strain data window is set to 10 sampling points. By comparing the deviation between the current sampled value and the window mean, extreme noise data that deviates more than 3 times the standard deviation is removed.

[0007] In a preferred embodiment, in step S2, the specific parameter configuration of the three-level storage system needs to be determined in combination with the data characteristics. High-frequency real-time data includes vibration acceleration and dynamic strain data, which are temporarily stored in the local SSD storage of the edge node. The storage capacity is configured as 20-50GB according to the number of connected devices, and the temporary storage time is set to 5-15 minutes. When the storage occupancy rate reaches 80%, the data cleaning mechanism is triggered.

[0008] In a preferred embodiment, in step S3, the data cleaning stage first performs format conversion, uniformly converting the CSV, JSON, and binary formats output by the sensors into Parquet columnar storage format. During the conversion process, data integrity is verified, and files with a missing field ratio exceeding 10% are directly marked as invalid. Units are uniformly standardized according to engineering specifications, with strain units converted to με, displacement to mm, temperature to ℃, and humidity to %RH, and the conversion accuracy retaining two decimal places. Missing values ​​are handled using linear interpolation. When the missing rate of a single data entry is less than 5%, interpolation is performed; if it exceeds 5%, the front-end equipment calibration process is triggered. The calibration cycle is adjusted according to the equipment type, with vibration sensors calibrated once a month and strain sensors calibrated once every two months.

[0009] In a preferred embodiment, in step S4, a physical topology network is constructed by combining the component types and their connection methods in the engineering structural design drawings. The overall structure is decomposed into several basic component nodes, each node corresponding to a structural unit with independent mechanical function, such as a beam unit, column unit, or connection node. An adjacency matrix is ​​used to describe the connection relationship between nodes. The matrix element values ​​are dynamically assigned according to the connection type: 1.0 for rigid connections, 0.6 for hinged connections, 0.3 for sliding connections, and 0 for no connections. At the same time, component material properties, such as elastic modulus and Poisson's ratio, are introduced as weighting coefficients of the matrix.

[0010] In a preferred embodiment, in step S4, when implementing the spatiotemporal anchoring algorithm, firstly, time-axis calibration is performed on the time-domain data of multiple monitoring points on the same topological node. A time synchronization method based on cross-correlation function is used to align monitoring data with different sampling frequencies. By setting a time tolerance threshold, typically 1 / 5 of the sampling period, data points with time deviations exceeding the threshold are removed. Next, a variable-length sliding time window is used, with the window length dynamically adjusted according to the structure's natural vibration characteristics. When the structure's response is stable, the window length is increased to 3 times the natural vibration period; when the response is severe, it is shortened to 1 times the natural vibration period. Time-domain characteristic parameters within the window, such as peak value, mean, and variance, are extracted, and the time evolution coefficients of the characteristic parameters are calculated. It reflects the trend changes in data; for the spatial response characteristics of different topological nodes, the spatial influence weight between nodes is calculated based on the topological adjacency matrix. The weight value is inversely proportional to the topological distance between nodes. The topological distance is defined as the number of components on the shortest connection path between two nodes. At the same time, the weight value is corrected by combining the weighting coefficient of material properties. The temporal characteristic parameters of the same node are fused with the spatial weighted characteristic parameters of adjacent nodes and arranged in the order of topological nodes to generate a feature matrix. The row vectors of the matrix correspond to the topological nodes, and the column vectors correspond to different mechanical characteristic indicators, such as strain amplitude, displacement rate of change, and vibration frequency. This matrix completely contains the mechanical state information of the entire structure, providing a data foundation for subsequent steps. The most innovative formula in the spatiotemporal anchoring fusion process is the formula for calculating the spatiotemporal fusion feature value of topological node i at time t: ; The parameters in the formula are defined as follows: F i (t) represents the fusion feature value of topological node i at time t; β is the time feature weighting coefficient, with a value ranging from 0.4 to 0.6. It is dynamically adjusted according to the stability of the node data, increasing to 0.6 when the data fluctuation is small and decreasing to 0.4 when the fluctuation is large. T i (t) is the time-domain feature vector of node i at time t, which consists of the peak value, mean, and variance within the sliding time window; Ψ i Let i be the set of spatially adjacent nodes of node i, that is, all nodes whose topological distance is not greater than 2; γ ij The spatial influence weight of node i on node j is determined by the topological distance d. i j is calculated. ; Δt is the time delay of mechanical response transmission between nodes, ranging from 0.08 to 0.15 seconds, and is determined based on the stiffness of the component; the greater the stiffness, the smaller the time delay. T j (t-Δt) is the time-domain feature vector of the adjacent node j at time t-Δt.

[0011] In a preferred embodiment, in step S5, the cumulative contribution rate threshold of principal component analysis during feature enhancement is set to 90%, that is, principal components whose sum of feature variances accounts for more than 90% of the total variance are retained. At the same time, sensitive indicators are screened in combination with structural mechanical properties, strain gradient is selected as the strain difference between the mid-span and the support of the component, displacement curvature is calculated as the second difference between adjacent monitoring points, and vibration frequency is extracted as the first 5 natural frequencies. The importance weights of these indicators are initialized through the structural finite element simulation results.

[0012] In a preferred embodiment, in step S6, the health baseline of the dynamic early warning model is constructed using historical data. Monitoring data within 3 months after the structure's completion is selected as the initial baseline, and then updated every 7 to 30 days. The update cycle is adjusted according to data stability. When the data fluctuation amplitude is less than 5% for two consecutive weeks, it is extended to 30 days; otherwise, it is shortened to 7 days. The sliding time window size is set according to the structure's natural vibration period. For structures with a natural vibration period of less than 1 second, the window is set to 1 minute; for those with a natural vibration period of 1 to 5 seconds, it is set to 3 minutes; and for those with a natural vibration period of more than 5 seconds, it is set to 5 minutes. The step size is fixed at 1 / 2 of the window length. The deviation is calculated using Euclidean distance. The modulus is calculated by subtracting the current feature vector from the health baseline vector. The threshold is divided according to the importance level of the structure. The threshold is set to 1.5 for primary structures (such as the main bridge), 2.0 for secondary structures (such as approach bridges), and 3.0 for tertiary structures (such as auxiliary structures). An abnormal event is determined when the deviation exceeds the threshold for 3 consecutive windows or when the deviation exceeds twice the threshold in a single instance.

[0013] In a preferred embodiment, in step S7, the base layer of the layered visualization template displays real-time monitoring curves. The sampling interval is adjusted according to the data type: vibration data at 1 second / point, strain data at 5 seconds / point, and environmental parameters at 30 seconds / point. The curve colors distinguish the data types: blue for strain, green for displacement, and red for vibration. The intermediate layer topology heatmap uses a red-yellow-green three-color mapping feature value. The color thresholds correspond to 60% (green), 80% (yellow), and 100% (red) of the design safety limit. The heatmap update frequency is synchronized with the intermediate frequency data storage cycle, i.e., it is updated once per hour.

[0014] In a preferred embodiment, in step S8, the anomaly level is classified according to the decrease in structural bearing capacity. Level 1 warning corresponds to a decrease in bearing capacity of more than 10%, Level 2 to 5% to 10%, and Level 3 to less than 5%. The level is determined by comparing the feature vector with the design limit value. The response process is strictly controlled in terms of activation time. Level 1 warnings are pushed through the project management terminal APP and SMS notifications, requiring the recipient to confirm within 15 minutes. If no confirmation is received, the notification is automatically escalated to the superior management department. Level 2 warnings are linked to the UAV inspection system. The UAV must complete takeoff preparation within 30 minutes. The inspection route covers the abnormal area and three adjacent components. The flight altitude is controlled at 5 to 10 meters above the structural surface, and the image resolution is not less than 1920×1080 pixels. Level 3 warnings trigger the expert remote consultation system.

[0015] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. In this invention, through standardized integration and dynamic cleaning of multi-source heterogeneous data, during the data access phase, edge nodes perform real-time preprocessing of various types of sensor data, such as vibration, strain, and displacement. Combined with moving average filtering and outlier removal mechanisms, the stability of the original data is ensured. After entering the cloud, through unified format conversion and unit calibration, the heterogeneous data output from different devices is integrated into a standardized columnar storage structure. Simultaneously, linear interpolation and drift correction techniques are used to handle missing values ​​and sensor drift errors, significantly improving data quality. This hierarchical processing mode retains the real-time advantages of the edge while leveraging cloud computing power to achieve deep cleaning of large-scale data, providing a more reliable data foundation for subsequent fusion analysis and minimizing misjudgments of structural states due to data anomalies.

[0016] 2. In this invention, the dynamic response efficiency and overall assessment accuracy of civil engineering structure monitoring are significantly improved through the synergistic effect of the elastic scalability of cloud computing architecture and spatiotemporal fusion algorithm. The cloud-edge collaborative storage architecture stores data in layers according to frequency and importance. High-frequency real-time data is processed rapidly at edge nodes, mid-frequency analysis data is efficiently queried in a distributed database in the cloud, and historical data is archived to low-cost object storage, achieving optimized allocation of storage resources. Combined with standardized data, the spatiotemporal anchoring algorithm dynamically fuses temporal and spatial response features to generate a feature matrix containing the overall mechanical state of the structure. Then, through key feature enhancement and dimensionality reduction optimization, it highlights changes in mechanical indicators of key structural components, achieving closed-loop management from data acquisition to safety decision-making, providing efficient and reliable technical support for the health monitoring and maintenance of civil engineering structures. Attached Figure Description

[0017] Figure 1 This is a schematic diagram illustrating the process principle of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] Reference Figure 1 , A cloud computing-based method for the fusion processing and visualization analysis of civil engineering structural monitoring data includes the following steps: S1: Real-time access and edge preprocessing of multi-source heterogeneous monitoring data. Edge computing nodes are deployed to connect to various sensors, drone inspection equipment, and manual recording terminals to collect data such as structural vibration, strain, displacement, and environmental temperature and humidity. Lightweight filtering algorithms are used to initially remove extreme noise data, and timestamps and spatial coordinates are added according to data type. The preprocessed data is then encrypted and transmitted to the cloud storage module in S2.

[0020] S2: Cloud-edge collaborative data tiered storage architecture. A three-tiered storage system is built based on data timeliness and importance: high-frequency real-time data is temporarily stored on edge nodes, mid-frequency analytical data is stored in a distributed database on the cloud platform, and historical archived data is migrated to a low-cost object storage system. Metadata tables containing data sources and accuracy ranges are generated synchronously, providing format indexes for S3's standardized processing and basic fields for S7's visualization configuration.

[0021] S3: Multimodal data cleaning and standardization integration. Format conversion, unit standardization, and missing value imputation are performed on the heterogeneous data stored in S2. A sliding window method is used to identify and correct sensor drift errors. The cleaned standard data is grouped according to the structural feature dimensions required by the S4 fusion module. Simultaneously, a data quality assessment report is fed back to S1 to optimize the sampling frequency and calibration cycle of the front-end acquisition equipment.

[0022] S4: Spatiotemporal Feature Anchoring and Fusion Based on Structural Topology. A physical topology network is constructed using engineering structural design drawings. The standardized data output from S3 is mapped to structural component nodes. A spatiotemporal anchoring algorithm is used to associate the temporal data of different monitoring points on the same component with the spatial response characteristics of different components, fusing them to generate a feature matrix containing the overall mechanical state of the structure. The topology association parameters extracted during the fusion process are synchronously updated to the metadata table of S2, providing prior structural information for feature enhancement in S5.

[0023] S5: Key Feature Enhancement and Dynamic Dimensionality Reduction Optimization. Based on the fused feature matrix output by S4, a combination of principal component analysis and structural mechanics feature selection is used to retain key indicators sensitive to structural safety. The dimensionality reduction is dynamically adjusted to balance data compression rate and feature integrity. The dimensionality-reduced feature vectors are directly transmitted to the anomaly detection module in S6, while the feature importance ranking results are fed back to S4 to optimize the weight allocation strategy during anchoring fusion.

[0024] S6: Dynamic Threshold Anomaly Detection for Structural State. A dynamic early warning model is constructed by combining the key features output from S5 with the structural design safety threshold. A sliding time window is used to compare the deviation of the current feature vector from the historical healthy baseline. Detected abnormal events trigger the early warning mechanism in S8 in real time, and abnormal samples are automatically stored in the anomaly database of S2 for iterative training of the model in the S5 feature enhancement module.

[0025] S7: Dynamic configuration of multi-dimensional visualization model. Based on the monitoring parameters in the S2 metadata table and the anomaly detection results output by S6, a layered visualization template is designed: the base layer displays the original monitoring curves, the intermediate layer presents the structural topology heatmap fused from S4, and the top layer overlays the anomaly warning markers from S6. It receives user interaction and adjustment commands for the visualization interface and synchronizes configuration preferences to the report generation module in S9.

[0026] S8: Tiered Early Warning and Cross-Platform Collaborative Response. Based on the anomaly level detected by S6, corresponding response procedures are initiated: Level 1 warnings are pushed to the project management terminal; Level 2 warnings trigger on-site inspection drones; and Level 3 warnings trigger the expert remote consultation system. Text records and image data of the warning handling process are stored in real-time in the S2 log database, providing case study material for the comprehensive report in S9.

[0027] S9: End-to-End Data Loop and Knowledge Graph Construction. Integrating the visualization results from S7 with the early warning response records from S8 generates a structural health monitoring report. Natural language processing technology is used to extract typical cases and handling experiences from the report, constructing a structural disease diagnosis knowledge graph. The association rules in the graph are periodically updated to the topology fusion algorithm of S4 and the feature weight model of S5, forming a data-driven continuous optimization loop.

[0028] In step S1, when deploying edge computing nodes, a data acquisition module compatible with mainstream sensor interfaces must be configured. The vibration sensor is a piezoelectric accelerometer with a sampling frequency of 50–200Hz. The strain sensor is a fiber Bragg grating type with a range of -2000–2000με. The displacement monitoring device is a laser displacement meter with an accuracy controlled at the 0.01mm level. The sampling interval for the environmental temperature and humidity sensor is set to 10–30 seconds. The edge node hardware configuration must meet at least a 4-core processor and 8GB of memory to support parallel access by multiple devices, with a maximum of 50 devices connected to a single node. In the data preprocessing stage, a moving average filtering algorithm is used. The window size is dynamically adjusted according to the data type. The vibration data window is set to 5 sampling points, and the strain data window is set to 10 sampling points. By comparing the deviation of the current sampled value with the window mean, extreme noise data deviating more than 3 times the standard deviation is removed. Timestamps are marked with millisecond-level precision, and the spatial coordinates uniformly adopt an engineering independent coordinate system, with coordinate errors controlled within ±5cm. The preprocessed data is encrypted using the AES-128 encryption algorithm, and the transmission rate is maintained at 1-5Mbps to ensure that the data is transmitted from the edge node to the cloud storage module within 5 seconds.

[0029] In step S2, the specific parameter configuration of the three-tier storage system needs to be determined based on the data characteristics. High-frequency real-time data, including vibration acceleration and dynamic strain data, is temporarily stored on local SSDs at edge nodes. The storage capacity is configured to 20-50GB depending on the number of connected devices, and the temporary storage duration is set to 5-15 minutes. A data cleanup mechanism is triggered when the storage occupancy rate reaches 80%. Mid-frequency analysis data, covering hourly statistical average strain values, cumulative displacement values, and environmental parameters, is stored in a distributed database on the cloud platform. The database uses a MySQL cluster, with single-table partitions divided by day. The number of replicas is set to 3 to ensure data redundancy, the size of a single data record is controlled to 1-5KB, and the query response time is required to be no more than 100ms. Historical archived data, including raw data and analysis results for more than 6 months, is migrated to an object storage system using the S3 compatible protocol. The data compression rate is no less than 30%, and the storage cost is controlled to within 50 yuan per TB per month. The metadata table should include fields such as data ID, source device number, collection time, accuracy range ±0.5%FS, and data status flag. The table structure should be updated quarterly based on the new device types to ensure a complete format index for subsequent standardized processing and visualization configuration.

[0030] In step S3, the data cleaning stage first involves format conversion, unifying the CSV, JSON, and binary formats output by the sensors into the Parquet columnar storage format. During the conversion process, data integrity is verified; files with more than 10% missing fields are marked as invalid. Units are standardized according to engineering specifications: strain is converted to με, displacement to mm, temperature to ℃, and humidity to %RH, with conversion accuracy retained to two decimal places. Missing values ​​are handled using linear interpolation. Interpolation is performed when the missing rate of a single data point is below 5%; otherwise, the front-end equipment calibration process is triggered. The calibration cycle is adjusted according to the equipment type: vibration sensors are calibrated every month, and strain sensors every two months. Sensor drift error correction uses a sliding window method with a window length of 30 seconds and a step size of 5 seconds. By calculating the slope of the linear trend of the data within the window, when the absolute value of the slope exceeds 0.5 με / min, it is considered drift and reverse correction is performed. The data quality assessment report includes data integrity rate, noise removal rate, and time synchronization error indicators. The integrity rate must reach above 95%, the noise removal rate must be no less than 90%, and the time synchronization error must be controlled within 10ms. The assessment results are fed back to the edge nodes in real time to optimize the sampling frequency. When the integrity rate is below 90%, the sampling frequency of the corresponding device will be increased by 20%.

[0031] In step S4, a physical topology network is constructed by combining the component types and connection methods in the engineering structural design drawings. The overall structure is decomposed into several basic component nodes, each corresponding to a structural unit with independent mechanical function, such as beam units, column units, and connection nodes. An adjacency matrix is ​​used to describe the connection relationship between nodes. The matrix element values ​​are dynamically assigned according to the connection type: 1.0 for rigid connections, 0.6 for hinged connections, 0.3 for sliding connections, and 0 for no connection. At the same time, component material properties, such as elastic modulus and Poisson's ratio, are introduced as weighting coefficients for the matrix. Various monitoring data are extracted from the standardized data output in S3, including strain time history data collected by strain sensors, displacement sequences recorded by displacement gauges, and acceleration signals from vibration accelerometers. According to the layout of the monitoring points in the structural design drawings, these data are mapped to the corresponding topology nodes. For example, strain monitoring data in the middle of a beam unit is associated with the topology node corresponding to that beam unit, and displacement monitoring data at the top of a column is associated with the topology node corresponding to the column unit, forming a basic data set for the nodes.

[0032] In step S4, when implementing the spatiotemporal anchoring algorithm, the time-domain data of multiple monitoring points on the same topological node are first calibrated on the time axis. A time synchronization method based on the cross-correlation function is used to align monitoring data with different sampling frequencies. By setting a time tolerance threshold, typically 1 / 5 of the sampling period, data points with time deviations exceeding the threshold are removed. Next, a variable-length sliding time window is used. The window length is dynamically adjusted according to the structure's natural vibration characteristics. When the structural response is stable, the window length is increased to 3 times the natural vibration period; when the response is severe, it is shortened to 1 times the natural vibration period. Temporal characteristic parameters within the window, such as peak value, mean, and variance, are extracted, and the time evolution coefficients of the characteristic parameters are calculated to reflect the trend changes in the data. For the spatial response characteristics of different topological nodes, the spatial influence weights between nodes are calculated based on the topological adjacency matrix. The weight value is inversely proportional to the topological distance between nodes. The topological distance is defined as the number of components on the shortest connection path between two nodes. At the same time, the weight value is corrected by combining the material property weighting coefficient. For example, the influence weight of steel component nodes on concrete component nodes needs to be multiplied by a material coupling coefficient of 0.8. The temporal feature parameters of the same node are fused with the spatial weighted feature parameters of adjacent nodes, and the feature matrix is ​​generated by arranging them in the order of the topological nodes. The row vectors of the matrix correspond to the topological nodes, and the column vectors correspond to different mechanical feature indicators, such as strain amplitude, displacement rate of change, vibration frequency, etc. This matrix completely contains the mechanical state information of the entire structure, providing a data foundation for subsequent steps. The most innovative formula in the spatiotemporal anchoring fusion process is the formula for calculating the spatiotemporal fusion feature value of topological node i at time t: ; The parameters in the formula are defined as follows: F i (t) represents the fusion feature value of topological node i at time t; β is the time feature weighting coefficient, with a value ranging from 0.4 to 0.6. It is dynamically adjusted according to the stability of the node data, increasing to 0.6 when the data fluctuation is small and decreasing to 0.4 when the fluctuation is large. T i (t) is the time-domain feature vector of node i at time t, which consists of the peak value, mean, and variance within the sliding time window; Ψ i Let i be the set of spatially adjacent nodes of node i, that is, all nodes whose topological distance is not greater than 2; γ ij The spatial influence weight of node i on node j is determined by the topological distance d. i j is calculated. ; Δt is the time delay of mechanical response transmission between nodes, ranging from 0.08 to 0.15 seconds, and is determined based on the stiffness of the component; the greater the stiffness, the smaller the time delay. T j(t-Δt) is the time-domain feature vector of the adjacent node j at time t-Δt.

[0033] In step S5, the cumulative contribution rate threshold of principal component analysis during feature enhancement is set to 90%, meaning that principal components whose sum of feature variances accounts for more than 90% of the total variance are retained. Simultaneously, sensitive indicators are screened based on structural mechanical properties. The strain gradient is selected from the strain difference between the mid-span and support of the component. The displacement curvature is calculated using the second-order difference between adjacent monitoring points. The vibration frequency is extracted using the first five natural frequencies. The importance weights of these indicators are initialized using the structural finite element simulation results. Dynamic dimensionality reduction is adjusted according to the sparsity of the feature matrix. When the sparsity (the proportion of zero-value elements) is greater than 0.3, the dimensionality is increased to 15-20 dimensions; when it is less than 0.1, it is reduced to 5-10 dimensions, ensuring that the data compression rate is maintained between 30% and 60%. Meanwhile, feature integrity is verified through reconstruction error, with the error value needing to be less than 5%. The feature importance ranking adopts the random forest algorithm. After training the fused feature matrix, the feature importance score is output. The top 70% of the features are retained. The score results are fed back to the spatiotemporal anchoring fusion module, which increases the weight of the top 20% of the features by 1.2 times and the weight of the 20% to 50% features by 1.1 times to enhance the influence of key features in the fusion process.

[0034] In step S6, the health baseline of the dynamic early warning model is constructed using historical data. Monitoring data within 3 months after the structure's completion is selected as the initial baseline, and then updated every 7 to 30 days. The update cycle is adjusted according to data stability; when the data fluctuation is less than 5% for two consecutive weeks, it is extended to 30 days, and vice versa. The sliding time window size is set according to the structure's natural vibration period: 1 minute for structures with a natural vibration period of less than 1 second, 3 minutes for 1 to 5 seconds, and 5 minutes for more than 5 seconds. The step size is fixed at 1 / 2 of the window length. Deviation is calculated using Euclidean distance, by subtracting the current feature vector from the health baseline vector. The threshold is divided according to the structure's importance level: 1.5 for Level 1 structures (e.g., main bridge), 2.0 for Level 2 (e.g., approach bridge), and 3.0 for Level 3 (e.g., auxiliary structures). An abnormal event is determined when the deviation exceeds the threshold for 3 consecutive windows or when a single deviation exceeds twice the threshold. The abnormal sample storage format includes the original feature vector, occurrence time, and corresponding topological node number. It is stored in an independent abnormal database. The size of a single sample is controlled at 1-2KB, and the storage period is no less than 1 year. It is used for model iteration of the feature enhancement module. Each iteration training requires more than 500 new abnormal samples.

[0035] In step S7, the base layer of the layered visualization template displays real-time monitoring curves. The sampling interval is adjusted according to the data type: vibration data at 1 second / point, strain data at 5 seconds / point, and environmental parameters at 30 seconds / point. Curve colors distinguish data types: blue for strain, green for displacement, and red for vibration. The intermediate layer topology heatmap uses a red-yellow-green three-color mapping feature value. The color thresholds correspond to 60% (green), 80% (yellow), and 100% (red) of the design safety limit. The heatmap update frequency is synchronized with the intermediate frequency data storage cycle, i.e., updated once per hour. The top-level warning markers display different styles according to the anomaly level: a flashing red triangle for level one, a static yellow circle for level two, and an orange square for level three. The marker position is accurate to the topology node coordinates, with an error of no more than 10cm. User interaction supports scaling ratios from 1:10 to 1:1000, with the panning range covering the entire engineering area. The time range selection can go back from 1 hour to 1 year. Custom views support saving 10-20 templates, including commonly used monitoring point combinations and time ranges. Configuration preferences are stored under the user account, and the cross-device synchronization response time is no more than 3 seconds.

[0036] In step S8, the anomaly level is classified based on the decrease in structural bearing capacity. Level 1 warning corresponds to a decrease of more than 10% in bearing capacity, Level 2 to 5%–10%, and Level 3 to less than 5%. The level is determined by comparing the feature vector with the design limit value. The response process is strictly controlled in terms of initiation time. Level 1 warnings are pushed via the project management terminal APP and SMS notifications, requiring the recipient to confirm within 15 minutes. If no confirmation is received, the notification is automatically escalated to the superior management department. Level 2 warnings are linked to the drone inspection system. The drone must complete takeoff preparation within 30 minutes, and the inspection route must cover the abnormal area and three adjacent components. The flight altitude is controlled at 5–10 meters above the structural surface, and the image resolution is not less than 1920×1080 pixels. Level 3 warnings trigger the expert remote consultation system, which automatically matches more than three engineers in relevant fields from the expert database. The system sends a meeting invitation and reserves 24 hours for consultation. The consultation record must include a problem description, preliminary diagnosis, and treatment suggestions, with a text content of no less than 200 words. Text records and image data of the early warning and handling process are stored in the log database in real time. The time-series database is indexed by event timestamps, with a single record storage size not exceeding 5MB and a retention period of 5 years, for subsequent structural performance evaluation and knowledge graph construction.

[0037] From the above, we can conclude that: In this invention, through standardized integration and dynamic cleaning of multi-source heterogeneous data, edge nodes perform real-time preprocessing of various types of sensor data, such as vibration, strain, and displacement, during the data access phase. Combined with moving average filtering and outlier removal mechanisms, the stability of the original data is ensured. After entering the cloud, heterogeneous data from different devices are integrated into a standardized columnar storage structure through unified format conversion and unit calibration. Simultaneously, linear interpolation and drift correction techniques are used to handle missing values ​​and sensor drift errors, significantly improving data quality. This hierarchical processing mode retains the real-time advantages of the edge while leveraging cloud computing power to achieve deep cleaning of large-scale data, providing a more reliable data foundation for subsequent fusion analysis and minimizing misjudgments of structural states due to data anomalies.

[0038] In this invention, the dynamic response efficiency and overall assessment accuracy of civil engineering structure monitoring are significantly improved through the synergistic effect of the elastic scalability of cloud computing architecture and spatiotemporal fusion algorithms. The cloud-edge collaborative storage architecture stores data in layers based on frequency and importance. High-frequency real-time data is processed rapidly at edge nodes, mid-frequency analysis data is efficiently queried in a distributed database in the cloud, and historical data is archived to low-cost object storage, achieving optimized allocation of storage resources. Combining standardized data, the spatiotemporal anchoring algorithm dynamically fuses temporal and spatial response features to generate a feature matrix containing the overall mechanical state of the structure. Through key feature enhancement and dimensionality reduction optimization, it highlights changes in mechanical indicators of key structural components, achieving closed-loop management from data acquisition to safety decision-making, providing efficient and reliable technical support for the health monitoring and maintenance of civil engineering structures.

[0039] It should be noted that, in this document, relational terms such as "first" and "second" are used merely 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 term "comprising" or any other variations thereof is 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 a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0040] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for data fusion processing and visualization analysis of civil engineering structure monitoring based on cloud computing architecture, characterized by: The method includes the following steps: S1: Deploy edge computing nodes to connect with various sensors, drone inspection equipment, and manual recording terminals to collect structural vibration, strain, displacement, and environmental temperature and humidity data. Use lightweight filtering algorithms to initially remove extreme noise data and mark timestamps and spatial coordinates according to data type. S2: Based on the timeliness and importance of data, a three-tier storage system is constructed: high-frequency real-time data is temporarily stored on edge nodes, mid-frequency analytical data is stored in a distributed database on the cloud platform, and historical archived data is migrated to a low-cost object storage system; S3: Perform format conversion, unit unification, and missing value imputation on the heterogeneous data stored in S2. Use the sliding window method to identify and correct sensor drift errors. The cleaned standard data is grouped according to the structural feature dimensions required by the S4 fusion module, and the data quality assessment report is fed back to S1. S4: Combine engineering structural design drawings to construct a physical topology network, map the standardized data output by S3 to structural component nodes, and use a spatiotemporal anchoring algorithm to associate the temporal data of different monitoring points of the same component with the spatial response characteristics of different components, and fuse them to generate a feature matrix containing the overall mechanical state of the structure. S5: Based on the fusion feature matrix output by S4, a method combining principal component analysis and structural mechanics feature screening is used to retain key indicators sensitive to structural safety. The data compression rate and feature integrity are balanced by dynamically adjusting the dimensionality reduction. S6: Combining the key features and structural design of the S5 output, a dynamic early warning model is constructed based on the safety threshold. A sliding time window is used to compare the deviation of the current feature vector from the historical health baseline. S7: Based on the monitoring parameters in the S2 metadata table and the anomaly detection results output by S6, design a layered visualization template: the base layer displays the original monitoring curves, the middle layer presents the structural topology heat map after S4 fusion, and the top layer overlays the anomaly warning markers of S6. S8: Initiate the corresponding response process based on the anomaly level detected by S6: Level 1 warning is pushed to the project management terminal, Level 2 warning is linked to the on-site inspection drone, and Level 3 warning triggers the expert remote consultation system. S9: Integrates the visualization results of S7 with the early warning response records of S8 to generate a structural health monitoring report. It uses natural language processing technology to extract typical cases and handling experience from the report, constructs a knowledge graph for structural disease diagnosis, and regularly updates the topology fusion algorithm and feature weight model to form a data-driven continuous optimization closed loop.

2. The method for data fusion processing and visualization analysis of civil engineering structure monitoring based on cloud computing architecture as described in claim 1, characterized in that: In step S1, when deploying edge computing nodes, a data acquisition module compatible with mainstream sensor interfaces must be configured. The vibration sensor is a piezoelectric accelerometer with a sampling frequency of 50–200 Hz. The strain sensor is a fiber optic grating with a range of -2000–2000 με. The displacement monitoring device is a laser displacement meter with an accuracy controlled at the 0.01 mm level. The sampling interval of the environmental temperature and humidity sensor is set to 10–30 seconds. The edge node hardware configuration must meet at least a 4-core processor and 8 GB of memory to support parallel access of multiple devices. The maximum number of devices that can be accessed by a single node is no more than 50. In the data preprocessing stage, a moving average filtering algorithm is used. The window size is dynamically adjusted according to the data type. The vibration data window is set to 5 sampling points, and the strain data window is set to 10 sampling points. By comparing the deviation of the current sampled value with the window mean, extreme noise data that deviates more than 3 times the standard deviation are removed.

3. The method for data fusion processing and visualization analysis of civil engineering structure monitoring based on cloud computing architecture as described in claim 1, characterized in that: In step S2, the specific parameter configuration of the three-level storage system needs to be determined in combination with the data characteristics. High-frequency real-time data includes vibration acceleration and dynamic strain data, which are temporarily stored in the local SSD storage of the edge node. The storage capacity is configured as 20-50GB according to the number of connected devices, and the temporary storage time is set to 5-15 minutes. When the storage occupancy rate reaches 80%, the data cleaning mechanism is triggered.

4. The method for data fusion processing and visualization analysis of civil engineering structure monitoring based on cloud computing architecture as described in claim 1, characterized in that: In step S3, the data cleaning stage first performs format conversion, uniformly converting the CSV, JSON, and binary formats output by the sensors into Parquet columnar storage format. During the conversion process, data integrity is verified, and files with a missing field ratio exceeding 10% are directly marked as invalid. Units are uniformly standardized according to engineering specifications, with strain units converted to με, displacement to mm, temperature to ℃, and humidity to %RH, and the conversion accuracy retaining two decimal places. Missing values ​​are handled using linear interpolation. When the missing rate of a single data point is less than 5%, interpolation is performed; if it exceeds 5%, the front-end equipment calibration process is triggered. The calibration cycle is adjusted according to the equipment type, with vibration sensors calibrated once a month and strain sensors calibrated once every two months.

5. The method for data fusion processing and visualization analysis of civil engineering structure monitoring based on cloud computing architecture as described in claim 1, characterized in that: In step S4, a physical topology network is constructed by combining the component types and their connection methods in the engineering structural design drawings. The overall structure is decomposed into several basic component nodes, each node corresponding to a structural unit with independent mechanical function. An adjacency matrix is ​​used to describe the connection relationship between nodes. The matrix element values ​​are dynamically assigned according to the connection type: 1.0 for rigid connection, 0.6 for hinged connection, 0.3 for sliding connection, and 0 for no connection. At the same time, component material properties are introduced.

6. The method for data fusion processing and visualization analysis of civil engineering structure monitoring based on cloud computing architecture as described in claim 1, characterized in that: In step S4, when implementing the spatiotemporal anchoring algorithm, firstly, the time-domain data of multiple monitoring points of the same topological node are calibrated on the time axis. A time synchronization method based on cross-correlation function is used to align monitoring data with different sampling frequencies. By setting a time tolerance threshold, usually 1 / 5 of the sampling period, data points with time deviations exceeding the threshold are removed. Next, a variable-length sliding time window is used. The window length is dynamically adjusted according to the structure's natural vibration characteristics. When the structure's response is stable, the window length is increased to 3 times the natural vibration period. When the response is severe, it is shortened to 1 times the natural vibration period. The time-domain feature parameters within the window are extracted. For the spatial response characteristics of different topological nodes, the spatial influence weight between nodes is calculated based on the topological adjacency matrix. The weight value is inversely proportional to the topological distance between nodes. The topological distance is defined as the number of components on the shortest connection path between two nodes. At the same time, the weight value is corrected by combining the weighting coefficient of material properties. The time-domain feature parameters of the same node are fused with the spatial weighted feature parameters of adjacent nodes. The feature matrix is ​​generated by arranging the topological nodes in order. The row vectors of the matrix correspond to the topological nodes, and the column vectors correspond to different mechanical feature indices. This matrix completely contains the mechanical state information of the entire structure, providing a data foundation for subsequent steps. The most innovative formula in the spatiotemporal anchoring fusion process is the formula for calculating the spatiotemporal fusion feature value of topological node i at time t: ； The parameters in the formula are defined as follows: F i (t) represents the fusion feature value of topological node i at time t; β is the time feature weighting coefficient, with a value ranging from 0.4 to 0.

6. It is dynamically adjusted according to the stability of the node data, increasing to 0.6 when the data fluctuation is small and decreasing to 0.4 when the fluctuation is large. T i (t) is the time-domain feature vector of node i at time t, which consists of the peak value, mean, and variance within the sliding time window; Ψ i Let i be the set of spatially adjacent nodes of node i, that is, all nodes whose topological distance is not greater than 2; γ ij The spatial influence weight of node i on node j is determined by the topological distance d. i j is calculated. ; Δt is the time delay of mechanical response transmission between nodes, ranging from 0.08 to 0.15 seconds, and is determined based on the stiffness of the component; the greater the stiffness, the smaller the time delay. T j (t-Δt) is the time-domain feature vector of the adjacent node j at time t-Δt.

7. The method for data fusion processing and visualization analysis of civil engineering structure monitoring based on cloud computing architecture as described in claim 1, characterized in that: In step S5, the cumulative contribution rate threshold of principal component analysis during feature enhancement is set to 90%, that is, principal components whose sum of feature variances accounts for more than 90% of the total variance are retained. At the same time, sensitive indicators are screened in combination with structural mechanical properties, strain gradient is selected as the strain difference between the mid-span and the support of the component, displacement curvature is calculated as the second difference between adjacent monitoring points, and vibration frequency is extracted as the first 5 natural frequencies. The importance weights of these indicators are initialized through the structural finite element simulation results.

8. The method for data fusion processing and visualization analysis of civil engineering structure monitoring based on cloud computing architecture as described in claim 1, characterized in that: In step S6, the health baseline of the dynamic early warning model is constructed using historical data. Monitoring data within 3 months after the structure's completion is selected as the initial baseline, and then updated every 7 to 30 days. The update cycle is adjusted according to data stability. When the data fluctuation is less than 5% for two consecutive weeks, it is extended to 30 days; otherwise, it is shortened to 7 days. The sliding time window size is set according to the structure's natural vibration period. For structures with a natural vibration period of less than 1 second, the window is set to 1 minute; for those with a natural vibration period of 1 to 5 seconds, it is set to 3 minutes; and for those with a natural vibration period of more than 5 seconds, it is set to 5 minutes. The step size is fixed at 1 / 2 of the window length. The deviation is calculated using Euclidean distance. The modulus is calculated by subtracting the current feature vector from the health baseline vector. The threshold is divided according to the importance level of the structure. The threshold for level 1 structures is set to 1.5, for level 2 structures to 2.0, and for level 3 structures to 3.

0. When the deviation exceeds the threshold for 3 consecutive windows or when the deviation exceeds twice the threshold in a single instance, it is judged as an abnormal event.

9. The method for data fusion processing and visualization analysis of civil engineering structure monitoring based on cloud computing architecture as described in claim 1, characterized in that: In step S7, the base layer of the layered visualization template displays real-time monitoring curves. The sampling interval is adjusted according to the data type: vibration data at 1 second / point, strain data at 5 seconds / point, and environmental parameters at 30 seconds / point. The curve colors distinguish the data types: blue for strain, green for displacement, and red for vibration. The intermediate layer topology heat map uses red-yellow-green three-color mapping feature values. The color thresholds correspond to 60%, 80%, and 100% of the design safety limits. The heat map update frequency is synchronized with the intermediate frequency data storage cycle, i.e., it is updated once per hour.

10. The method for data fusion processing and visualization analysis of civil engineering structure monitoring based on cloud computing architecture as described in claim 1, characterized in that: In step S8, the anomaly level is classified according to the decrease in structural bearing capacity. Level 1 warning corresponds to a decrease in bearing capacity of more than 10%, Level 2 to 5% to 10%, and Level 3 to less than 5%. The level is determined by comparing the feature vector with the design limit value. The response process is strictly controlled in terms of start time. Level 1 warning is pushed through the project management terminal APP and SMS notification, requiring the recipient to confirm within 15 minutes. If no confirmation is made, the notification is automatically upgraded to the superior management department. The Level 2 early warning system is linked to the drone inspection system. The drone must complete takeoff preparation within 30 minutes, and the inspection route covers the abnormal area and three adjacent components. The flight altitude is controlled at 5 to 10 meters away from the structural surface, and the image resolution is not less than 1920×1080 pixels. The Level 3 early warning system triggers the remote expert consultation system.

Images