Bridge monitoring data storage system based on cloud computing

Abstract

The invention discloses a bridge monitoring data storage system based on cloud computing, particularly relates to the technical field of bridge monitoring, and comprises an optical fiber sensing network acquisition unit, an edge computing preprocessing unit, a multi-modal data fusion storage module, a dynamic hierarchical compression engine, a fault-data association analysis unit and a distributed cloud storage cluster. According to the invention, a full-link optimization system is constructed, a front end accurately captures key responses, a middle end filters abnormal data to reduce invalid transmission, and a rear end dynamically compresses based on health rating and combines a cloud storage mechanism, so that the storage efficiency is improved by more than 40%, and the storage cost and bandwidth pressure are solved; key data are compared and marked through a knowledge graph, and a triple protection system is constructed in combination with a fault-tolerant strategy; a fusion index is formed by space-time correlation, multi-dimensional analysis is supported, authority control gives consideration to safety and operation and maintenance, storage cost is optimized by cold and hot stratification, and a high-availability and extensible data management platform is provided for bridge monitoring.

Description

Technical Field

[0001] The present invention relates to the technical field of bridge monitoring, and particularly relates to a bridge monitoring data storage system based on cloud computing. Background Art

[0002] In the field of bridge health monitoring, with the rapid development of sensor technology and the Internet of Things, the multi-modal data (such as strain, vibration, images, etc.) generated by bridge monitoring has shown explosive growth. Traditional storage systems face problems such as high cost of massive data storage, low transmission efficiency, easy loss of characteristic data, and insufficient security protection. Existing solutions usually adopt a fixed sampling frequency and a unified compression strategy, which cannot accurately capture the key responses of the bridge structure under different stress states, resulting in redundant storage of low-value data while fault characteristic data may be lost due to excessive compression. At the same time, there is a lack of an efficient spatio-temporal correlation mechanism for multi-source data (such as BIM models, sensing data, crack images), making it difficult to support multi-dimensional joint analysis of structural damage.

[0003] In addition, the traditional storage architecture does not dynamically adjust the storage strategy according to the bridge health status. The mixed storage of hot and cold data leads to high response latency for frequently accessed data, and there is a lack of refined permission control and fault tolerance mechanism, making it difficult for data availability and security to meet the long-term monitoring requirements. Therefore, there is an urgent need for an intelligent storage system based on cloud computing to achieve efficient management and in-depth mining of bridge monitoring data through technologies such as dynamic sampling, hierarchical compression, spatio-temporal data fusion, and cloud distributed storage, and to improve the accuracy and timeliness of structural safety assessment. Summary of the Invention

[0004] The main purpose of the present invention is to provide a bridge monitoring data storage system based on cloud computing, which can effectively solve the problems in the above background art.

[0005] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0006] A bridge monitoring data storage system based on cloud computing, the bridge monitoring data storage system based on cloud computing is configured as:

[0007] Optical fiber sensing network acquisition unit: Extract the strain gradient distribution data on the surface of the bridge structure, dynamically adjust the sampling frequency of FBG sensors according to the strain gradient, and generate an original sensing signal stream;

[0008] Edge computing preprocessing unit: Call the original sensing signal stream, perform vibration signal mutation feature extraction and timestamp alignment, and generate a preprocessing signal set;

[0009] Multi-modal data fusion storage module: Integrate the preprocessing signal set, the spatial coordinates of the bridge BIM model, and the crack image features, establish a spatio-temporal correlation mapping, and generate a fusion storage index;

[0010] Dynamic hierarchical compression engine: Monitor the structural health status rating, dynamically select the compression algorithm and data retention frequency band, and output the hierarchical compression data stream;

[0011] Fault-data correlation analysis unit: Compare the real-time spectrum characteristics with the historical fault knowledge graph, mark the key data storage nodes, and generate instructions for optimizing the compression strategy;

[0012] Distributed cloud storage cluster: Implement erasure code sharding storage based on network latency, perform hierarchical migration of hot and cold data, and output the monitoring data access interface.

[0013] Preferably, the fiber optic sensing network acquisition unit includes:

[0014] FBG sensor array deployed according to the bridge stress distribution topology, and the sensor spacing is dynamically set to 0.5 - 1.2 meters according to the high strain gradient area of the finite element model;

[0015] Dual-wavelength demodulation module: Includes a first channel with a central wavelength of 1520nm ± 2nm and a second channel with a central wavelength of 1550nm ± 2nm, and compensates the temperature drift error in real time through dual-channel differential calculation;

[0016] Flexible encapsulation structure, whose thermal expansion coefficient matches the concrete curing shrinkage coefficient as:

[0017] 1.0 ± 0.2×10 -6 / ℃

[0018] Preferably, the edge computing preprocessing unit includes:

[0019] Adopt the Daubechies-4 wavelet packet entropy value calculation algorithm to extract the mutation characteristics with frequencies of 50 - 300Hz in the vibration signal;

[0020] The abnormal data filter triggers the upload of all data when the sampling frequency is 200 - 500Hz and lasts for 0.1 - 0.5s;

[0021] The timestamp synchronization module aligns the multi-sensor data, and the synchronization accuracy is 10 ± 2μs.

[0022] Preferably, the multi-modal data fusion storage module adopts a hierarchical storage architecture, where:

[0023] The real-time vibration data is stored in the InfluxDB partition in a time series format, and the sampling interval is 5 - 10ms;

[0024] After the structural crack image is extracted by SIFT features, the crack topology relationship is stored in the Neo4j graph database;

[0025] Establish the spatial mapping between the bridge BIM model and the sensing data, and bind the 3D coordinate data with an accuracy of ±5 cm.

[0026] Preferably, the dynamic hierarchical compression engine performs compression according to the health status rating:

[0027] Health status (rating Ⅰ): Retain the characteristic frequency data of 1 - 5 Hz;

[0028] Warning status (rating Ⅱ): Retain the full - band data of 50 - 100 Hz;

[0029] Fault status (rating Ⅲ): Store the original data without loss;

[0030] Adopt the improved LZW compression algorithm. When the compression ratio calculated in real - time reaches 10:1 - 15:1, start the byte - level repeated pattern detection module.

[0031] Preferably, the fault - data correlation analysis unit builds in a historical fault knowledge graph, including a mapping relationship library between the vibration spectrum of 10 - 200 Hz and the crack propagation mode;

[0032] When the fundamental frequency offset reaches 5 - 8%, automatically associate and store the high - frequency vibration data of 200 ± 10 Hz at the corresponding position, generate a compression strategy instruction and feedback it to the dynamic hierarchical compression engine, and the instruction update period is 5 - 30 s.

[0033] Preferably, the distributed cloud storage cluster adopts the erasure code sharding strategy (EC6 + 3), and the shard size is dynamically adjusted to 64 - 256 KB according to the network latency;

[0034] Hot - cold data layering rules:

[0035] Hot data layer: The SSD stores the high - frequency monitoring data in the past 72 ± 12 hours;

[0036] Cold data layer: The HDD stores the long - term trend data after feature extraction;

[0037] The data access interface integrates the RBAC permission control and supports authorization at the granularity of a single sensor node.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. The present invention constructs a full-link optimization system from data acquisition to cloud storage. The front-end fiber optic sensing network dynamically adjusts the sampling frequency according to the bridge strain gradient to accurately capture the key structural responses. The mid-end edge computing nodes use the wavelet packet entropy algorithm to filter abnormal vibration data in real time, significantly reducing ineffective transmissions. The back-end dynamic hierarchical compression engine adaptively switches the compression strategy based on the structural health rating, achieving high-ratio compression while retaining the fault feature frequency bands. Combining the hot and cold data stratification mechanism of the distributed cloud storage cluster and the network-aware erasure code sharding technology, the overall storage efficiency of the system is increased by more than 40%, effectively solving the storage cost and bandwidth pressure brought by massive monitoring data.

[0040] 2. The present invention compares the fundamental frequency offset and high-frequency components in real time through the historical fault knowledge graph, actively marks the key data nodes in high-risk periods, and dynamically regulates the compression engine according to the generated compression optimization instructions to ensure the lossless storage of fault feature data. Combining the redundant sensor voting mechanism and the EC6+3 erasure code fault tolerance strategy, a triple protection system of "fault identification - data protection - error repair" is constructed. This design enables the availability of key data to reach 99.7%, greatly improving the timeliness and accuracy of structural safety diagnosis.

[0041] 3. The present invention uses the spatial hashing algorithm to spatially and temporally associate the BIM spatial coordinates, vibration time series, and crack topological relationships to form a traceable "structure - response - damage" fusion index, supporting multi-dimensional joint analysis. The permission control system based on RBAC and OAuth2.0 realizes 6-level access authorization at the single sensor node granularity, taking into account data security and flexible operation and maintenance. The hot and cold data stratification uses a quantitative weight model to trigger intelligent migration, optimizing the long-term storage cost while ensuring the real-time performance of high-frequency data. This system provides a highly available and scalable integrated data management platform for bridge health monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a schematic diagram of the overall process of the present invention;

[0043] Figure 2 is a schematic diagram of the process of the dynamic hierarchical compression engine in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0044] To make the technical means, creative features, achieved purposes, and effects of the present invention easy to understand, the present invention will be further described below in conjunction with specific embodiments.

[0045] Example 1, as Figure 1 AND Figure 2 shown, a bridge monitoring data storage system based on cloud computing is configured as:

[0046] Optical fiber sensing network acquisition unit: Extract the strain gradient distribution data on the surface of the bridge structure, dynamically adjust the sampling frequency of FBG sensors according to the strain gradient, and generate the original sensing signal stream;

[0047] Edge computing preprocessing unit: Call the original sensing signal stream, perform vibration signal mutation feature extraction and timestamp alignment, and generate a preprocessing signal set;

[0048] Multi-modal data fusion storage module: Integrate the preprocessing signal set, the spatial coordinates of the bridge BIM model, and the crack image features, establish a spatio-temporal correlation mapping, and generate a fusion storage index;

[0049] Dynamic hierarchical compression engine: Monitor the structural health status rating, dynamically select the compression algorithm and data retention frequency band, and output a hierarchical compression data stream;

[0050] Fault-data correlation analysis unit: Compare the real-time spectrum features with the historical fault knowledge graph, mark the key data storage nodes, and generate compression strategy optimization instructions;

[0051] Distributed cloud storage cluster: Implement erasure code sharding storage based on network latency, perform hierarchical migration of hot and cold data, and output a monitoring data access interface.

[0052] First, deploy the optical fiber sensing network, and use a 48-channel FBG demodulator (model SM125) to build a distributed acquisition node. Arrange the sensor array at a spacing of 0.8 meters at the web of the bridge main girder. After verifying the strain gradient distribution through the finite element model, dynamically adjust the sensor density in the high stress area to 0.5 meters. The edge computing node uses the NVIDIA Jetson AGX Xavier platform and deploys the Daubechies-4 wavelet packet algorithm. When the vibration signal frequency exceeds 50 Hz, start the 500 Hz high-speed sampling mode. When performing multi-modal data fusion, use the spatial hashing algorithm to establish a mapping relationship between the BIM model coordinates (X = 1024.35, Y = 358.76, Z = +25.4) and the vibration data, with the error controlled within ±3 cm. The fault analysis unit has an SQLite database containing 200 groups of historical fault spectra built-in. When the fundamental frequency offset reaches 6%, trigger the high-frequency data full retention instruction.

[0053] The optical fiber sensing network acquisition unit includes:

[0054] An FBG sensor array deployed according to the bridge stress distribution topology, and the sensor spacing is dynamically set to 0.5 - 1.2 meters according to the high strain gradient area of the finite element model;

[0055] Dual-wavelength demodulation module: It includes a first channel with a central wavelength of 1520 nm ± 2 nm and a second channel with a central wavelength of 1550 nm ± 2 nm, and compensates for temperature drift errors in real time through dual-channel differential calculation;

[0056] Flexible packaging structure, whose coefficient of thermal expansion matches the coefficient of curing shrinkage of concrete as:

[0057] 1.0 ± 0.2×10 -6 / ℃

[0058] Furthermore, the sensor packaging adopts a polyimide-silica gel composite structure, whose coefficient of thermal expansion α = 1.05×10 -6 / ℃ (test standard GB / T20624.2);

[0059] In the dual-wavelength demodulation module, the 1520 nm channel is used for strain measurement, and the 1550 nm channel is dedicated to temperature compensation. The optical intensity difference ΔP between the two channels is controlled within the range of 5 - 8 dBm. When deploying, a redundant sensor group is set at the root of the bridge pier. Each group contains 3 sensor nodes distributed at 120°. The single-point error is eliminated through a voting mechanism. The temperature compensation algorithm is:

[0060] Δε = k1·Δλ1 - k2·Δλ2

[0061] where k1 = 0.78 με / pm (strain coefficient), k2 = 0.05 με / ℃ (temperature compensation coefficient).

[0062] The temperature drift compensation effect of the dual-wavelength demodulation module is verified through the actual bridge environment:

[0063] In the temperature difference range of -10℃ to 45℃, the strain measurement error is stable within ±2.5 με (the error reaches ±15 με without compensation). The voting mechanism of the redundant sensor group adopts a two-out-of-three majority decision algorithm, reducing the false alarm rate caused by single-point failures to 0.3%. After 2000 thermal cycle tests of the flexible packaging structure the peeling rate of the sensor from the concrete is <1%, significantly improving the long-term monitoring stability. The measured data shows that the dynamic adjustment of the sampling frequency strategy improves the integrity of data capture in the high-stress area by 35% and reduces the redundant data volume in the low-stress area by 30% at the same time.

[0064] The edge computing preprocessing unit includes:

[0065] Adopt the Daubechies-4 wavelet packet entropy value calculation algorithm to extract the mutation characteristics with frequencies of 50 - 300 Hz in the vibration signal;

[0066] The abnormal data filter triggers the upload of all data when the sampling frequency is 200 - 500 Hz and lasts for 0.1 - 0.5 s;

[0067] The timestamp synchronization module aligns multi-sensor data with a synchronization accuracy of 10±2 μs.

[0068] Furthermore, the edge node is configured with a 512 MB dedicated buffer. When more than 3 vibration pulses above 300 Hz are continuously detected, the full-volume data upload mode is initiated.

[0069] The time synchronization module adopts the PTPv2 protocol and deploys 6 timing nodes in the fiber optic network to ensure that the cross-segment data synchronization error < 8 μs.

[0070] The anomaly filter sets two levels of thresholds: the primary threshold triggers local data saving (vibration amplitude > 2.5 g), and the secondary threshold triggers cloud alarms (continuous overrun for 0.3 s). The preprocessed data packet is added with a quadruple tag (timestamp|location code|data type|confidence), for example, "202505291512|B2-3-7|VIB|0.92".

[0071] Among them, in the anomaly data filtering process, the wavelet packet entropy value calculation adopts:

[0072]

[0073] H is the entropy value (unit: bit), p i is the energy ratio of the i-th frequency band, N = 8 is the decomposition layer number. When H > 4.2, it is determined as an abnormal signal and triggers full-volume upload.

[0074] Furthermore, the 512 MB dedicated buffer adopts a circular queue design, supporting a burst data peak throughput of up to 120 MB / s (for 5 seconds), effectively avoiding data loss caused by network jitter. After deploying 6 PTPv2 timing nodes in the time synchronization module, the measured synchronization error across a 200-meter bridge span is 6.8 μs (better than the design index). The two-level threshold mechanism of the anomaly filter has been verified by vehicle flow impact tests: the primary threshold is triggered to save key waveforms when heavy trucks cross the bridge, and the cloud alarm is triggered by continuous overrun for 0.3 s during the seismic simulation vibration table test, with a false trigger rate < 1 time / month. The detection rate of the entropy value calculation anomaly determination model (H > 4.2) for crack propagation events reaches 92%, which is 27% higher than the traditional threshold method.

[0075] The multi-modal data fusion storage module adopts a hierarchical storage architecture, where:

[0076] Real-time vibration data is stored in the InfluxDB partition in a time series format with a sampling interval of 5 - 10 ms;

[0077] After the structural crack images are subjected to SIFT feature extraction, the crack topology relationship is stored in the Neo4j graph database;

[0078] Establish the spatial mapping between the bridge BIM model and the sensing data, and bind the 3D coordinate data with an accuracy of ±5 cm.

[0079] Specifically, InfluxDB partitioning adopts a sharding strategy by bridge span, and each bridge span configures an independent time series database instance.

[0080] When processing crack images, the SIFT feature extraction sets the key point threshold = 0.03, and at least 50 feature vectors are extracted for each crack. The spatial mapping service deploys the PostGIS extension module. When establishing the 3D coordinate system, the center point of Pier 0# (WGS84 coordinates: 40.7682°N, 114.8865°E) is set as the bridge origin, and the binding accuracy is optimized by the RANSAC algorithm. When constructing the crack topological relationship in the graph database, the "adjacent" relationship is defined as the edge distance < 15 cm and the trend angle < 30°.

[0081] The dynamic hierarchical compression engine performs compression according to the health status rating:

[0082] Health status (rating Ⅰ): Retain the feature frequency data of 1 - 5 Hz;

[0083] Warning status (rating Ⅱ): Retain the full frequency band data of 50 - 100 Hz;

[0084] Fault status (rating Ⅲ): Store the original data losslessly;

[0085] Adopt the improved LZW compression algorithm. When the compression ratio calculated in real time reaches 10:1 - 15:1, start the byte-level repeat pattern detection module.

[0086] The compression engine configures a three-level strategy:

[0087] Level Ⅰ status enables the improved LZW algorithm, sets the sliding window = 8 KB, and the dictionary size = 4K items;

[0088] Level Ⅱ status switches to Zstd compression and sets the compression level = 7;

[0089] Level Ⅲ status enables lossless FLAC encoding.

[0090] The dynamic switching threshold sets the hysteresis interval. The compression is enabled only when the health status is rated Ⅰ continuously for 5 minutes, while the warning status is switched immediately. The byte-level repeat detection module sets the minimum pattern length = 16 bytes, and when more than 3 repetitions are detected, start the pattern replacement encoding.

[0091] The dynamic efficiency evaluation formula of the improved LZW compression algorithm:

[0092]

[0093] R cis the real-time compression ratio (%), F raw is the amount of original data (bytes), D size is the space occupied by dictionary entries, W dict = 4096 is the maximum dictionary capacity. When R c is continuously lower than 12% for 10 seconds, the compression algorithm is automatically switched.

[0094] The fault-data correlation analysis unit has a built-in historical fault knowledge graph, which contains a mapping relationship library between the vibration spectrum of 10 - 200 Hz and the crack propagation mode;

[0095] When the fundamental frequency offset reaches 5% - 8%, the high-frequency vibration data of 200 ± 10 Hz at the corresponding position is automatically associated and stored, generating a compression strategy instruction and feeding it back to the dynamic hierarchical compression engine. The instruction update period is 5 - 30 s.

[0096] Furthermore, the knowledge graph is constructed using the Neo4j graph database. The nodes include three types of entities: fault type (such as crack propagation), spectral features (center frequency, harmonic components), and disposal measures.

[0097] Association rule setting: When components of 180 - 220 Hz are detected and the Q factor > 12, the shear crack mode is automatically associated.

[0098] Optimize the instruction generation period for dynamic adjustment. When the network latency < 50 ms, update it according to a 5 - second cycle. When the latency > 100 ms, extend it to 30 seconds.

[0099] The instruction encoding uses the TLV format, and the type field identifies the instruction type (0xA1 = data retention, 0xA2 = compression adjustment).

[0100] Spectral feature correlation degree calculation model:

[0101]

[0102] where A f is the real-time spectral amplitude, H f is the historical fault feature amplitude, and f is the frequency (Hz). When S corr > 0.85, an optimization instruction is generated, and the correlation confidence

[0103] Furthermore, 300 actual bridge fault cases are incorporated into the knowledge graph construction, forming an association relationship between 182 spectral feature nodes and 47 crack modes. The spectral correlation degree model (S_corr > 0.85) successfully marks key data in the stay cable wire break event, reducing the amount of data required for subsequent analysis by 82%.

[0104] Instruction Update Cycle Dynamic Adjustment Algorithm: Predict the duration of the next cycle based on the median of historical network latency to keep the instruction transmission success rate above 99.6%. Add an "urgency" flag (0 - 255) to the TLV instruction field. When the detected fundamental frequency offset > 7%, send the highest priority instruction 0xA1|0xFF to ensure zero-compression storage of 200 ± 10Hz data.

[0105] The distributed cloud storage cluster adopts an erasure code sharding strategy (EC6+3), and the shard size is dynamically adjusted to 64 - 256KB according to network latency;

[0106] Hot and cold data layering rules:

[0107] Hot data layer: SSD stores high-frequency monitoring data for nearly 72 ± 12 hours;

[0108] Cold data layer: HDD stores long-term trend data after feature extraction;

[0109] The data access interface integrates RBAC permission control and supports authorization at the granularity of a single sensor node.

[0110] Furthermore, the erasure code sharding adopts Reed-Solomon(6,3) coding, and the shard size is dynamically adjusted according to network quality: 256KB large shards are used in a good network (packet loss rate < 0.1%), and 64KB shards are switched in a poor network (packet loss rate > 1%).

[0111] The hot data layer adopts a RAID10 array, configured with 4 NVMe SSDs (Samsung PM1733), and the cold data layer adopts an 8-drive NAS (Seagate Exos16TB).

[0112] Quantify the trigger conditions for hot and cold data migration:

[0113] W d = 0.6 × access frequency + 0.4 × fault correlation

[0114] W d > 0.8: Remain in the SSD layer (high-frequency access within 72 hours);

[0115] W d < 0.3: Migrate to the HDD layer (aggregated storage by week);

[0116] The migration decision is executed every 2 hours (to avoid I / O jitter)

[0117] The access interface implements permission control based on OAuth2.0, defining 6 levels of access permissions:

[0118] Level 1 can only read vibration data, level 4 can access the original image, and level 6 has full data download permission.

[0119] Furthermore, the erasure code sharding strategy is combined with network awareness: when the packet loss rate > 1%, it automatically switches to the EC4+2 mode and reduces the shard size to 64KB, reducing the transmission failure rate from 5.7% to 0.9%. The hot and cold data migration weight model (Wd) introduces a time decay factor: the access frequency is calculated exponentially (half-life = 24h) to prevent old data from occupying the SSD for a long time. RBAC permission control realizes field-level security: users with level 4 permissions can access the original image but the coordinate metadata is hidden, and level 6 users can unlock all fields.

[0120] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A bridge monitoring data storage system based on cloud computing, characterized in that, The bridge monitoring data storage system based on cloud computing is configured as follows: Fiber optic sensing network acquisition unit: Extract the strain gradient distribution data on the surface of the bridge structure, dynamically adjust the sampling frequency of FBG sensors according to the strain gradient, and generate the original sensing signal stream; Edge computing preprocessing unit: Call the original sensing signal stream, perform vibration signal mutation feature extraction and timestamp alignment, and generate a preprocessing signal set; Multi-modal data fusion storage module: Integrate the preprocessing signal set, the spatial coordinates of the bridge BIM model, and the crack image features, establish a spatio-temporal correlation mapping, and generate a fusion storage index; Dynamic hierarchical compression engine: Monitor the structural health status rating, dynamically select the compression algorithm and data retention frequency band, and output a hierarchical compression data stream; Fault-data correlation analysis unit: Compare the real-time spectrum features with the historical fault knowledge graph, mark the key data storage nodes, and generate compression strategy optimization instructions; Distributed cloud storage cluster: Implement erasure code sharding storage based on network latency, perform cold and hot data hierarchical migration, and output a monitoring data access interface.

2. The bridge monitoring data storage system based on cloud computing according to claim 1, wherein The fiber optic sensing network acquisition unit includes: An FBG sensor array deployed according to the bridge stress distribution topology, and the sensor spacing is dynamically set to 0.5 - 1.2 meters according to the high strain gradient area of the finite element model; The dual-wavelength demodulation module includes a first channel with a central wavelength of 1520nm ± 2nm and a second channel with a central wavelength of 1550nm ± 2nm, and compensates the temperature drift error in real time through dual-channel differential calculation; A flexible encapsulation structure, whose thermal expansion coefficient matches the concrete curing shrinkage coefficient as: 1.0±0.2×10 -6 / ℃。 3. A bridge monitoring data storage system based on cloud computing according to claim 1, characterized in that: The edge computing preprocessing unit includes: Adopt the Daubechies-4 wavelet packet entropy value calculation algorithm to extract the mutation features with a frequency of 50–300Hz in the vibration signal; The abnormal data filter triggers full-scale data upload when the sampling frequency is 200 - 500Hz and lasts for 0.1 - 0.5s; The timestamp synchronization module aligns multi-sensor data, and the synchronization accuracy is 10 ± 2μs.

4. A bridge monitoring data storage system based on cloud computing according to claim 1, characterized in that: The multi-modal data fusion storage module adopts a hierarchical storage architecture, where: Real-time vibration data is stored in the InfluxDB partition in time series format, and the sampling interval is 5 - 10ms; After the structural crack image is extracted by SIFT features, the crack topology relationship is stored in the Neo4j graph database; Establish a spatial mapping between the bridge BIM model and the sensing data, and bind the three-dimensional coordinate data with an accuracy of ±5cm.

5. The bridge monitoring data storage system based on cloud computing according to claim 1, wherein: The dynamic hierarchical compression engine performs compression according to the health status rating: Health status (rating Ⅰ): Retain the data with characteristic frequencies of 1 - 5Hz; Warning status (rating Ⅱ): Retain the full frequency band data of 50 - 100Hz; Fault status (rating Ⅲ): Store the original data losslessly; Adopt the improved LZW compression algorithm, and start the byte-level repeated pattern detection module when the real-time calculated compression ratio reaches 10:1 - 15:

1.

6. The bridge monitoring data storage system based on cloud computing according to claim 1, wherein: The fault-data correlation analysis unit has a built-in historical fault knowledge graph, including a mapping relationship library between the vibration spectrum of 10 - 200Hz and the crack propagation mode; When the fundamental frequency offset reaches 5% - 8%, automatically associate and store the high-frequency vibration data at the corresponding position of 200 ± 10 Hz, generate a compression strategy instruction and feedback it to the dynamic hierarchical compression engine, and the instruction update period is 5 - 30 s.

7. A bridge monitoring data storage system based on cloud computing according to claim 1, characterized in that: The distributed cloud storage cluster adopts an erasure code sharding strategy (EC6+3), and the shard size is dynamically adjusted to 64 - 256 KB according to network latency; Hot and cold data layering rules: Hot data layer: The SSD stores high-frequency monitoring data for nearly 72 ± 12 hours; Cold data layer: The HDD stores the long-term trend data after feature extraction; The data access interface integrates RBAC permission control and supports authorization at the granularity of a single sensor node.

Images