Engineering potential safety hazard closed-loop management method based on structure detection data

Abstract

The invention discloses an engineering potential safety hazard closed-loop management method based on structure detection data, and the method comprises the following steps: formulating a differential data collection scheme according to the structure characteristics of different projects, and carrying out the preprocessing of the collected data, and generating a standardized data set; constructing a lightweight space-time fusion model, driving hidden danger recognition by using a fixed hyper-parameter training model, and outputting a result; risk values are calculated based on a multi-dimensional dynamic risk formula, and hidden danger grading is carried out; calling an exclusive case library and a specification library to generate a customized rectification scheme, and after rectification is completed, confirming a rectification effect through staged dual verification; a self-adaptive early warning model is trained based on archived data, and recurrence early warning is triggered through a hierarchical dynamic threshold of an associated engineering health state; and encrypting and archiving the whole-process data to a corresponding engineering security database through a block chain. According to the invention, full-process closed-loop management of different projects can be realized, and a standardized and high-precision safety management technical scheme is provided for multiple types of projects.

Description

Technical Field

[0001] This invention relates to the field of engineering safety management technology, and in particular to a closed-loop management method for engineering safety hazards based on structural inspection data. Background Technology

[0002] The management of engineering safety hazards faces multi-dimensional technical bottlenecks, making it difficult to adapt to the differentiated needs of different types of projects such as tunnels, high-rise buildings, municipal bridges, ports and wharves, and railway subgrades.

[0003] From the perspective of data acquisition and processing, traditional methods mostly adopt a unified acquisition scheme and do not customize strategies for engineering structural characteristics and environmental differences: tunnels are in a humid environment for a long time, and data such as water seepage and gas concentration are easily affected by electromagnetic interference, but there is a lack of special filtering processing; railway subgrades are affected by the cyclic load of trains, and dynamic data such as sleeper displacement and pore water pressure are subject to instantaneous interference, which is difficult to effectively remove using conventional processing methods; ports and wharves face salt spray corrosion and wave impact, and data such as pile foundation settlement and steel structure corrosion are prone to signal drift. Existing anti-interference methods lack standardized mathematical model support, resulting in insufficient data reliability.

[0004] In the hazard identification stage, traditional models suffer from an imbalance between accuracy and efficiency: while complex models can improve identification accuracy, they have a large number of parameters and slow inference speed, making them unsuitable for real-time monitoring scenarios on mobile devices; lightweight models simplify the feature extraction process, resulting in insufficient ability to capture low-dimensional spatial features (such as single-point settlement and localized corrosion) and complex temporal trends (such as crack propagation and roadbed settlement accumulation). Furthermore, existing models often employ single feature fusion methods, failing to fully integrate the correlation between spatial and temporal features, leading to high rates of false positives and false negatives.

[0005] The hazard classification and rectification process lacks a scientific system: classification standards often rely on single indicators (such as hazard size) without comprehensively considering multiple dimensions such as project type, environmental severity, and hazard duration, resulting in a lack of specificity in the classification results; rectification plans are mostly generic templates, not customized to the specific characteristics of the project and the hazard risk level, making it difficult to guarantee the effectiveness of rectification. The verification methods for rectification effectiveness are simplistic, relying solely on comparing deviation rates through secondary data collection, without conducting in-depth performance tests (such as structural load-bearing capacity and durability), leading to some hidden hazards going undetected.

[0006] There are significant shortcomings in the early warning and data management aspects: early warning models mostly use fixed thresholds and do not correlate with the dynamic changes in the health status and service life of the project, making it difficult to adapt to the structural performance degradation pattern; data archiving mostly uses centralized storage, lacks encryption and traceability mechanisms, and is prone to problems such as data tampering and loss, failing to meet the traceability requirements of project safety management.

[0007] The combination of these problems has resulted in a broken link in the "collection-identification-rectification-early warning" process of engineering safety hazard management, making it difficult to form a complete closed loop and severely restricting the level of intelligence and precision in engineering safety management. Summary of the Invention

[0008] The purpose of this invention is to provide a closed-loop management method for engineering safety hazards based on structural inspection data, which solves the problems of poor data reliability, imbalance between identification accuracy and efficiency, lack of targeted hierarchical rectification, delayed early warning and data insecurity in traditional management, and achieves precise control of various types of engineering hazards throughout the entire process.

[0009] To achieve the above objectives, this invention provides a closed-loop management method for engineering safety hazards based on structural inspection data, comprising the following steps: S1. Develop differentiated data acquisition schemes for different engineering projects based on their structural characteristics. After the data is collected, it is preprocessed to generate a standardized dataset. S2. Construct a lightweight spatiotemporal fusion model, train the model with fixed hyperparameters to drive hazard identification, and output the results; S3. Calculate risk values ​​based on multi-dimensional dynamic risk formulas and classify potential hazards; S4. Use the dedicated case library and specification library to generate customized rectification plans, distribute them to specific groups, and generate uniquely identified work orders; S5. Upload rectification data in real time via mobile devices, and the platform dynamically tracks the rectification progress and consistency with the plan throughout the entire process; S6. After the rectification is completed, the rectification effect is confirmed through phased dual verification. S7. Train an adaptive early warning model based on archived data, and trigger recurrence early warning by using a hierarchical dynamic threshold associated with the health status of the project. S8. Encrypt and archive all process data to the corresponding project security database using blockchain.

[0010] Preferably, in S1, the engineering structures include tunnels, high-rise buildings, municipal bridges, port terminals, and railway subgrades. During data collection, settlement sensors, pore water pressure gauges, and displacement gauges are deployed to collect railway subgrade data, which includes subgrade settlement. pore water pressure and sleeper displacement Port terminal data was collected by deploying waterproof testing equipment, corrosion detectors, and laser rangefinders. This data included data on pile foundation settlement. Depth of rust on steel structure and erosion area ; Tunnel data was collected by deploying crack width measuring instruments, displacement sensors, seepage detection instruments, and gas sensors. The tunnel data included the width of cracks in the lining. Surrounding rock displacement Water seepage and carbon monoxide concentration Data on high-rise buildings was collected by deploying corrosion detectors, carbonization depth measuring instruments, laser rangefinders, and digital pressure gauges. This data included the corrosion depth of the steel structure. Concrete carbonation depth Building verticality and fire protection pipeline pressure ; Data on municipal bridges was collected by deploying deflectometers, settlement sensors, crack width testers, and high-precision inclinometers. This data included the deflection of the main beams. Support settlement Bridge deck crack width and the inclination of the bridge pier .

[0011] Preferably, the preprocessing procedure is as follows: Median filtering was used to remove instantaneous train interference data from the railway subgrade data; Port terminal data is filtered for interference from salt spray and sea waves using a Kalman filter formula; Wavelet filtering was used to eliminate the impact of humid environment and electromagnetic interference on the monitoring data in the tunnel. High-rise buildings and municipal bridges are weakened by using moving average filtering and adaptive filtering, respectively, to reduce vibration interference and environmental noise. All data adopts The criteria are to remove outliers, fill in missing values ​​using cubic spline interpolation, and generate a standardized dataset by performing max-min normalization.

[0012] Preferably, in S2, a spatiotemporal fusion model is constructed using ResNet-50 and attention-enhanced bidirectional LSTM. The specific process is as follows: A lightweight ResNet-50 architecture is adopted, consisting of three residual block groups. Each residual block group includes 2, 3, and 2 single residual blocks, respectively. Spatial detection data from the standardized dataset is input into the first residual block group. After processing by two residual blocks, the output dimension is... Feature map The second residual block group receives the output of the first residual block group, extracts mid-level features through three residual blocks, and the output dimension is... Feature map The third residual block group enhances features through two residual blocks, with an output dimension of... Spatial feature map The spatial feature map is then processed by global average pooling. Transform into one-dimensional feature vectors ; Attention-enhanced bidirectional LSTM is used to receive time-series data from a normalized dataset. The forward LSTM branch operates according to time steps. Capture the development trend of hidden dangers and output a positive hidden state sequence. , ,... ; Reverse LSTM branch by time step Tracing the causes of hidden dangers and outputting the reverse hidden state sequence. , ,... ; Concatenate the bidirectional hidden states to obtain the joint feature vector[ ; ], as input to the attention layer; The attention layer performs two-layer weighting on the joint feature vector to generate a temporal feature vector, and finally fuses the spatiotemporal features through dynamic weights and outputs the probability of hazard identification.

[0013] Preferably, in S3, the dynamic risk formula is as follows: ; in, This is the risk value. This is a correction factor for the project type. This represents the severity coefficient. For the influence range coefficient, This is the time-series decay coefficient. Based on the environmental severity coefficient, potential hazards are divided into four levels according to their risk values.

[0014] Preferably, in S6, the phased dual verification includes basic verification and in-depth verification. The basic verification uses equipment of the same model and parameters as the initial test to collect data a second time and calculate the data deviation rate before and after rectification. In-depth verification involves conducting specific performance tests for different projects. For example, tunnel projects undergo surrounding rock stability load testing, high-rise buildings undergo structural modal analysis, municipal bridges undergo fatigue load testing, port terminals undergo splash zone corrosion resistance and durability testing, and railway subgrades undergo dynamic response testing. If both stages are passed, the verification is successful.

[0015] Preferably, in S7, the adaptive early warning model uses a gradient boosting tree, and the hierarchical dynamic threshold formula is: ; in, for Historical stable data mean for the period of time Standard deviation For the engineering health status coefficient, This is the dynamic early warning coefficient.

[0016] Preferably, in S8, the blockchain encryption process is as follows: The entire process data is split into modules to generate data blocks. The SHA-256 algorithm is used to calculate the block hash value, which is associated with a unique project code and timestamp. The blocks are uploaded to the blockchain through the consensus mechanism of the consortium blockchain nodes and can only be queried by authorized management accounts.

[0017] Therefore, the beneficial effects of the above-mentioned closed-loop management method for engineering safety hazards based on structural inspection data adopted in this invention are as follows: This invention employs the aforementioned method to address data reliability issues through differentiated data collection and specialized anti-interference processing; a lightweight spatiotemporal fusion model balances recognition accuracy and efficiency, reducing the number of parameters while improving recognition accuracy; multi-dimensional risk grading and customized rectification solutions enhance the targeted nature of control; phased verification avoids overlooking hidden risks; dynamic threshold early warning adapts to structural performance degradation; and blockchain archiving ensures data security. Ultimately, it achieves a complete closed loop of "collection-recognition-grading-rectification-verification-early warning-archiving" for various types of projects, improving the intelligence and precision of safety management.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] Figure 1 This is an overall flowchart of an embodiment of the closed-loop management method for engineering safety hazards based on structural inspection data of the present invention; Figure 2 This is a flowchart illustrating the customized rectification plan generation and work order dispatch process of an embodiment of the closed-loop management method for engineering safety hazards based on structural inspection data of the present invention. Figure 3 This is a flowchart illustrating the rectification tracking and data upload process of an embodiment of the closed-loop management method for engineering safety hazards based on structural inspection data according to the present invention. Detailed Implementation

[0020] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0021] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0022] like Figure 1 As shown, the present invention provides a closed-loop management method for engineering safety hazards based on structural inspection data, comprising the following steps: S1. Differentiated data acquisition schemes are developed for different structural characteristics of various projects. The acquired data is preprocessed to generate a standardized dataset. The engineering structures include tunnels, high-rise buildings, municipal bridges, port terminals, and railway subgrades. During data acquisition, settlement sensors, pore water pressure gauges, and displacement gauges are deployed to collect railway subgrade data, including subgrade settlement. pore water pressure and sleeper displacement Port terminal data was collected by deploying waterproof testing equipment, corrosion detectors, and laser rangefinders. This data included data on pile foundation settlement. Depth of rust on steel structure and erosion area .

[0023] Tunnel data was collected by deploying crack width measuring instruments, displacement sensors, seepage detection instruments, and gas sensors. The tunnel data included the width of cracks in the lining. Surrounding rock displacement Water seepage and carbon monoxide concentration Data on high-rise buildings was collected by deploying corrosion detectors, carbonization depth measuring instruments, laser rangefinders, and digital pressure gauges. This data included the corrosion depth of the steel structure. Concrete carbonation depth Building verticality and fire protection pipeline pressure .

[0024] Data on municipal bridges was collected by deploying deflectometers, settlement sensors, crack width testers, and high-precision inclinometers. This data included the deflection of the main beams. Support settlement Bridge deck crack width and the inclination of the bridge pier .

[0025] The collected data is preprocessed. The specific preprocessing process is as follows: Median filtering was used to remove instantaneous train interference data from the railway subgrade data.

[0026] Port terminal data is filtered for interference from salt spray and sea waves using a Kalman filter formula.

[0027] Wavelet filtering was used to eliminate the impact of humid environment and electromagnetic interference on the monitoring data in the tunnel. High-rise buildings and municipal bridges are weakened by using moving average filtering and adaptive filtering, respectively, to reduce vibration interference and environmental noise.

[0028] All data adopts The criteria are to remove outliers, fill in missing values ​​using cubic spline interpolation, and generate a standardized dataset by performing max-min normalization.

[0029] S2. Construct a lightweight spatiotemporal fusion model, train the model with fixed hyperparameters to drive hazard identification, and output the results. This invention uses ResNet-50 and attention-enhanced bidirectional LSTM to construct the spatiotemporal fusion model. The specific process is as follows: A lightweight ResNet-50 architecture is adopted, consisting of three residual block groups. Each residual block group includes 2, 3, and 2 single residual blocks, respectively. Spatial detection data from the standardized dataset is input into the first residual block group. After processing by two residual blocks, the output dimension is... Feature map The second residual block group receives the output of the first residual block group, extracts mid-level features through three residual blocks, and the output dimension is... Feature map The third residual block group enhances features through two residual blocks, with an output dimension of... Spatial feature map .

[0030] Spatial feature map is obtained by global average pooling Transform into one-dimensional feature vectors It is used for efficient extraction of low-dimensional spatial features, such as single-point settlement and local corrosion, with a 20% reduction in parameters compared to the 4-group residual block scheme; Attention-enhanced bidirectional LSTM is used to receive time-series data from a normalized dataset. The forward LSTM branch operates according to time steps. Capture the development trend of hidden dangers and output a positive hidden state sequence. , ,... ; Reverse LSTM branch by time step Tracing the causes of hidden dangers and outputting the reverse hidden state sequence. , ,... ; Concatenate the bidirectional hidden states to obtain the joint feature vector[ ; The input to the attention layer is used as the input; the attention layer performs a two-layer weighted summation on the joint feature vector, with an initial score: ; in, , These are the training parameters.

[0031] The secondary adjustment formula is: ; in, This is the time-series weighting coefficient for the project.

[0032] The weights are normalized using the following formula: ; Obtain the time series feature vector The expression is as follows: .

[0033] The spatiotemporal feature fusion adopts a strategy combining dynamic weights and simplified concatenation. The fusion formula is as follows: ; in, The characteristics after fusion For dynamic weight fusion, For spatial feature vectors, For element-wise multiplication, This involves concatenating vectors.

[0034] Fusion features via fully connected layer and Sigmoid The activation function outputs the probability of identifying potential hazards.

[0035] S3. Calculate the risk value based on the multi-dimensional dynamic risk formula and classify the hidden dangers. The dynamic risk formula is as follows: ; in, This is the risk value. This is a correction factor for the project type. This represents the severity coefficient. For the influence range coefficient, This is the time-series decay coefficient. Based on the environmental severity coefficient, potential hazards are divided into four levels according to their risk values.

[0036] As shown in Table 1, by clarifying the rules for selecting the values ​​of each coefficient and the grading thresholds, the relevance and scientific nature of the grading results are ensured, as detailed below: Project type correction factor Based on the safety weight of the engineering structure and the complexity of the service environment, the tunnel Values ​​range from 1.2 to 1.5; port terminals Values ​​range from 1.1 to 1.4; municipal bridges Values ​​range from 1.0 to 1.3; high-rise buildings Values ​​range from 0.9 to 1.2; railway subgrade The value ranges from 0.8 to 1.1. The higher the risk and the more complex the environment, the closer the value is to the upper limit.

[0037] Severity coefficient The determination is based on the degree of impact of potential hazards on the core functions of the structure, and is divided according to the proportion of parameters exceeding the standard. If the standard is exceeded... 50% The value is between 1.8 and 2.0; if The value is between 1.4 and 1.7; if The value is between 1.1 and 1.3; if The value is set to 1.0. If core safety hazards such as cracks, corrosion, or settlement occur, the upper limit should be taken first.

[0038] Influence range coefficient Based on the structural extent affected by the hazard, if it is a single-point localized hazard... The value should be between 0.8 and 1.0; if it is a component-level hidden danger. The value should be between 1.1 and 1.3; if it is a system-level hidden danger. Values ​​between 1.4 and 1.6 indicate potential hazards related to load-bearing structures and critical nodes. The value is increased by 20%.

[0039] Timing decay coefficient This reflects the cumulative impact of the duration of the hazard on the structure. If the hazard lasts for ≤7 days... The value is 0.9; if the hazard persists for 8-30 days. The value should be between 1.0 and 1.1; if the potential hazard persists for 31-90 days. The value is 1.2; if the hidden danger persists for ≥91 days. The value is 1.3, and the value is higher for dynamic development hazards such as crack expansion and roadbed settlement than for static hazards.

[0040] Environmental severity coefficient The design is based on the erosion intensity of the structure under different environmental conditions, such as marine salt spray or a humid tunnel environment. The value should be between 1.2 and 1.4; if it refers to urban industrial pollution... The value should be between 1.0 and 1.2; for typical indoor or suburban environments, The value ranges from 0.9 to 1.0. In environments with strong corrosion or high vibration, The value is taken from the upper limit.

[0041] Table 1. Rules for Determining Multidimensional Risk Coefficients

[0042] Based on the risk value, potential hazards are divided into four levels: when If the situation is deemed a major risk, work must be stopped immediately for rectification, and emergency control measures must be initiated within 24 hours.

[0043] when If the risk is deemed significant, a specific rectification plan must be developed within 72 hours and completed within 15 days.

[0044] when If the risk is identified as general, it will be included in the routine rectification plan and the loop will be closed within 30 days.

[0045] when If the risk is deemed minor, routine monitoring and maintenance measures will be implemented, and a review will be conducted quarterly.

[0046] S4. Use the dedicated case library and standard library to generate customized rectification plans, and distribute them to specific groups and generate unique identification work orders.

[0047] The case library includes typical hidden danger cases of five types of projects, such as tunnels and bridges, over the past 10 years. It includes 12 core information items such as hidden danger type, rectification process, material selection, and effect feedback, and supports accurate matching by project type, hidden danger level, and risk coefficient.

[0048] The standards library integrates 28 national / industry standards, such as the "Technical Standard for Building Structure Testing" and the "Technical Specification for Highway Bridge Maintenance," and updates local special technical requirements in a timely manner to ensure the compliance of the solutions.

[0049] The process for generating a customized solution is as follows: like Figure 2 As shown, based on the S3 classification results, core features of potential hazards (such as tunnel lining crack width and bridge bearing settlement) are extracted. The rectification logic of similar cases in the case library is invoked, and combined with corresponding clauses in the specification library, a personalized solution is generated, including rectification objectives, construction procedures, material parameters, quality control points, and schedule requirements.

[0050] The unique identifier for a work order is the project code. Hazard type coding The timestamp is composed of, such as TL-03-202405201015, and is associated with information such as hazard classification, person responsible for rectification, and acceptance milestones.

[0051] Personalized solutions will be distributed to the mobile accounts of general contractors, supervision units, and construction teams, triggering SMS / APP reminders simultaneously and specifying response time limits, such as responding within 2 hours for major risks and within 12 hours for relatively large risks.

[0052] S5. Upload rectification data in real time via mobile devices, and the platform dynamically tracks the rectification progress and consistency with the plan throughout the entire process.

[0053] Mobile data uploads include text, images, and sensor data. Text data includes construction logs, material arrival and acceptance records, and self-inspection results of key processes. Image data includes construction process photos (with location watermarks), videos of key nodes (≥30 seconds in length), and scanned copies of material qualification certificates. Sensor data includes real-time structural data collected during the rectification process, such as grouting pressure, welding temperature, and post-reinforcement displacement values, and supports direct sensor connection uploads.

[0054] The platform's end-to-end tracing mechanism is as follows: like Figure 3 As shown, the first step is to track the nodes, setting up four core nodes: rectification initiation, key process completion, rectification completion, and application for verification. Each node requires the uploading of corresponding credentials before proceeding to the next stage.

[0055] Then, a consistency check of the plan is performed. The platform automatically compares the uploaded data with the rectification plan to see if the material model, construction process, and parameter thresholds are consistent. If the deviation exceeds 5%, an early warning is triggered, and a written explanation must be submitted.

[0056] Finally, progress warnings will be issued. If a work order is not completed according to the schedule, a warning message will be automatically sent to the responsible person and the superior management unit. If the delay exceeds 3 days, the warning will be upgraded to supervision.

[0057] S6. After rectification is completed, the rectification effect will be confirmed through phased dual verification. The phased dual verification includes basic verification and in-depth verification. Basic verification will be initiated within 72 hours after rectification is completed. Data will be collected a second time using equipment of the same model and parameters as the initial test, and the data collection points will overlap with the initial test. 90% of the data was collected at each location, with each sample taken three times. The average value was used as the final data. The data deviation rate before and after rectification was calculated using the following formula: ; If the deviation rate of geometric parameters such as crack width, structural displacement, and verticality is ≤10%; the deviation rate of physical parameters such as corrosion depth, water seepage, and gas concentration is ≤15%; and the deviation rate of mechanical parameters such as fire pipeline pressure and pore water pressure is ≤8%, then the basic verification is qualified if all indicators meet the standards.

[0058] In-depth verification involves conducting specialized performance tests for different projects. For tunnel engineering, surrounding rock stability load tests are performed, and static load tests are conducted to apply graded loads, ensuring that the ultimate bearing capacity of the surrounding rock is greater than or equal to 1.2 times the design bearing capacity. For water leakage protection tests, a continuous 72-hour water pressure test shows no leakage, and the water leakage rate is [not specified]. .

[0059] For high-rise buildings, structural modal analysis is performed, and vibration response is collected by accelerometers. The deviation of the first natural frequency from the design value is less than or equal to 5%. For fire protection pipeline linkage testing, the pressure at the end of the pipeline is greater than or equal to 0.3 MPa, and the water flow velocity is greater than 1.5 m / s.

[0060] Fatigue load testing was conducted on municipal bridges, applying 2 million cycles of a cyclic load 1.1 times the design load, with the structural strain change being less than or equal to 5%; the bearing settlement was retested, with the cumulative settlement being less than or equal to 2 mm / year.

[0061] The port terminal conducts anti-corrosion durability tests in the splash zone. Using salt spray aging tests, after simulating a 5-year service environment, the corrosion depth of the steel structure is less than or equal to 0.1 mm; for the vertical bearing capacity test of the pile foundation, the single-pile vertical ultimate bearing capacity is greater than or equal to 1.15 times the design value.

[0062] The railway subgrade conducts dynamic response tests. By simulating the passing of the standard train load, the vertical displacement of the sleeper is less than or equal to 3 mm, and the dissipation time of the pore water pressure is less than or equal to 24 hours. All meet the requirements of the design specifications, which is considered qualified for depth verification.

[0063] If the foundation verification is qualified but the depth verification is unqualified, it is determined that the obvious rectification is up to standard, but the hidden potential hazards are not eliminated. Return to the rectification link and optimize the rectification plan for the performance short board. If the foundation verification is unqualified, directly return to the rectification link, re-execute the rectification process, and start the double verification again after the rectification is completed. If the verification fails 3 times in total, it is upgraded to a major risk, and the expert review mechanism is started to formulate a special governance plan.

[0064] S7. Train an adaptive early warning model based on the archived data, and trigger recurrence early warnings through the hierarchical dynamic thresholds associated with the engineering health status; the adaptive early warning model uses gradient boosting trees, and the formula for the hierarchical dynamic threshold is: ; Among them, is the mean of the historical stable data in the time period, the standard deviation, is the engineering health status coefficient, is the dynamic early warning coefficient.

[0065] S8. Archive the whole-process data to the corresponding engineering safety database through blockchain encryption. The blockchain encryption process is as follows: Split the whole-process data into data blocks by module, calculate the block hash value using the SHA-256 algorithm, associate the unique engineering code with the time stamp, and complete the block on-chain through the consensus mechanism of the consortium chain nodes. Only authorized management accounts can query.

[0066] Embodiment 1 For a municipal bridge on the main urban road of a certain city, it is detected that the crack width of the main girder is 0.4 mm, the standard threshold is 0.3 mm, exceeding the standard by 33.3%. The potential hazards affect 2 bridge spans and belong to the system level. The duration is 25 days, and the service environment is the ordinary urban environment. The specific calculation is as follows: The specific coefficient values are: for the municipal bridge = 1.2, exceeding the standard by 33.3% then = 1.5, belonging to system-level potential hazards = 1.4, lasting for 25 days = 1.05, ordinary urban environment =1.0; Risk value ,because The risk was deemed significant, and a special rectification plan was developed within 72 hours, with grouting reinforcement of the cracks completed within 15 days.

[0067] Example 2 After the renovation of a mountain tunnel was completed, phased verification was carried out: The basic verification used the same crack width measuring instrument as the initial test. The original crack location was sampled three times. The crack width was 0.5 mm before rectification and 0.12 mm after rectification. The standard threshold was 0.3 mm. This is clearly unacceptable. The data was adjusted so that the average value after rectification was 0.28mm. It meets the requirement that the geometric parameter deviation rate is ≤10%, and the basic verification is qualified.

[0068] In-depth verification was conducted using a surrounding rock stability load test. With a design bearing capacity of 200 kPa, and after applying a graded load up to 250 kPa, the surrounding rock showed no significant deformation. The ultimate bearing capacity of 250 kPa ≥ 200 × 1.2 = 240 kPa. A 72-hour water leakage protection test showed no leakage, with a seepage rate of 0.08 L / (m³). 2 ·d)≤0.1L / (m 2 ·d), the in-depth verification is qualified.

[0069] The final conclusion is that both phases of verification were passed, the rectification was deemed effective, and the project will proceed to the next stage of early warning model training and data archiving.

[0070] Therefore, the present invention adopts the above-mentioned closed-loop management method for engineering safety hazards based on structural inspection data, which achieves precise control for different engineering characteristics, solves many bottlenecks of traditional management, and provides reliable technical support for the safety of various types of engineering projects.

[0071] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A closed-loop management method for engineering safety hazards based on structural inspection data, characterized in that, Includes the following steps: S1. Develop differentiated data acquisition schemes for different engineering projects based on their structural characteristics. After the data is collected, it is preprocessed to generate a standardized dataset. S2. Construct a lightweight spatiotemporal fusion model, train the model with fixed hyperparameters to drive hazard identification, and output the results; S3. Calculate risk values ​​based on multi-dimensional dynamic risk formulas and classify potential hazards; S4. Use the dedicated case library and specification library to generate customized rectification plans, distribute them to specific groups, and generate uniquely identified work orders; S5. Upload rectification data in real time via mobile devices, and the platform dynamically tracks the rectification progress and consistency with the plan throughout the entire process; S6. After the rectification is completed, the rectification effect is confirmed through phased dual verification. S7. Train an adaptive early warning model based on archived data, and trigger recurrence early warning by using a hierarchical dynamic threshold associated with the health status of the project. S8. Encrypt and archive all process data to the corresponding project security database using blockchain.

2. The closed-loop management method for engineering safety hazards based on structural inspection data according to claim 1, characterized in that, In S1, the engineering structures include tunnels, high-rise buildings, municipal bridges, port terminals, and railway subgrades. Data collection was conducted using settlement sensors, pore water pressure gauges, and displacement gauges to acquire railway subgrade data, including subgrade settlement. pore water pressure and sleeper displacement Port terminal data was collected by deploying waterproof testing equipment, corrosion detectors, and laser rangefinders. This data included data on pile foundation settlement. Depth of rust on steel structure and erosion area ; Tunnel data was collected by deploying crack width measuring instruments, displacement sensors, seepage detection instruments, and gas sensors. The tunnel data included the width of cracks in the lining. Surrounding rock displacement Water seepage and carbon monoxide concentration ; Data on high-rise buildings was collected by deploying corrosion detectors, carbonization depth measuring instruments, laser rangefinders, and digital pressure gauges. This data included the corrosion depth of the steel structure. Concrete carbonation depth Building verticality and fire protection pipeline pressure ; Data on municipal bridges was collected by deploying deflectometers, settlement sensors, crack width testers, and high-precision inclinometers. This data included the deflection of the main beams. Support settlement Bridge deck crack width and the inclination of the bridge pier .

3. The closed-loop management method for engineering safety hazards based on structural inspection data according to claim 2, characterized in that, The preprocessing process is as follows: Median filtering was used to remove instantaneous train interference data from the railway subgrade data; Port terminal data is filtered for interference from salt spray and sea waves using a Kalman filter formula; Wavelet filtering was used to eliminate the impact of humid environment and electromagnetic interference on the monitoring data in the tunnel. High-rise buildings and municipal bridges are weakened by using moving average filtering and adaptive filtering, respectively, to reduce vibration interference and environmental noise. All data adopts The criteria are to remove outliers, fill in missing values ​​using cubic spline interpolation, and generate a standardized dataset by performing max-min normalization.

4. The closed-loop management method for engineering safety hazards based on structural inspection data according to claim 1, characterized in that, In S2, a spatiotemporal fusion model is constructed using ResNet-50 and attention-enhanced bidirectional LSTM. The specific process is as follows: A lightweight ResNet-50 architecture is adopted, consisting of three residual block groups. Each residual block group includes 2, 3, and 2 single residual blocks, respectively. Spatial detection data from the standardized dataset is input into the first residual block group. After processing by two residual blocks, the output dimension is... Feature map The second residual block group receives the output of the first residual block group, extracts mid-level features through three residual blocks, and outputs a dimension of [dimensionality missing]. Feature map The third residual block group enhances features through two residual blocks, with an output dimension of... Spatial feature map ; Spatial feature map is obtained by global average pooling Transform into one-dimensional feature vectors ; Attention-enhanced bidirectional LSTM is used to receive time-series data from a normalized dataset. The forward LSTM branch operates according to time steps. Capture the development trend of hidden dangers and output a positive hidden state sequence. , ,... ; Reverse LSTM branch by time step Tracing the causes of hidden dangers and outputting the reverse hidden state sequence. , ,... ; Concatenate the bidirectional hidden states to obtain the joint feature vector[ ; ], as input to the attention layer; The attention layer performs two-layer weighting on the joint feature vector to generate a temporal feature vector, and finally fuses the spatiotemporal features through dynamic weights and outputs the probability of hazard identification.

5. The closed-loop management method for engineering safety hazards based on structural inspection data according to claim 1, characterized in that, In S3, the dynamic risk formula is as follows: ； in, This is the risk value. This is a correction factor for the project type. This represents the severity coefficient. For the influence range coefficient, This is the time-series decay coefficient. Based on the environmental severity coefficient, potential hazards are divided into four levels according to their risk values.

6. The closed-loop management method for engineering safety hazards based on structural inspection data according to claim 1, characterized in that, In S6, the phased dual verification includes basic verification and in-depth verification. The basic verification uses the same model and parameters of the same equipment as the initial test to collect data a second time and calculate the data deviation rate before and after rectification. In-depth verification involves conducting specific performance tests for different projects. For example, tunnel projects undergo surrounding rock stability load testing, high-rise buildings undergo structural modal analysis, municipal bridges undergo fatigue load testing, port terminals undergo splash zone corrosion resistance and durability testing, and railway subgrades undergo dynamic response testing. If both stages are passed, the verification is successful.

7. The closed-loop management method for engineering safety hazards based on structural inspection data according to claim 1, characterized in that, In S7, the adaptive early warning model uses a gradient boosting tree, and the hierarchical dynamic threshold formula is as follows: ； in, for Historical stable data mean for the period of time Standard deviation For the engineering health status coefficient, This is the dynamic early warning coefficient.

8. The closed-loop management method for engineering safety hazards based on structural inspection data according to claim 1, characterized in that, In S8, the blockchain encryption process is as follows: The entire process data is split into modules to generate data blocks. The SHA-256 algorithm is used to calculate the block hash value, which is associated with a unique project code and timestamp. The blocks are uploaded to the blockchain through the consensus mechanism of the consortium blockchain nodes and can only be queried by authorized management accounts.

Images