Engineering construction safety monitoring method

Abstract

This invention relates to the field of civil engineering technology, specifically to a method for monitoring construction safety, comprising the following steps: Step 1: Construction of a four-dimensional monitoring system and deployment of sensors, S1.1 Determining monitoring dimensions and indicators: Structural safety dimension: monitoring foundation pit settlement (accuracy ±0.1mm), slope displacement (accuracy ±0.5mm), tower crane verticality (deviation ≤0.3‰), support axial force (range 0-5000kN), and bridge deflection (accuracy ±0.2mm); Equipment operation dimension: monitoring tower crane lifting capacity, lifting height, slewing angle, wire rope wear (accuracy ±0.1mm), construction elevator operating speed, and braking performance; This invention constructs a four-dimensional monitoring system of "structural safety + equipment operation + personnel behavior + environmental parameters," integrating data from multiple sensors to improve the accuracy of identifying complex hidden dangers. Employing edge computing + 5G transmission technology, the data acquisition and platform diagnostic response time is ≤3 seconds, and the hazard warning response time is ≤10 seconds, gaining crucial time for handling emergencies.

Description

Technical Field

[0001] This invention relates to the field of civil engineering technology, specifically to a method for monitoring safety during engineering construction. Background Technology

[0002] Construction safety monitoring is a core means of ensuring the safety of personnel, equipment, and structures during the construction of buildings, rail transit, bridges, tunnels, and other engineering projects. Existing construction safety monitoring technologies have significant drawbacks:

[0003] Single monitoring dimensions and lagging hazard identification: Traditional monitoring focuses on a single indicator, such as monitoring only the settlement of foundation pits and the verticality of tower cranes. It lacks collaborative monitoring of personnel violations, equipment operating status and environmental risks, and cannot identify complex hazards of "structural deformation + personnel violations". The failure rate is as high as 15% or more.

[0004] High rate of manual intervention and poor real-time performance: Data collection relies on manual reading and recording, data transmission is delayed, and the emergency response time exceeds 30 minutes, making it difficult to cope with sudden accidents such as foundation pit collapse and tower crane overturning.

[0005] Data is isolated and lacks linkage, resulting in a lack of closed-loop processing: Monitoring data is stored in different systems and cannot be linked with the construction management platform and emergency command system. Early warnings are only at the "reminder" level and have not formed a closed-loop process of "monitoring-diagnosis-early warning-processing-verification".

[0006] Poor environmental adaptability and weak anti-interference ability: The sensor is easily affected by construction vibration, electromagnetic interference, and severe weather, resulting in low data accuracy and an error exceeding 20% ​​in strong electromagnetic environments;

[0007] Low level of intelligence and lack of predictive ability: It can only realize "over-limit alarm", but cannot predict the trend of structural deformation and the probability of equipment failure based on historical data, making it difficult to avoid risks in advance.

[0008] Existing technologies (such as single-point IoT monitoring and video surveillance) have achieved partial automated data collection, but have not formed a complete process system of "multi-source sensor fusion + AI intelligent diagnosis + smart platform linkage". In particular, they lack the ability to identify complex risks in complex construction scenarios. Therefore, a method for monitoring engineering construction safety is proposed. Summary of the Invention

[0009] In view of this, the present invention provides a method for monitoring the safety of engineering construction, so as to solve or alleviate the technical problems existing in the prior art, and at least provide a beneficial option.

[0010] The technical solution of this invention is implemented as follows: a method for monitoring safety during engineering construction, comprising the following steps:

[0011] Step 1: Construction of the four-dimensional monitoring system and deployment of sensors

[0012] S1.1 Determine monitoring dimensions and indicators:

[0013] Structural safety dimensions: monitoring foundation pit settlement (accuracy ±0.1mm), slope displacement (accuracy ±0.5mm), tower crane verticality (deviation ≤0.3‰), support axial force (range 0-5000kN), and bridge deflection (accuracy ±0.2mm);

[0014] Equipment operation dimensions: monitoring tower crane lifting capacity, lifting height, slewing angle, wire rope wear (accuracy ±0.1mm), construction elevator operating speed, and braking performance;

[0015] Personnel behavior dimension: Monitoring personnel's wearing of safety helmets, fastening of safety belts, entry into restricted areas, and unauthorized use of open flames;

[0016] Environmental parameters include: wind speed (range 0-60m / s), rainfall (accuracy ±0.1mm), temperature (-40℃~+85℃), humidity (0—100%RH), dust concentration (range 0-10mg / m³), and harmful gas concentration (such as CO, H2S, range 0-100ppm).

[0017] S1.2 Sensor Selection and Deployment:

[0018] Structural safety sensors: High-precision hydrostatic level, GNSS displacement monitor, and vibrating wire axial force gauge are used and deployed around the foundation pit, tower crane body, and key parts of the main beam of the bridge, with a protection level ≥ IP67;

[0019] Equipment operation sensors: Utilizing tension sensors, angle sensors, and laser rangefinders, these sensors are integrated into equipment such as tower cranes and construction elevators, and support integration with equipment control systems.

[0020] Personnel behavior sensors: AI smart cameras (≥8 million pixels) and UWB positioning tags (positioning accuracy ±10cm) are deployed at construction area entrances and exits and high-risk work areas;

[0021] Environmental sensors: Ultrasonic anemometers, rain gauges, dust detectors, and gas sensors are used and deployed at high points, pit edges, and densely populated areas of the construction site.

[0022] S1.3 System Calibration: All sensors are calibrated using standard instruments and time synchronization is achieved using the PTPv2 clock synchronization protocol with a synchronization accuracy of ≤100ns;

[0023] Step 2: Multi-source data acquisition and edge computing preprocessing

[0024] S2.1 Data Acquisition: Each sensor acquires data at a preset frequency, with the structural safety sensor sampling frequency ≥1Hz, the equipment operation sensor ≥10Hz, the personnel behavior camera frame rate ≥25fps, and the environmental sensor ≥0.5Hz.

[0025] S2.2 Edge Computing Preprocessing: Deploy edge computing gateways at the construction site to achieve localized data processing.

[0026] Data cleaning: Outliers are removed using the 3σ criterion, and vibration interference is eliminated using Kalman filtering;

[0027] Feature extraction: Settlement rate and displacement acceleration are extracted from structural data; load fluctuation coefficient and wire rope wear rate are extracted from equipment data; violation behavior feature values ​​are extracted from personnel data; wind speed change rate and cumulative rainfall value are extracted from environmental data.

[0028] Data compression: Wavelet transform is used to compress data with a compression ratio of ≥10:1 to ensure transmission efficiency;

[0029] S2.3 Secure Transmission: It adopts dual-link redundant transmission of "5G + Industrial Ethernet". The data is uploaded to the smart monitoring platform after being encrypted by AES-256. It supports breakpoint resume and local storage of data (capacity ≥1TB) when the network is interrupted.

[0030] Step 3: AI-integrated diagnosis and hierarchical early warning

[0031] S3.1 Multi-source data fusion: The intelligent monitoring platform fuses pre-processed structured data with unstructured data (video images) to construct a four-dimensional monitoring database;

[0032] S3.2 AI Intelligent Diagnosis:

[0033] Real-time hazard identification: A CNN model is used to identify personnel violations and abnormal equipment conditions; an LSTM model is used to analyze structural deformation trends and equipment failure probabilities, with a diagnostic delay of ≤1 second;

[0034] Composite Hazard Identification: Construct an association rule mining model to identify composite risks such as "excessive settlement of foundation pit + rainstorm" and "vertical deviation of tower crane + strong wind", with an accuracy rate of ≥99%;

[0035] S3.3 Graded Early Warning: Divided into three levels according to the severity of the hidden dangers:

[0036] Level I Warning (Red, Emergency Situation): Such as the settlement rate of the foundation pit ≥5mm / h, the verticality deviation of the tower crane ≥0.5‰, or personnel entering the restricted area and starting a fire, directly endangering safety;

[0037] Level II Warning (Yellow, Significant Potential Hazard): If the settlement rate is ≥2mm / h, the wire rope wear is ≥3mm, or the wind speed is ≥15m / s, it may develop into a dangerous situation.

[0038] Level III Warning (Blue, General Hazard): If personnel are not wearing safety helmets or dust concentration exceeds the standard, a reminder to rectify the situation is required.

[0039] S3.4 Warning methods: Platform pop-up window + audible and visual alarm + mobile APP push + SMS notification. Warning information includes hazard type, location, risk level, and handling suggestions.

[0040] Step 4: Intelligent Collaborative Response and Closed-Loop Verification

[0041] S4.1 Hierarchical linkage response:

[0042] Level I warning: The platform immediately cuts off the power supply to relevant equipment (such as tower cranes and construction elevators), locks the access control of high-risk work areas, automatically generates emergency work orders and dispatches them to project leaders and emergency rescue teams, and simultaneously reports to the housing and construction department;

[0043] Level II Warning: A rectification work order is generated and dispatched to the responsible team, requiring them to handle the situation within 2 hours;

[0044] Level III Warning: On-site broadcast reminder, safety officer on-site supervision and rectification;

[0045] S4.2 Process Tracking: Maintenance personnel upload photos and videos of the process through a mobile app, and the platform tracks the progress in real time.

[0046] S4.3 Closed-loop verification: After the handling is completed, the platform instructs the sensors to recollect data, and the AI ​​model verifies whether the hidden danger has been eliminated. If the verification is successful, the hidden danger status is updated to "closed-loop". If it fails, the warning level is upgraded.

[0047] Step 5: Model Iteration and Monitoring Optimization

[0048] S5.1AI Model Incremental Training: New hidden danger cases and handling data are added to the training set every month to incrementally train the CNN-LSTM model. The iteration cycle is ≤72 hours, and the diagnostic accuracy continues to improve.

[0049] S5.2 Monitoring Parameter Optimization: Adjust the threshold values ​​of monitoring indicators according to the construction stage, such as lowering the settlement warning threshold during the foundation pit excavation stage; adjust the environmental warning thresholds according to the season, such as lowering the wind speed warning threshold during the typhoon season.

[0050] S5.3 Sensor Deployment Optimization: Based on fault statistics, adjust the location and number of sensors. For example, if there is abnormal settlement in a certain area, increase the deployment density of hydrostatic levels.

[0051] Preferably, the UWB positioning tag in step 1 supports integration with a safety helmet, has a one-click alarm function, and the alarm signal upload delay is ≤1 second.

[0052] More preferably, the edge computing gateway in step 2 adopts an industrial-grade ARM architecture, has a protection level of ≥IP65, and supports wide operating temperature range of -40℃ to +70℃.

[0053] Further preferably, the composite hazard identification model in step 3 supports custom association rules, and risk association parameters can be configured according to different project types (buildings / rail transit / bridges).

[0054] Preferably, the emergency work order in step 4 includes the three-dimensional coordinates of the hazard location, an emergency response plan, and an emergency resource distribution map, which can be viewed offline.

[0055] More preferably, the harmful gas sensor in step 1 supports the simultaneous detection of multiple gases, with a response time ≤30 seconds and a resolution ≤0.1ppm.

[0056] Preferably, the early warning information in step 3 supports linkage with the construction site broadcasting system, access control system, and equipment control system to achieve automated emergency response.

[0057] In a further preferred embodiment, the incremental training of the model in step 5 adopts the transfer learning method. When there is insufficient new engineering data, it can be quickly adapted based on existing engineering data to reduce the amount of training data required.

[0058] More preferably, the data compression in step 2 adopts a combination of wavelet transform and Huffman coding, with a compression ratio of up to 15:1, without losing key feature data.

[0059] The embodiments of the present invention have the following advantages due to the adoption of the above technical solutions:

[0060] I. This invention constructs a four-dimensional monitoring system of "structural safety + equipment operation + personnel behavior + environmental parameters", integrates data from multiple sensors, improves the accuracy of identifying complex hidden dangers, and adopts edge computing + 5G transmission technology. The data acquisition and platform diagnosis response time is ≤3 seconds, and the emergency warning response time is ≤10 seconds, winning golden time for handling sudden accidents.

[0061] Second, the innovative CNN-LSTM fusion model of this invention can not only identify real-time hidden dangers, but also predict structural deformation trends and equipment failure probabilities, and issue early warnings 12-24 hours in advance, improving the hidden danger prevention and control rate by 80%. Closed-loop linkage and management efficiency are improved: it is deeply linked with the construction management platform and emergency command system, automatically generates disposal work orders, dispatches emergency resources, and automatically reviews and verifies after disposal, forming a closed-loop process, improving the efficiency of operation and maintenance collaboration by 200%.

[0062] Third, the sensor of this invention adopts a vibration-resistant and electromagnetic interference-resistant design, combined with a data filtering algorithm, achieving a data accuracy of ≥98% in strong electromagnetic and high vibration environments, making it suitable for complex construction scenarios such as buildings, rail transit, bridges and tunnels.

[0063] The above overview is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of the invention will become readily apparent from the accompanying drawings and the following detailed description. Attached Figure Description

[0064] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0065] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0066] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0067] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0068] like Figure 1 As shown in the figure, this embodiment of the invention provides a method for monitoring safety during engineering construction, including the following steps:

[0069] Step 1: Construction of the four-dimensional monitoring system and deployment of sensors

[0070] S1.1 Determine monitoring dimensions and indicators:

[0071] Structural safety dimensions: monitoring foundation pit settlement (accuracy ±0.1mm), slope displacement (accuracy ±0.5mm), tower crane verticality (deviation ≤0.3‰), support axial force (range 0-5000kN), and bridge deflection (accuracy ±0.2mm);

[0072] Equipment operation dimensions: monitoring tower crane lifting capacity, lifting height, slewing angle, wire rope wear (accuracy ±0.1mm), construction elevator operating speed, and braking performance;

[0073] Personnel behavior dimension: Monitoring personnel's wearing of safety helmets, fastening of safety belts, entry into restricted areas, and unauthorized use of open flames;

[0074] Environmental parameters include: wind speed (range 0-60m / s), rainfall (accuracy ±0.1mm), temperature (-40℃~+85℃), humidity (0—100%RH), dust concentration (range 0-10mg / m³), and harmful gas concentration (such as CO, H2S, range 0-100ppm).

[0075] S1.2 Sensor Selection and Deployment:

[0076] Structural safety sensors: High-precision hydrostatic level, GNSS displacement monitor, and vibrating wire axial force gauge are used and deployed around the foundation pit, tower crane body, and key parts of the main beam of the bridge, with a protection level ≥ IP67;

[0077] Equipment operation sensors: Utilizing tension sensors, angle sensors, and laser rangefinders, these sensors are integrated into equipment such as tower cranes and construction elevators, and support integration with equipment control systems.

[0078] Personnel behavior sensors: AI smart cameras (≥8 million pixels) and UWB positioning tags (positioning accuracy ±10cm) are deployed at construction area entrances and exits and high-risk work areas;

[0079] Environmental sensors: Ultrasonic anemometers, rain gauges, dust detectors, and gas sensors are used and deployed at high points, pit edges, and densely populated areas of the construction site.

[0080] S1.3 System Calibration: All sensors are calibrated using standard instruments and time synchronization is achieved using the PTPv2 clock synchronization protocol with a synchronization accuracy of ≤100ns;

[0081] Step 2: Multi-source data acquisition and edge computing preprocessing

[0082] S2.1 Data Acquisition: Each sensor acquires data at a preset frequency, with the structural safety sensor sampling frequency ≥1Hz, the equipment operation sensor ≥10Hz, the personnel behavior camera frame rate ≥25fps, and the environmental sensor ≥0.5Hz.

[0083] S2.2 Edge Computing Preprocessing: Deploy edge computing gateways at the construction site to achieve localized data processing.

[0084] Data cleaning: Outliers are removed using the 3σ criterion, and vibration interference is eliminated using Kalman filtering;

[0085] Feature extraction: Settlement rate and displacement acceleration are extracted from structural data; load fluctuation coefficient and wire rope wear rate are extracted from equipment data; violation behavior feature values ​​are extracted from personnel data; wind speed change rate and cumulative rainfall value are extracted from environmental data.

[0086] Data compression: Wavelet transform is used to compress data with a compression ratio of ≥10:1 to ensure transmission efficiency;

[0087] S2.3 Secure Transmission: It adopts dual-link redundant transmission of "5G + Industrial Ethernet". The data is uploaded to the smart monitoring platform after being encrypted by AES-256. It supports breakpoint resume and local storage of data (capacity ≥1TB) when the network is interrupted.

[0088] Step 3: AI-integrated diagnosis and hierarchical early warning

[0089] S3.1 Multi-source data fusion: The intelligent monitoring platform fuses pre-processed structured data with unstructured data (video images) to construct a four-dimensional monitoring database;

[0090] S3.2 AI Intelligent Diagnosis:

[0091] Real-time hazard identification: A CNN model is used to identify personnel violations and abnormal equipment conditions; an LSTM model is used to analyze structural deformation trends and equipment failure probabilities, with a diagnostic delay of ≤1 second;

[0092] Composite Hazard Identification: Construct an association rule mining model to identify composite risks such as "excessive settlement of foundation pit + rainstorm" and "vertical deviation of tower crane + strong wind", with an accuracy rate of ≥99%;

[0093] S3.3 Graded Early Warning: Divided into three levels according to the severity of the hidden dangers:

[0094] Level I Warning (Red, Emergency Situation): Such as the settlement rate of the foundation pit ≥5mm / h, the verticality deviation of the tower crane ≥0.5‰, or personnel entering the restricted area and starting a fire, directly endangering safety;

[0095] Level II Warning (Yellow, Significant Potential Hazard): If the settlement rate is ≥2mm / h, the wire rope wear is ≥3mm, or the wind speed is ≥15m / s, it may develop into a dangerous situation.

[0096] Level III Warning (Blue, General Hazard): If personnel are not wearing safety helmets or dust concentration exceeds the standard, a reminder to rectify the situation is required.

[0097] S3.4 Warning methods: Platform pop-up window + audible and visual alarm + mobile APP push + SMS notification. Warning information includes hazard type, location, risk level, and handling suggestions.

[0098] Step 4: Intelligent Collaborative Response and Closed-Loop Verification

[0099] S4.1 Hierarchical linkage response:

[0100] Level I warning: The platform immediately cuts off the power supply to relevant equipment (such as tower cranes and construction elevators), locks the access control of high-risk work areas, automatically generates emergency work orders and dispatches them to project leaders and emergency rescue teams, and simultaneously reports to the housing and construction department;

[0101] Level II Warning: A rectification work order is generated and dispatched to the responsible team, requiring them to handle the situation within 2 hours;

[0102] Level III Warning: On-site broadcast reminder, safety officer on-site supervision and rectification;

[0103] S4.2 Process Tracking: Maintenance personnel upload photos and videos of the process through a mobile app, and the platform tracks the progress in real time.

[0104] S4.3 Closed-loop verification: After the handling is completed, the platform instructs the sensors to recollect data, and the AI ​​model verifies whether the hidden danger has been eliminated. If the verification is successful, the hidden danger status is updated to "closed-loop". If it fails, the warning level is upgraded.

[0105] Step 5: Model Iteration and Monitoring Optimization

[0106] S5.1AI Model Incremental Training: New hidden danger cases and handling data are added to the training set every month to incrementally train the CNN-LSTM model. The iteration cycle is ≤72 hours, and the diagnostic accuracy continues to improve.

[0107] S5.2 Monitoring Parameter Optimization: Adjust the threshold values ​​of monitoring indicators according to the construction stage, such as lowering the settlement warning threshold during the foundation pit excavation stage; adjust the environmental warning thresholds according to the season, such as lowering the wind speed warning threshold during the typhoon season.

[0108] S5.3 Sensor Deployment Optimization: Based on fault statistics, adjust the location and number of sensors. For example, if there is abnormal settlement in a certain area, increase the deployment density of hydrostatic levels.

[0109] In one embodiment, the UWB positioning tag in step 1 supports integration with a safety helmet, has a one-click alarm function, and the alarm signal upload delay is ≤1 second.

[0110] In one embodiment, the edge computing gateway in step 2 adopts an industrial-grade ARM architecture, has a protection level of ≥IP65, and supports wide operating temperature range of -40℃ to +70℃.

[0111] In one embodiment, the composite hazard identification model in step 3 supports custom association rules, and risk association parameters can be configured according to different project types (buildings / rail transit / bridges).

[0112] In one embodiment, the emergency work order in step 4 includes the three-dimensional coordinates of the hazard location, the emergency response plan, and the emergency resource distribution map, which can be viewed offline.

[0113] In one embodiment, the hazardous gas sensor in step 1 supports the simultaneous detection of multiple gases, with a response time ≤30 seconds and a resolution ≤0.1ppm.

[0114] In one embodiment, the early warning information in step 3 can be linked with the construction site broadcasting system, access control system, and equipment control system to achieve automated emergency response.

[0115] In one embodiment, the incremental training of the model in step 5 adopts the transfer learning method. When there is insufficient new engineering data, it can be quickly adapted based on existing engineering data to reduce the amount of training data required.

[0116] In one embodiment, the data compression in step 2 uses a combination of wavelet transform and Huffman coding, achieving a compression ratio of up to 15:1 without losing key feature data.

[0117] In one embodiment, this embodiment takes the construction of the foundation pit of Swan Lake Station of Phase I of Hefei Metro Line 8 as the application scenario. The foundation pit is constructed by open excavation, with a depth of 28m, a length of 120m, and a width of 25m. It is adjacent to Swan Lake Park and high-rise building complex, and the construction risk level is Level I (extremely high risk).

[0118] Traditional monitoring uses a combination of manual total station measurement and manual inspection, which has problems such as data lag (a single monitoring session takes 2 hours), weak ability to identify complex hazards, and slow emergency response. This embodiment applies a safety monitoring method of "multi-source sensor fusion + AI intelligent diagnosis + smart linkage" to verify the practicality and advancement of the method and ensure the safety of personnel, structures and equipment during foundation pit construction.

[0119] II. Implementation of Hardware and Software Configuration

[0120] (a) Hardware configuration

[0121] Sensor system

[0122] Structural safety sensors: 15 static levels (model: SJJ-900, accuracy ±0.1mm), deployed on top of the retaining piles of the foundation pit, spaced 8m apart; 6 GNSS displacement monitors (model: TrimbleR10, positioning accuracy ±5mm), deployed on the rooftops of 3 surrounding high-rise buildings; 20 vibrating wire axial force gauges (model: ZX-3000, range 0-5000kN), installed at both ends of the steel supports;

[0123] Equipment operation sensors: 2 tower crane tension sensors (model: LF-100, range 0-10t), installed on 2 QTZ80 tower crane hooks; 2 wire rope wear sensors (model: WMC-200, accuracy ±0.1mm), wound on the tower crane wire rope; 1 construction elevator speed sensor (model: SD-50, range 0-2m / s), integrated into the construction elevator control system;

[0124] Personnel behavior sensors: 10 AI smart cameras (model: Hikvision DS-2CD864FWD, 8 megapixels, 25fps) deployed at the pit entrance and exit, tower crane operation area, and edge protection area; 50 UWB positioning tags (model: DW1000, positioning accuracy ±10cm) integrated into the safety helmets of construction workers.

[0125] Environmental sensors: 2 ultrasonic anemometers (model: FS-300, range 0-60m / s), installed on the top of the tower crane; 3 gas sensors (model: GT-1000, CO / H2S detection, resolution 0.1ppm), deployed at the bottom of the foundation pit; 1 rain gauge (model: YL-69, accuracy ±0.1mm), installed at the highest point of the construction site;

[0126] Edge computing and transmission systems

[0127] One edge computing gateway (model: Advantech UNO-2484G, industrial-grade ARM architecture, IP65 protection rating) is deployed in the construction site monitoring room and supports wide temperature range of -40℃ to +70℃.

[0128] Communication module: 5G industrial module (Huawei ME909s-821) + industrial Ethernet switch (Huawei S5720) to build a dual-link redundant transmission network;

[0129] Smart monitoring platform

[0130] Server configuration: CPU Intel Xeon Gold 6348, memory 128GB, hard drive 8TB SSD, deployed in the Hefei Rail Transit Group Data Center;

[0131] Mobile App: Developed for Android / iOS, for use by project managers, maintenance teams, and emergency response teams;

[0132] (II) Software Configuration

[0133] Data preprocessing software: integrates 3σ criterion data cleaning, Kalman filtering, and wavelet transform compression algorithms, and supports localized deployment on edge gateways;

[0134] AI diagnostic models: CNN image recognition model (trained based on PyTorch, with a training set containing 50,000 images of personnel violations), LSTM time series prediction model (time series window length 24h), and composite hazard association rule mining model.

[0135] Work order management system: Supports automatic generation, dispatch, tracking, and archiving, and is linked to the emergency response plan database;

[0136] III. Implementation Steps and Operational Data

[0137] (I) Step 1: Construction of the four-dimensional monitoring system and deployment of sensors

[0138] Determination of monitoring indicators and thresholds

[0139] Monitoring Dimensions Core Indicators Level I Warning Threshold Level II Warning Threshold Level III Warning Threshold

[0140] Structural safety foundation pit settlement rate ≥5mm / h ≥2mm / h ≥1mm / h

[0141] Structural safety steel support axial force ≥4000kN ≥3000kN ≥2000kN

[0142] Tower crane overload rate during equipment operation ≥10% ≥5%

[0143] Unauthorized entry into restricted areas.

[0144] Environmental parameters: wind speed ≥15m / s ≥10m / s

[0145] Sensor Deployment and Calibration

[0146] The static level instrument was deployed linearly along the top of the foundation pit retaining piles and calibrated using second-order leveling. The initial settlement value was set to 0. The GNSS displacement monitoring instrument completed the pairing of the base station and the mobile station and calibrated the positioning accuracy to ±5mm.

[0147] The AI ​​camera has completed angle adjustments to ensure coverage of all high-risk work areas; the UWB positioning tag has completed the demarcation of the electronic fence, and the area within 2m of the edge of the foundation pit has been designated as a restricted zone;

[0148] All sensors are synchronized via a PTPv2 clock, with a measured synchronization accuracy of 85ns, which meets the requirement of ≤100ns.

[0149] (II) Step 2: Multi-source data acquisition and edge computing preprocessing

[0150] Data collection

[0151] The structural safety sensor has a sampling frequency of 1Hz, the equipment operation sensor has a sampling frequency of 10Hz, the AI ​​camera has a frame rate of 25fps, and the environmental sensor has a sampling frequency of 0.5Hz.

[0152] At 14:00 on August 15, 2026, the static level instrument at the southeast corner of the foundation pit detected a settlement value of 0.8mm. At 14:30, a sudden rainstorm occurred, and the sampling frequency of the rain gauge was automatically increased to 1Hz.

[0153] Edge computing preprocessing

[0154] Data cleaning: The 3σ criterion was used to remove abnormal axial force values ​​caused by tower crane vibration (peak value 3500kN, exceeding the normal range of 2000-3000kN), and Kalman filtering was used to optimize settlement data and eliminate noise interference.

[0155] Feature extraction: The calculated settlement rate of the foundation pit is 2.2 mm / h (14:00-14:30), the axial force growth rate of the steel support is 50 kN / h, and the wind speed change rate is 3 m / s / h;

[0156] Data compression: Using wavelet transform + Huffman coding, the compression ratio reaches 12:1, compressing 1 hour of monitoring data from 500MB to 42MB, with a 5G link transmission time of only 1.2 seconds;

[0157] Secure transmission

[0158] When heavy rain causes 5G signal fluctuations, it automatically switches to the industrial Ethernet link with a data transmission latency of 450ms and no data loss; during a 10-minute network outage, the edge gateway locally stores 120MB of data, which is automatically retransmitted after the network is restored.

[0159] (III) Step 3: AI-integrated diagnosis and hierarchical early warning

[0160] Multi-source data fusion

[0161] The intelligent monitoring platform integrates settlement data, axial force data, rainfall data, and wind speed data to construct a four-dimensional monitoring database and automatically associates the tag "Southeast corner of the foundation pit - 20260815 - Rainstorm Condition".

[0162] AI-powered intelligent diagnosis

[0163] The CNN model identified two construction workers who were not wearing safety helmets (Level III warning) and one person who entered the restricted area of ​​the foundation pit (Level I warning); the LSTM model, based on 3 hours of settlement data, predicted that the settlement in the next 12 hours would reach 15 mm, far exceeding the Level I threshold.

[0164] The composite hazard identification model determines that "the foundation pit settlement rate of 2.2 mm / h (Level II) + rainstorm (rainfall of 50 mm / h) + wind speed of 12 m / s (Level II)" constitutes a composite risk and is automatically upgraded to a Level I warning.

[0165] Tiered early warning

[0166] At 14:32, a red Level I warning pop-up window appeared on the platform, and the sound and light alarm at the construction site was activated; the mobile APP pushed the warning information to the project manager and emergency rescue team: "Level I warning for composite risk in the southeast corner of the foundation pit, settlement rate 2.2mm / h, heavy rain and strong winds, there is a risk of collapse, it is recommended to evacuate personnel immediately and stop construction."

[0167] A Level III warning message was simultaneously pushed out: "Two people were not wearing safety helmets and need to be rectified on-site."

[0168] (iv) Step 4: Intelligent linkage and closed-loop verification

[0169] Hierarchical linkage response

[0170] Level I Early Warning Response: The platform automatically cuts off the power to the tower crane and construction elevator, and locks the access control at the pit entrance; it automatically generates an emergency work order, including the three-dimensional coordinates of the southeast corner of the pit (X:312567.8, Y:56321.2, Z:-28m), an emergency response plan ("Immediately evacuate personnel → Stop dewatering → Load counterpressure → Reinforce steel supports"), and an emergency resource distribution map (allocate 300t sandbags and 5 steel supports from nearby locations).

[0171] The emergency rescue team received the order at 14:35, arrived at the scene at 14:40, completed the evacuation of personnel at 14:45, and started the load counter-pressure reinforcement at 15:00.

[0172] Level III Warning Response: The safety officer arrived at the scene at 14:38, urged the two personnel to wear safety helmets, and uploaded rectification photos to the platform;

[0173] Process tracking

[0174] Maintenance personnel uploaded reinforcement process data in real time via a mobile app: the sandbag stacking height was 1.5m, and the axial force of the newly added steel support was stable at 2500kN; the platform displayed the settlement rate changes in real time, and the settlement rate dropped to 0.3mm / h at 15:30;

[0175] Closed-loop verification

[0176] At 16:00, the platform instructed the sensors to recollect data: the settlement rate was 0.2 mm / h, the axial force was stable at 2400 kN, and the wind speed dropped to 5 m / s; the AI ​​model verified that the hidden danger had been eliminated, and at 16:05, the status of the hidden danger was updated to "closed loop".

[0177] After the hazard of not wearing a safety helmet is rectified, the safety officer confirms the closed loop, and the platform archives the rectification record;

[0178] (v) Step 5: Model Iteration and Monitoring Optimization

[0179] Incremental training of AI models

[0180] At the end of August 2026, the complex hazard cases of this rainstorm were added to the training set, and the LSTM model was incrementally trained with an iteration cycle of 60 hours. After optimization, the accuracy of the model in predicting settlement under rainstorm conditions improved from 92% to 97%.

[0181] Monitoring parameter optimization

[0182] In response to the characteristics of the rainy season, the warning threshold for Level II settlement of foundation pits has been adjusted from 2 mm / h to 1.5 mm / h to trigger the warning in advance; the sampling frequency of rain gauges has been increased to 1 Hz (automatically triggered under rainstorm conditions).

[0183] Sensor deployment optimization

[0184] Three new static levels were added to the southeast corner of the foundation pit, increasing the deployment density to a spacing of 5 meters; one new anemometer was added to the top of the tower crane to achieve all-round wind speed monitoring.

[0185] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in the present invention, and these should all be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for monitoring safety during engineering construction, characterized in that: Includes the following steps: Step 1: Construction of the four-dimensional monitoring system and deployment of sensors S1.1 Determine monitoring dimensions and indicators: Structural safety dimensions: monitoring foundation pit settlement (accuracy ±0.1mm), slope displacement (accuracy ±0.5mm), tower crane verticality (deviation ≤0.3‰), support axial force (range 0-5000kN), and bridge deflection (accuracy ±0.2mm); Equipment operation dimensions: monitoring tower crane lifting capacity, lifting height, slewing angle, wire rope wear (accuracy ±0.1mm), construction elevator operating speed, and braking performance; Personnel behavior dimension: Monitoring personnel's wearing of safety helmets, fastening of safety belts, entry into restricted areas, and unauthorized use of open flames; Environmental parameters include: wind speed (range 0-60m / s), rainfall (accuracy ±0.1mm), temperature (-40℃~+85℃), humidity (0—100%RH), dust concentration (range 0-10mg / m³), and harmful gas concentration (such as CO, H2S, range 0-100ppm). S1.2 Sensor Selection and Deployment: Structural safety sensors: High-precision hydrostatic level, GNSS displacement monitor, and vibrating wire axial force gauge are used and deployed around the foundation pit, tower crane body, and key parts of the main beam of the bridge, with a protection level ≥ IP67; Equipment operation sensors: Utilizing tension sensors, angle sensors, and laser rangefinders, these sensors are integrated into equipment such as tower cranes and construction elevators, and support integration with equipment control systems. Personnel behavior sensors: AI smart cameras (≥8 million pixels) and UWB positioning tags (positioning accuracy ±10cm) are deployed at construction area entrances and exits and high-risk work areas; Environmental sensors: Ultrasonic anemometers, rain gauges, dust detectors, and gas sensors are used and deployed at high points, pit edges, and densely populated areas of the construction site. S1.3 System Calibration: All sensors are calibrated using standard instruments and time synchronization is achieved using the PTPv2 clock synchronization protocol with a synchronization accuracy of ≤100ns; Step 2: Multi-source data acquisition and edge computing preprocessing S2.1 Data Acquisition: Each sensor acquires data at a preset frequency, with the structural safety sensor sampling frequency ≥1Hz, the equipment operation sensor ≥10Hz, the personnel behavior camera frame rate ≥25fps, and the environmental sensor ≥0.5Hz. S2.2 Edge Computing Preprocessing: Deploy edge computing gateways at the construction site to achieve localized data processing. Data cleaning: Outliers are removed using the 3σ criterion, and vibration interference is eliminated using Kalman filtering; Feature extraction: Settlement rate and displacement acceleration are extracted from structural data; load fluctuation coefficient and wire rope wear rate are extracted from equipment data; feature values ​​of violations are extracted from personnel data. Extract wind speed change rate and cumulative rainfall from environmental data; Data compression: Wavelet transform is used to compress data with a compression ratio of ≥10:1 to ensure transmission efficiency; S2.3 Secure Transmission: It adopts dual-link redundant transmission of "5G + Industrial Ethernet". The data is uploaded to the smart monitoring platform after being encrypted by AES-256. It supports breakpoint resume and local storage of data (capacity ≥1TB) when the network is interrupted. Step 3: AI-integrated diagnosis and hierarchical early warning S3.1 Multi-source data fusion: The intelligent monitoring platform fuses pre-processed structured data with unstructured data (video images) to construct a four-dimensional monitoring database; S3.2 AI Intelligent Diagnosis: Real-time hazard identification: A CNN model is used to identify personnel violations and abnormal equipment conditions; an LSTM model is used to analyze structural deformation trends and equipment failure probabilities, with a diagnostic delay of ≤1 second; Composite Hazard Identification: Construct an association rule mining model to identify composite risks such as "excessive settlement of foundation pit + rainstorm" and "vertical deviation of tower crane + strong wind", with an accuracy rate of ≥99%. S3.3 Graded Early Warning: Divided into three levels according to the severity of the hidden dangers: Level I Warning (Red, Emergency Situation): Such as the settlement rate of the foundation pit ≥5mm / h, the verticality deviation of the tower crane ≥0.5‰, or personnel entering the restricted area and starting a fire, directly endangering safety; Level II Warning (Yellow, Significant Potential Hazard): If the settlement rate is ≥2mm / h, the wire rope wear is ≥3mm, or the wind speed is ≥15m / s, it may develop into a dangerous situation. Level III Warning (Blue, General Hazard): If personnel are not wearing safety helmets or dust concentration exceeds the standard, a reminder to rectify the situation is required. S3.4 Warning methods: Platform pop-up window + audible and visual alarm + mobile APP push + SMS notification. Warning information includes hazard type, location, risk level, and handling suggestions. Step 4: Intelligent Collaborative Response and Closed-Loop Verification S4.1 Hierarchical linkage response: Level I warning: The platform immediately cuts off the power supply to relevant equipment (such as tower cranes and construction elevators), locks the access control of high-risk work areas, automatically generates emergency work orders and dispatches them to project leaders and emergency rescue teams, and simultaneously reports to the housing and construction department; Level II Warning: A rectification work order is generated and dispatched to the responsible team, requiring them to handle the situation within 2 hours; Level III Warning: On-site broadcast reminder, safety officer on-site supervision and rectification; S4.2 Process Tracking: Maintenance personnel upload photos and videos of the process through a mobile app, and the platform tracks the progress in real time. S4.3 Closed-loop verification: After the handling is completed, the platform instructs the sensors to recollect data, and the AI ​​model verifies whether the hidden danger has been eliminated. If the verification is successful, the hidden danger status is updated to "closed-loop". If it fails, the warning level is upgraded. Step 5: Model Iteration and Monitoring Optimization S5.1AI Model Incremental Training: New hidden danger cases and handling data are added to the training set every month to incrementally train the CNN-LSTM model. The iteration cycle is ≤72 hours, and the diagnostic accuracy continues to improve. S5.2 Monitoring Parameter Optimization: Adjust the threshold values ​​of monitoring indicators according to the construction stage, such as lowering the settlement warning threshold during the foundation pit excavation stage; adjust the environmental warning thresholds according to the season, such as lowering the wind speed warning threshold during the typhoon season. S5.3 Sensor Deployment Optimization: Based on fault statistics, adjust the location and number of sensors. For example, if there is abnormal settlement in a certain area, increase the deployment density of hydrostatic levels.

2. The method for monitoring construction safety according to claim 1, characterized in that: The UWB positioning tag mentioned in step 1 supports integration with safety helmets, has a one-click alarm function, and the alarm signal upload delay is ≤1 second.

3. The method for monitoring construction safety according to claim 1, characterized in that: The edge computing gateway in step 2 adopts an industrial-grade ARM architecture, has a protection level of ≥IP65, and supports wide operating temperature range of -40℃ to +70℃.

4. The method for monitoring construction safety according to claim 1, characterized in that: The composite hazard identification model described in step 3 supports custom association rules, and risk association parameters can be configured according to different project types (buildings / rail transit / bridges).

5. The method for monitoring construction safety according to claim 1, characterized in that: The emergency work order in step 4 includes the three-dimensional coordinates of the hazard location, the emergency response plan, and the emergency resource distribution map, which can be viewed offline.

6. The method for monitoring construction safety according to claim 1, characterized in that: The harmful gas sensor described in step 1 supports simultaneous detection of multiple gases, with a response time ≤30 seconds and a resolution ≤0.1ppm.

7. The method for monitoring construction safety according to claim 1, characterized in that: The early warning information mentioned in step 3 supports linkage with the construction site broadcasting system, access control system, and equipment control system to achieve automated emergency response.

8. The method for monitoring construction safety according to claim 1, characterized in that: The incremental training of the model in step 5 adopts the transfer learning method. When there is insufficient new engineering data, it can be quickly adapted based on existing engineering data to reduce the amount of training data required.

9. The method for monitoring construction safety according to claim 1, characterized in that: The data compression in step 2 uses a combination of wavelet transform and Huffman coding, achieving a compression ratio of up to 15:1 without losing key feature data.

Images